Overview

To facilitate local memory allocation on FPGA devices, the Vitis Vision library functions are provided in templates with compile-time parameters. Data is explicitly copied from cv::Mat to xf::cv::Mat and is stored in physically contiguous memory to achieve the best possible performance. After processing, the output in xf::cv::Mat is copied back to cv::Mat to write it into the memory.

xf::cv::Mat Image Container Class

xf::cv::Mat is a template class that serves as a container for storing image data and its attributes.

Note

The xf::cv::Mat image container class is similar to the cv::Mat class of the OpenCV library.

Class Definition

template <int T, int ROWS, int COLS, int NPC, int XFCVDEPTH = 2>
class Mat {
   public:
        unsigned char allocatedFlag; // flag to mark memory allocation in this class
        int rows, cols, size;        // actual image size

        typedef XF_TNAME(T, NPC) DATATYPE;

#if defined(__SYNTHESIS__) && !defined(__SDA_MEM_MAP__)
        hls::stream<DATATYPE, XFCVDEPTH> data;
#else
        DATATYPE* data;
#endif

        Mat(); // default constructor
        Mat(Size _sz);
        Mat(int _rows, int _cols);
        Mat(int _size, int _rows, int _cols);
        Mat(int _rows, int _cols, void* _data);
        Mat(const Mat&); // copy constructor

        ~Mat();

        Mat& operator=(const Mat&); // Assignment operator
        //  XF_TNAME(T, XF_NPPC1) operator() (unsigned int r, unsigned int c);
        //  XF_CTUNAME(T, NPC) operator() (unsigned int r, unsigned int c, unsigned int ch);
        XF_TNAME(T, NPC) read(int index);
        float read_float(int index);
        void write(int index, XF_TNAME(T, NPC) val);
        void write_float(int index, float val);

        void init(int _rows, int _cols, bool allocate = true);
        void copyTo(void* fromData);
        unsigned char* copyFrom();

        const int type() const;
        const int depth() const;
        const int channels() const;

        template <int DST_T>
        void convertTo(Mat<DST_T, ROWS, COLS, NPC, XFCVDEPTH>& dst, int otype, double alpha = 1, double beta = 0);
};

Parameter Descriptions

The following table lists the xf::cv::Mat class parameters and their descriptions:

Table xf::cv::Mat Class Parameter Descriptions

Parameter

Description

rows

The number of rows in the image or height of the image.

cols

The number of columns in the image or width of the image.

size

The number of words stored in the data member. The value is calculated using rows*cols/(number of pixels packed per  word).

allocatedFlag

Flag for memory allocation status

*data

class parameters and the pointer to the words that store the pixels of the image.

The following table lists the member functions and their descriptions:

Table xf::cv::Mat Member Function Descriptions

Member Functions

Description

Mat()

This default constructor initializes the Mat object sizes, using the template parameters ROWS and COLS.

Mat(int _rows, int _cols)

This constructor initializes the Mat object using arguments _rows and _cols.

Mat(const xf::cv::Mat &_src)

This constructor helps clone a Mat object to another. New memory will be allocated for the newly created constructor.

Mat(int _rows, int _cols, void *_data)

This constructor initializes the Mat object using arguments _rows, _cols, and _data. The *data member of the Mat object points to the memory allocated for _data argument, when this constructor is used. No new memory is allocated for the *data member.

convertTo(Mat <DST_T,ROWS, COLS, NPC> &dst, int otype, double alpha=1, double beta=0)

Refer to

xf::cv::convertTo

copyTo(* fromData)

Copies the data from Data pointer into physically contiguous memory allocated inside the constructor.

copyFrom()

Returns the pointer to the first location of the *data member.

read(int index)

Readout a value from a given location and return it as a packed (for multi-pixel/clock) value.

read_float(in t index)

Readout a value from a given location and return it as a float value

write(int index, XF_TNAME(T,NP C) val)

Writes a packed (for multi-pixel/clock) value into the given location.

write_float(i nt index, float val)

Writes a float value into the given location.

type()

Returns the type of the image.

depth()

Returns the depth of the image

channels()

Returns number of channels of the image

~Mat()

This is a default destructor of the Mat object.

Template parameters of the xf::cv::Mat class are used to set the depth of the pixel, number of channels in the image, number of pixels packed per word, maximum number of rows and columns of the image. The following table lists the template parameters and their descriptions:

Table xf::cv::Mat Template Parameter Descriptions

Parameters

Description

TYPE

Type of the pixel data. For example, XF_8UC1 stands for 8-bit unsigned and one channel pixel. More types can be found in include/common/xf_params.h.

HEIGHT

Maximum height of an image.

WIDTH

Maximum width of an image.

NPC

The number of pixels to be packed per word. For instance, XF_NPPC1 for 1 pixel per word; and XF_NPPC8 for 8 pixels per word.

Pixel-Level Parallelism

The amount of parallelism to be implemented in a function from Vitis Vision is kept as a configurable parameter. In most functions, there are two options for processing data.

  • Single-pixel processing

  • Processing eight pixels in parallel

The following table describes the options available for specifying the level of parallelism required in a particular function:

Table Options Available for Specifying the Level of Parallelism

Option

Description

XF_NPPC1

Process 1 pixel per clock cycle

XF_NPPC2

Process 2 pixels per clock cycle

XF_NPPC4

Process 4 pixels per clock cycle

XF_NPPC8

Process 8 pixels per clock cycle

Macros to Work With Parallelism

There are two macros that are defined to work with parallelism.

  • The XF_NPIXPERCYCLE(flags) macro resolves to the number of pixels processed per cycle.

    • XF_NPIXPERCYCLE(XF_NPPC1) resolves to 1

    • XF_NPIXPERCYCLE(XF_NPPC2) resolves to 2

    • XF_NPIXPERCYCLE(XF_NPPC4) resolves to 4

    • XF_NPIXPERCYCLE(XF_NPPC8) resolves to 8

  • The XF_BITSHIFT(flags) macro resolves to the number of times to shift the image size to right to arrive at the final data transfer size for parallel processing.

    • XF_BITSHIFT(XF_NPPC1) resolves to 0

    • XF_BITSHIFT(XF_NPPC2) resolves to 1

    • XF_BITSHIFT(XF_NPPC4) resolves to 2

    • XF_BITSHIFT(XF_NPPC8) resolves to 3

Data Types

Data types will differ, depending on the combination of the depth of pixels and the number of channels in the image. The generic nomenclature of the parameter is listed below.

XF_<Number of bits per pixel><signed (S) or unsigned (U) or float (F)>C<number of channels>

For example, for an 8-bit pixel - unsigned - 1 channel the data type is XF_8UC1.

The following table lists the available data types for the xf::cv::Mat class:

Table xf::cv::Mat Class - Available Data Types

Option

Number of bits per Pixel

Unsigned/ Signed/ Float Type

Number of Channels

XF_2UC1

2

Unsigned

1

XF_8UC1

8

Unsigned

1

XF_8UC2

8

Unsigned

2

XF_8UC3

8

Unsigned

3

XF_8UC4

8

Unsigned

4

XF_10UC1

8

Unsigned

1

XF_10UC2

8

Unsigned

2

XF_10UC3

8

Unsigned

3

XF_10UC4

8

Unsigned

4

XF_12UC1

8

Unsigned

1

XF_12UC2

8

Unsigned

2

XF_12UC3

8

Unsigned

3

XF_12UC4

8

Unsigned

4

XF_16UC1

16

Unsigned

1

XF_16SC1

16

Signed

1

XF_32UC1

32

Unsigned

1

XF_32FC1

32

Float

1

XF_32FC3

32

Float

3

XF_32SC1

32

Signed

1

Manipulating Data Type

Based on the number of pixels to process per clock cycle and the type parameter, there are different possible data types. The Vitis Vision library uses these datatypes for internal processing and inside the xf::cv::Mat class. The following are a few supported types:

  • XF_TNAME(TYPE,NPPC) resolves to the data type of the data member of the xf::cv::Mat object. For instance, XF_TNAME(XF_8UC1,XF_NPPC8) resolves to ap_uint<64>.

  • Word width = pixel depth * number of channels * number of pixels to process per             cycle (NPPC).

  • XF_DTUNAME(TYPE,NPPC) resolves to the data type of the pixel. For instance, XF_DTUNAME(XF_32FC1,XF_NPPC1) resolves to float.

  • XF_PTSNAME(TYPE,NPPC) resolves to the ‘C’ data type of the pixel. For instance, XF_PTSNAME             (XF_16UC1,XF_NPPC2) resolves to unsigned             short.

Note

ap_uint<>, ap_int<>, ap_fixed<>, and ap_ufixed<> types belong to the high-level synthesis (HLS) library. For more information, see the Vivado Design Suite User Guide: High-Level Synthesis (UG902).

xf::cv::imread

The function xf::cv::imread loads an image from the specified file path, copies into xf::cv::Mat and returns it. If the image cannot be read (because of missing file, improper permissions, unsupported or invalid format), the function exits with a non-zero return code and an error statement.

Note

In an HLS standalone mode like Cosim, use cv::imread followed by copyTo function, instead of xf::cv::imread.

API Syntax

template<int PTYPE, int ROWS, int COLS, int NPC>
xf::cv::Mat<PTYPE, ROWS, COLS, NPC> imread (char *filename, int type)

Parameter Descriptions

The table below describes the template and the function parameters.

Table xf::cv::imread Parameter Description

Parameter

Description

PTYPE

Input pixel type. Value should be in accordance with the ‘type’ argument’s value.

ROWS

Maximum height of the image to be read

COLS

Maximum width of the image to be read

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

filename

Name of the file to be loaded

type

Flag that depicts the type of image. The values are:

  • ‘0’ for gray scale

  • ‘1’ for color image

xf::cv::imwrite

The function xf::cv::imwrite saves the image to the specified file from the given xf::cv::Mat. The image format is chosen based on the file name extension. This function internally uses cv::imwrite for the processing. Therefore, all the limitations of cv::imwrite are also applicable to xf::cv::imwrite.

API Syntax

template <int PTYPE, int ROWS, int COLS, int NPC>
void imwrite(const char *img_name, xf::cv::Mat<PTYPE, ROWS, COLS, NPC> &img)

Parameter Descriptions

The table below describes the template and the function parameters.

Table xf::cv::imwrite Parameter Description

Parameter

Description

PTYPE

Input pixel type. Supported types are: XF_8UC1, XF_16UC1, XF_8UC4, and XF_16UC4

ROWS

Maximum height of the image to be read

COLS

Maximum width of the image to be read

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

img_name

Name of the file with the extension

img

xf::cv::Mat array to be saved

xf::cv::absDiff

The function xf::cv::absDiff computes the absolute difference between each individual pixels of an xf::cv::Mat and a cv::Mat, and returns the difference values in a cv::Mat.

API Syntax

template <int PTYPE, int ROWS, int COLS, int NPC>
void absDiff(cv::Mat &cv_img, xf::cv::Mat<PTYPE, ROWS, COLS, NPC>& xf_img, cv::Mat &diff_img )

Parameter Descriptions

The table below describes the template and the function parameters.

Table xf::cv::absDiff Parameter Description

Parameter

Description

PTYPE

Input pixel type

ROWS

Maximum height of the image to be read

COLS

Maximum width of the image to be read

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1, XF_NPPC4, and XF_NPPC8 for 1-pixel, 4-pixel, and 8-pixel parallel operations respectively.

cv_img

cv::Mat array to be compared

xf_img

xf::cv::Mat array to be compared

diff_img

Output difference image(cv::Mat)

xf::cv::convertTo

The xf::cv::convertTo function performs bit depth conversion on each individual pixel of the given input image. This method converts the source pixel values to the target data type with appropriate casting.

dst(x,y)= cast<target-data-type>(α(src(x,y)+β))

Note: The output and input Mat cannot be the same. That is, the converted image cannot be stored in the Mat of the input image.

API Syntax

template<int DST_T> void convertTo(xf::cv::Mat<DST_T,ROWS, COLS, NPC> &dst, int ctype, double alpha=1, double beta=0)

Parameter Descriptions

The table below describes the template and function parameters.

Table xf::cv::convertTo Parameter Description

Parameter

Description

DST_T

Output pixel type. Possible values are XF_8UC1, XF_16UC1, XF_16SC1, and XF_32SC1.

ROWS

Maximum height of image to be read

COLS

Maximum width of image to be read

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1, XF_NPPC4, and XF_NPPC8 for 1-pixel, 4-pixel, and 8-pixel parallel operations respectively. XF_32SC1 and XF_NPPC8 combination is not supported.

dst

Converted xf Mat

ctype

Conversion type : Possible values are listed here.

//Down-convert:

  • XF_CONVERT_16U_TO_8U

  • XF_CONVERT_16S_TO_8U

  • XF_CONVERT_32S_TO_8U

  • XF_CONVERT_32S_TO_16U

  • XF_CONVERT_32S_TO_16S

//Up-convert:

  • XF_CONVERT_8U_TO_16U

  • XF_CONVERT_8U_TO_16S

  • XF_CONVERT_8U_TO_32S

  • XF_CONVERT_16U_TO_32S

  • XF_CONVERT_16S_TO_32S

alpha

Optional scale factor

beta

Optional delta added to the scaled values

Vitis Vision Library Functions

The Vitis Vision library is a set of select OpenCV functions optimized for Zynq-7000, Zynq UltraScale+ MPSoC, Alveo U200 and U50 devices. The maximum resolution supported for all the functions is 4K, except Houghlines and HOG (RB mode).

Note

Resolution Conversion (Resize) in 8 pixel per cycle mode, Dense Pyramidal LK Optical Flow, and Dense Non-Pyramidal LK Optical Flow functions are not supported on the Zynq-7000 SoC ZC702 devices, due to the higher resource utilization.

Note

Number of pixel per clock depends on the maximum bus width a device can support. For example: Zynq-7000 SoC has 64-bit interface and so for a pixel type 16UC1, maximum of four pixel per clock (XF_NPPC4) is possible.

Absolute Difference

API Syntax

The absdiff function finds the pixel wise absolute difference between two input images and returns an output image. The input and the output images must be the XF_8UC1 type.


image0

Where,

  • Iout(x, y) is the intensity of output image at (x,y) position.

  • Iin1(x, y) is the intensity of first input image at (x,y) position.

  • Iin2(x, y) is the intensity of second input image at (x,y) position.

template<int SRC_T, int ROWS, int COLS, int NPC=1>
void absdiff(
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> dst )

Parameter Descriptions

The following table describes the template and the function parameters.

Table absdiff Parameter Description

Parameter

Description

SRC_T

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

src2

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Table absdiff Function Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

62

67

17

8 Pixel

150

0

0

67

234

39

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Table absdiff Function Performance Estimate Summary

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.69

Deviation from OpenCV

There is no deviation from OpenCV, except that the absdiff function supports 8-bit pixels.

Accumulate

The accumulate function adds an image (src1) to the accumulator image (src2), and generates the accumulated result image (dst).

image1

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void accumulate (
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int DST_T, int ROWS, int COLS, int NPC> dst )

Parameter Descriptions

The following table describes the template and the function parameters.

Table accumulate Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

DST_T

Output pixel type. Only 16-bit, unsigned, 1 and 3 channels are supported (XF_16UC1 and XF_16UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Recommend using a multiple of 8, for an 8-pixel operation.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

src2

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

62

55

12

8 Pixel

150

0

0

389

285

61

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process 4K 3 Channel image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

207

72

32

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Deviation from OpenCV

In OpenCV the accumulated image is stored in the second input image. The src2 image acts as both input and output, as shown below:
image2

Whereas, in the Vitis Vision implementation, the accumulated image is stored separately, as shown below:


image3

Accumulate Squared

The accumulateSquare function adds the square of an image (src1) to the accumulator image (src2) and generates the accumulated result (dst).


image4

The accumulated result is a separate argument in the function, instead of having src2 as the accumulated result. In this implementation, having a bi-directional accumulator is not possible as the function makes use of streams.

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void accumulateSquare (
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int DST_T, int ROWS, int COLS, int NPC> dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table accumulateSquare Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

DST_T

Output pixel type. Only 16-bit, unsigned, 1 and 3 channels are supported (XF_16UC1 and XF_16UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

src2

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

71

52

14

8 Pixel

150

0

8

401

247

48

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process 4K 3 Channel image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

3

227

86

37

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.6

Deviation from OpenCV

In OpenCV the accumulated squared image is stored in the second input image. The src2 image acts as input as well as output.


image5
Whereas, in the Vitis Vision implementation, the accumulated squared image is stored separately. image6

Accumulate Weighted

The accumulateWeighted function computes the weighted sum of the input image (src1) and the accumulator image (src2) and generates the result in dst.


image7

The accumulated result is a separate argument in the function, instead of having src2 as the accumulated result. In this implementation, having a bi-directional accumulator is not possible, as the function uses streams.

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void accumulateWeighted (
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int DST_T, int ROWS, int COLS, int NPC> dst,
float alpha )

Parameter Descriptions

The following table describes the template and the function parameters.

Table accumulateWeighted Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

DST_T

Output pixel type. Only 16-bit, unsigned, 1 and 3 channels are supported (XF_16UC1 and XF_16UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Recommend multiples of 8, for an 8-pixel operation.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

src2

Input image

dst

Output image

alpha

Weight applied to input image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Table accumulateWeighted Function Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

5

295

255

52

8 Pixel

150

0

19

556

476

88

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K 3 Channel image.

Table accumulateWeighted Function Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

9

457

387

95

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Table accumulateWeighted Function Performance Estimate Summary

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Deviation from OpenCV

The resultant image in OpenCV is stored in the second input image. The src2 image acts as input as well as output, as shown below:

image8

Whereas, in Vitis Vision implementation, the accumulated weighted image is stored separately.

image9

AddS

The AddS function performs the addition operation between pixels of input image src and given scalar value scl and stores the result in dst.

dst(x,y)= src(x,y) + scl

Where (x,y) is the spatial coordinate of the pixel.

API Syntax

template<int POLICY_TYPE, int SRC_T, int ROWS, int COLS, int NPC =1>
void addS(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, unsigned char _scl[XF_CHANNELS(SRC_T,NPC)],xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table 26. AddS Parameter Description

Parameter

Description

SRC_T

Input pixel type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

First input image

_scl

Input scalar value, the size should be number of channels.

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the AddS function in both the resource optimized (8 pixel) mode and normal mode, as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

100

101

LUT

52

185

CLB

20

45

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Addweighted

The addweighted function calculates a weighted sum of two input images src1, src2 and generates the result in dst.

dst(x,y)= src1(x,y)*alpha+src2(x,y)*beta+ gamma

API Syntax

template< int SRC_T , int DST_T,   int ROWS, int COLS, int NPC=1>
void addWeighted(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, float alpha, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src2, float beta, float gamma, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table 29. Addweighted Parameter Description

Parameter

Description

SRC_T

Input Pixel Type. 8-bit, unsigned,1 channel is supported (XF_8UC1)

DST_T

Output Pixel Type. 8-bit, unsigned,1 channel is supported (XF_8UC1)

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

First Input image

Alpha

Weight applied on first image

_src2

Second Input image

Beta

Weight applied on second image

gamma

Scalar added to each sum

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the Addweighted function in Resource optimized (8 pixel) mode and normal mode, as generated in Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

11

25

FF

903

680

LUT

851

1077

CLB

187

229

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Autoexposurecorrection

Auto exposure correction improves contrast and brightness of the image and also corrects the exposure of the input frame. The algorithm uses luminence histogram equalization to improve overall exposure and contrast of the image. Luminence (V) is extracted after converting input image to HSV color space. Once the algorthm is applied the image is converted back to RGB color space.

API Syntax

template <int SRC_T, int DST_T, int SIN_CHANNEL_TYPE, int ROWS, int COLS, int NPC = 1>
void autoexposurecorrection(xf::cv::Mat<SRC_T, ROWS, COLS, NPC>& src,
                            xf::cv::Mat<DST_T, ROWS, COLS, NPC>& dst,
                            unsigned int hist_array1[1][256],
                            unsigned int hist_array2[1][256])

Parameter Descriptions

The following table describes template parameters and arguments of the function.

Table AEC Parameter Description

Parameter

Description

SRC_T

Input pixel type. 8-bit unsigned 3 channel is supported (XF_8UC3).

DST_T

Output pixel type. 8-bit unsigned 3 channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

SIN_CHANNEL_TYPE

Single channel type. should be XF_8UC1

NPC

Number of pixels to be processed per cycle; possible options is XF_NPPC1, XF_NPPC2 AND so on

src

Input image

dst

Output image

hist_array1

Histogram array

hist_array2

Histogram array

Resource Utilization

The following table summarizes the resource utilization of kernel in different configurations, generated using Vitis HLS 2020.2 tool, to process a FULL HD image.

Table AEC Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

4

18

6713

2996

1103

2 pixel

300

4

27

7618

3705

1257

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vitis HLS 2020.2 tool, to process a FULL HD image.

Operating Mode

Operating Frequency (MHz)

Latency Estimate Max (ms)

1 pixel

300

7

2 pixel

300

3.7

Autowhitebalance

Grayworld whitebalancing algorithm:

This algorithm scales the values of pixels based on a gray-world assumption which states that the average of all channels should result in a gray image. It adds a modification which thresholds pixels based on their saturation value and only uses pixels below the provided threshold in finding average pixel values. Saturation is calculated using the following for a 3-channel RGB image per pixel I and is in the range [0, 1]:

image161

A threshold of 1 means that all pixels are used to white-balance, while a threshold of 0 means no pixels are used. Lower thresholds are useful in white-balancing saturated images.

Simple whitebalancing algorithm:

A simple white balance algorithm that works by independently stretching each of the input image channels to the specified range(maximum and minimum). Computes channel wise intensity histogram and ignores p% maximum and minimum values and finally normalize each channel with min and max. For increased robustness it ignores the top and bottom \(p\%\ \ (4\%\ is\ fixed)\) of pixel values.

API Syntax for Simple white balance

API Syntax for Grayworld white balance

template <int SRC_T, int DST_T, int ROWS, int COLS, int NPC = 1, int WB_TYPE>
void AWBChannelGain(xf::cv::Mat<SRC_T, ROWS, COLS, NPC>& src,
            xf::cv::Mat<DST_T, ROWS, COLS, NPC>& dst,
            float thresh,
            int i_gain[3])

template <int SRC_T, int DST_T, int ROWS, int COLS, int NPC = 1, int WB_TYPE, int S_DEPTH = 2>
void AWBGainUpdate(xf::cv::Mat<SRC_T, ROWS, COLS, NPC>& src1,
           xf::cv::Mat<DST_T, ROWS, COLS, NPC, S_DEPTH>& src2,
           float thresh,
           int i_gain[3])

Parameter Descriptions

The following table describes the template and the function parameters.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vitis HLS 2020.1 tool for the Xilinx xc7vx485t-ffg1157-1 FPGA, to process a 4K image.

Table Autowhitebalance Resource Utilization Summary

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

14

10

4798

4953

1757

2 pixel

300

14

10

8335

8535

2901

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vitis HLS 2020.1 tool for the Xilinx xc7vx485t-ffg1157-1 FPGA, to process a 4K image.

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

Max (ms)

1 pixel

300

55.2 for still image(27.9 for video stream)

2 pixel

300

28 for still image(14.2 for video stream)

Badpixelcorrection

An image sensor may have a certain number of defective/bad pixels that may be the result of manufacturing faults or variations in pixel voltage levels based on temperature or exposure. The Badpixelcorrection module removes the defective pixels in the image using below operation.

If the middle pixel value is lesser than minimum neighborhood value, will consider minimum neighborhood value as mid pixel, otherwise mid pixel value is greater than maximum neighborhood value, will consider maximum neighborhood as mid pixel.

API Syntax

template<int TYPE, int ROWS, int COLS, int NPPC=1, int BORDER_T=XF_BORDER_CONSTANT, int USE_URAM=0>void badpixelcorrection(xf::cv::Mat<TYPE, ROWS, COLS, NPPC> &_src,xf::cv::Mat<TYPE, ROWS, COLS, NPPC> &_dst)

The following table describes the template and the function parameters.

Table badpixelcorrection Parameter Description

Parameter

Description

TYPE

Input and Output Pixel Type.

ROWS

Maximum height of input and output image (Must be multiple of NPPC)

COLS

Maximum width of input and output image (Must be multiple of NPPC)

NPPC

Number of Pixels to be processed per cycle.

BORDER_T

Border Type supported is XF_BORDER_CONSTANT

USE_URAM

Enable to map storage structures to UltraRAM.

_src

Input Bayer image

_dst

Output Bayer image

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.2 tool for the Xilinx xc7vx485t-ffg1157-1 FPGA, to process a 4K image.

Table badpixelcorrection Resource Utilization Summary

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

SLICE

1 pixel

300

10

0

979

744

355

2 pixel

300

10

0

1148

1177

458

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vivado HLS 2019.2 tool for the Xilinx xc7vx485t-ffg1157-1, to process 4K image.

Table badpixelcorrection Resource Utilization Summary

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

Max (ms)

1 pixel

300

27.8

2 pixel

300

14.2

Brute-force (Bf) Feature Matcher

Bf matcher takes the descriptor of one feature in first set and is matched with all other features in second set and the closest one is returned.

API Syntax

template <int PU = 1, int MAX_KEYPOINTS = 10000>
void bfMatcher(ap_uint<256> desc_list_q[MAX_KEYPOINTS],
               ap_uint<256> desc_list_t[MAX_KEYPOINTS],
               ap_int<16> match_list[MAX_KEYPOINTS],
               ap_uint<32> num_keypoints_q,
               ap_uint<32> num_keypoints_t,
               float ratio_thresh)

Parameter Descriptions

The following table describes template paramters and arguments of the function.

Table brute-force matcher Parameter Description

Parameter

Description

PU

Parallel units / compute units. Number of parallel matches computed. Default is ‘1’. Increasing this parameter results in lesser compute time, but also consumes more hardware resources.

MAX_KEYPOINTS

Maximum keypoints in the query and training feature sets.

desc_list_q

Feature descriptor query list of 256-bit type.

desc_list_t

Feature descriptor training list of 256-bit type.

match_list

Index of corresponding matches for query list in training set.

num_keypoints_q

Total number keypoints in the query set. This must not exceed MAX_KEYPOINTS.

num_keypoints_t

Total number keypoints in the training set. This must not exceed MAX_KEYPOINTS.

ratio_thresh

Ratio threshold for lowe’s test, for strong matches.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vitis 2020.2 tool, for MAX_KEYPOINTS of 10000.

Table brute-force matcher Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

PU = 1

300

162

0

5152

8453

PU = 2

300

176

0

9471

16320

PU = 10

300

176

0

17708

48839

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vitis 2020.2 tool, for MAX_KEYPOINTS of 10000.

Table brute-force matcher Function Performance Estimate Summary

Operating Mode

Operating Frequency (MHz)

Latency Estimate Max (ms)

PU = 1

300

333.4

PU = 2

300

168.6

PU = 10

300

34.285

Bilateral Filter

In general, any smoothing filter smoothens the image which will affect the edges of the image. To preserve the edges while smoothing, a bilateral filter can be used. In an analogous way as the Gaussian filter, the bilateral filter also considers the neighboring pixels with weights assigned to each of them. These weights have two components, the first of which is the same weighing used by the Gaussian filter. The second component takes into account the difference in the intensity between the neighboring pixels and the evaluated one.

The bilateral filter applied on an image is:
image10
Where
image11
and image12 is a gaussian filter with variance image13.

The gaussian filter is given by: image14

API Syntax

template<int FILTER_SIZE, int BORDER_TYPE, int TYPE, int ROWS, int COLS, int NPC=1>
void bilateralFilter (
xf::cv::Mat<int TYPE, int ROWS, int COLS, int NPC> src,
xf::cv::Mat<int TYPE, int ROWS, int COLS, int NPC> dst,
float sigma_space, float sigma_color )

Parameter Descriptions

The following table describes the template and the function parameters.

Table bilateralFilter Parameter Description

Parameter

Description

FILTER_SIZE

Filter size. Filter size of 3 (XF_FILTER_3X3), 5 (XF_FILTER_5X5) and 7 (XF_FILTER_7X7) are supported

BORDER_TYPE

Border type supported is XF_BORDER_CONSTANT

TYPE

Input and output pixel type. Only 8-bit, unsigned, 1 channel, and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; this function supports XF_NPPC1 and XF_NPPC8.

src

Input image

dst

Output image

sigma_space

Standard deviation of filter in spatial domain

sigma_color

Standard deviation of filter used in color space

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to progress a grayscale HD (1080x1920) image.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

3x3

300

6

22

4934

4293

5x5

300

12

30

5481

4943

7x7

300

37

48

7084

6195

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to progress a 4K 3 channel image.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

3x3

300

12

32

8342

7442

5x5

300

27

57

10663

8857

7x7

300

49

107

12870

12181

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Filter Size

Latency Estimate

300 MHz

Max Latency (ms)

1 pixel

3x3

7.18

5x5

7.20

7x7

7.22

Deviation from OpenCV

Unlike OpenCV, Vitis Vision only supports filter sizes of 3, 5 and 7.

Bit Depth Conversion

The convertTo function converts the input image bit depth to the required bit depth in the output image.

API Syntax

template <int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void convertTo(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src_mat, xf::cv::Mat<DST_T, ROWS, COLS, NPC> &_dst_mat, ap_uint<4> _convert_type, int _shift)

Parameter Descriptions

The following table describes the template and the function parameters.

Table 36. convertTo Parameter Description

Parameter

Description

SRC_T

Input pixel type. 8-bit, unsigned, 1 channel (XF_8UC1),

16-bit, unsigned, 1 channel (XF_16UC1),

16-bit, signed, 1 channel (XF_16SC1),

32-bit, signed, 1 channel (XF_32SC1) are supported.

DST_T

Output pixel type. 8-bit, unsigned, 1 channel (XF_8UC1),

16-bit, unsigned, 1 channel (XF_16UC1),

16-bit, signed, 1 channel (XF_16SC1),

32-bit, signed, 1 channel (XF_32SC1) are supported.

ROWS

Height of input and output images

COLS

Width of input and output images

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively. XF_NPPC8 is not supported with the 32-bit input and output pixel type.

_src_mat

Input image

_dst_mat

Output image

_convert_ty pe

This parameter specifies the type of conversion required. (See XF_convert_bit_depth_e enumerated type in file xf_params.h for possible values.)

_shift

Optional scale factor

Possible Conversions

The following table summarizes supported conversions. The rows are possible input image bit depths and the columns are corresponding possible output image bit depths (U=unsigned, S=signed).

Table 37. convertTo Function Supported Conversions

INPUT/OUTPUT

U8

U16

S16

U32

S32

U8

NA

yes

yes

NA

yes

U16

yes

NA

NA

NA

yes

S16

yes

NA

NA

NA

yes

U32

NA

NA

NA

NA

NA

S32

yes

yes

yes

NA

NA

Resource Utilization

The following table summarizes the resource utilization of the convertTo function, generated using Vivado HLS 2019.1 tool for the Xilinx® Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

8

581

523

119

8 Pixel

150

0

8

963

1446

290

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

8

591

541

124

8 Pixel

150

0

8

915

1500

308

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.69

Bitwise AND

The bitwise_and function performs the bitwise AND operation for each pixel between two input images, and returns an output image.

image15

Where,

  • image16 is the intensity of output image at (x, y) position

  • image17 is the intensity of first input image at (x, y) position

  • image18 is the intensity of second input image at (x, y) position

API Syntax

template<int SRC_T, int ROWS, int COLS, int NPC=1>
void bitwise_and (
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> dst )

Parameter Descriptions

The following table describes the template and the function parameters.

Table 41. bitwise_and Parameter Description

Parameter

Description

SRC_T

Input and output pixel type. Supports 1 channel and 3 channels (XF_8UC1 and XF_8UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8 pixel mode)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations, respectively.

src1

Input image

src2

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

62

44

10

8 Pixel

150

0

0

59

72

13

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K 3Channel image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

155

61

22

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Bitwise NOT

The bitwise_not function performs the pixel wise bitwise NOT operation for the pixels in the input image, and returns an output image. image19

Where,

  • image20 is the intensity of output image at (x, y) position

  • image21 is the intensity of input image at (x, y) position

API Syntax

template<int SRC_T, int ROWS, int COLS, int NPC=1>
void bitwise_not (
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> dst )

Parameter Descriptions

The following table describes the template and the function parameters.

Table 45. bitwise_not Parameter Description

Parameter

Description

SRC_T

Input and output pixel type. Supports 1 channel and 3 channels (XF_8UC1 and XF_8UC3).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations, respectively.

src

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Table 46. bitwise_not Function Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

97

78

20

8 Pixel

150

0

0

88

97

21

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K 3Channel image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

155

61

22

… rubric:: Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Bitwise OR

The bitwise_or function performs the pixel wise bitwise OR

operation between two input images, and returns an output image. image22

Where,

  • image23 is the intensity of output image at (x, y) position

  • image24 is the intensity of first input image at (x, y) position

  • image25 is the intensity of second input image at (x, y) position

API Syntax

template<int SRC_T, int ROWS, int COLS, int NPC=1>
void bitwise_or (
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> dst )

Parameter Descriptions

The following table describes the template and the function parameters.

Table 49. bitwise_or Parameter Description

Parameter

Description

SRC_T

Input and output pixel type. Supports 1 channel and 3 channels (XF_8UC1 and XF_8UC3).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of 8, for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

src2

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

62

44

10

8 Pixel

150

0

0

59

72

13

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K 3Channel image

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

155

61

22

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Bitwise XOR

The bitwise_xor function performs the pixel wise bitwise XOR operation between two input images, and returns an output image, as shown below:

image26

Where,

  • image27 is the intensity of output image at (x, y) position

  • image28 is the intensity of first input image at (x, y) position

  • image29 is the intensity of second input image at (x, y) position

API Syntax

template<int SRC_T, int ROWS, int COLS, int NPC=1>
void bitwise_xor(
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> dst )

Parameter Descriptions

The following table describes the template and the function parameters.

Table 53. bitwise_xor Parameter Description

Parameter

Description

SRC_T

Input and output pixel type. Supports 1 channel and 3 channels (XF_8UC1 and XF_8UC3).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of 8, for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

src2

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image:

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

62

44

10

8 Pixel

150

0

0

59

72

13

Performance Estimate

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4k Channel image

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

155

61

22

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image:

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Blacklevelcorrection

Black level leads to the whitening of image in dark region and perceived loss of overall contrast. The Blacklevelcorrection algorithm corrects the black and white levels of the overall image.

API Syntax

template <int SRC_T,int MAX_ROWS,int MAX_COLS,int NPPC = XF_NPPC1,int MUL_VALUE_WIDTH = 16,int FL_POS = 15,int USE_DSP = 1>
void blackLevelCorrection(xf::cv::Mat<SRC_T, MAX_ROWS, MAX_COLS, NPPC>& _Src,
                          xf::cv::Mat<SRC_T, MAX_ROWS, MAX_COLS, NPPC>& _Dst,
                          XF_CTUNAME(SRC_T, NPPC) black_level,
                          float mul_value)

Parameter Descriptions

The following table describes the template and the function parameters.

Table blacklevelcorrection correction Parameter Description

Parameter

Description

MUL_VALUE_WIDTH

Width of multiplication factor.

FL_POS

Number of fractional bits in multiplication factor.

USE_DSP

Enables usage of DSP for multiplication.

SRC_T

Input pixel type. 8/10/12/16-bit unsigned 1 channel are supported (XF_8UC1, XF_10UC1, XF_12UC1, XF_16UC1).

DST_T

Output pixel type. 8/10/12/16-bit unsigned 1 channel are supported (XF_8UC1, XF_10UC1, XF_12UC1, XF_16UC1).

MAX_ROWS

Maximum height of input and output image.

MAX_COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1, XF_NPPC2 AND so on

_Src

Input image

_Dst

Output image

black_level

Black level value

mul_value

Multiplication factor for blacklevel correction; which is computed as maxlevel/(maxlevel-blacklevel)

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vitis HLS 2020.2 tool, to process a FULL HD image.

Table blacklevelcorrection Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel-8U

300

0

0

279

271

70

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vitis HLS 2020.2 tool, to process a FULL HD image.

Table blacklevelcorrection Function Performance Estimate Summary

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

Max (ms)

1 pixel

300

7

2 pixel

300

3.6

Box Filter

The boxFilter function performs box filtering on the input image. Box filter acts as a low-pass filter and performs blurring over the image. The boxFilter function or the box blur is a spatial domain linear filter in which each pixel in the resulting image has a value equal to the average value of the neighboring pixels in the image. image30

API Syntax

template<int BORDER_TYPE,int FILTER_TYPE, int SRC_T, int ROWS, int COLS,int NPC=1,bool USE_URAM=false>
void boxFilter(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst_mat)

Parameter Descriptions

The following table describes the template and the function parameters.

Table 57. boxFilter Parameter Description

Parameter

Description

FILTER_SIZE

Filter size. Filter size of 3(XF_FILTER_3X3), 5(XF_FILTER_5X5) and 7(XF_FILTER_7X7) are supported

BORDER_TYPE

Border Type supported is XF_BORDER_CONSTANT

SRC_T

Input and output pixel type. 8-bit, unsigned, 16-bit unsigned and 16-bit signed, 1 channel is supported (XF_8UC1)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

USE_URAM

Enable to map storage structures to UltraRAM

_src_mat

Input image

_dst_mat

Output image

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

3x3

300

3

1

545

519

104

5x5

300

5

1

876

870

189

7x7

300

7

1

1539

1506

300

8 Pixel

3x3

150

6

8

1002

1368

264

5x5

150

10

8

1576

3183

611

7x7

150

14

8

2414

5018

942

The following table summarizes the resource utilization of the kernel in different configurations, generated using the Vivado HLS™ 2019.1 tool for the xczu7ev-ffvc1156-2-e FPGA, to process a grayscale 4K (3840x2160) image with UltraRAM enable.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

URAM

DSP_48Es

FF

LUT

1 Pixel

3x3

300

0

1

1

821

521

5x5

300

0

1

1

1204

855

7x7

300

0

1

1

2083

1431

8 Pixel

3x3

150

0

3

8

1263

1480

5x5

150

0

5

8

1771

3154

7x7

150

0

7

8

2700

5411

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode
Operating Frequency

(MHz)

Filter Size

Latency Estimate

Max Latency (ms)

1 pixel

300

3x3

7.2

300

5x5

7.21

300

7x7

7.22

8 pixel

150

3x3

1.7

150

5x5

1.7

150

7x7

1.7

BoundingBox

The boundingbox function highlights the region of interest (ROI) from the input image using below equations.

P(X,Y) ≤ P(xi, yi) ≤ P(X,Y’)

P(X’,Y) ≤ P(xi, yi) ≤ P(X’,Y’)

Where,

  • P(xi, yi) - Current pixel location

  • P(X,Y) - Top left corner of ROI

  • P(X,Y’) - Top right corner of ROI

  • P(X’,Y) - Bottom left corner of ROI

  • P(X’,Y’) - Bottom Right of ROI

API Syntax

template<int SRC_T, int ROWS, int COLS, int MAX_BOXES=1, int NPC=1>
void boundingbox(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat, xf::cv::Rect_<int> *roi , xf::cv::Scalar<4,unsigned char > *color, int num_box)

Parameter Descriptions

The following table describes the template and the function parameters.

Table 61. boundingbox Parameter Description

Parameter

Description

SRC_T

Input pixel Type. Only 8-bit, unsigned, 1 channel and 3 channel is supported (XF_8UC1,XF_8UC3).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of NPC.

MAX_BOXES

Maximum number of boxes, fixed to 5.

NPC

Number of pixels to be processed per cycle, possible options are XF_NPPC1 only.

_src_mat

Input image

roi

ROI is a xf::cv::Rect object that consists of the top left corner of the rectangle along with the height and width of the rectangle.

color

The xf::cv::Scalar object consists of color information for each box (ROI).

num_box

Number of boxes to be detected. It should be equal or less than MAX_BOXES.

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

5

4

2521

1649

409

Performance Estimate

The following table summarizes the performance of the kernel in 1-pixel mode as generated using Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a grayscale 4K (2160x3840) image for highlighting 3 different boundaries (480x640, 100x200, 300x300).

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

0.15

Vitis Vision Reference

The xf::cv::boundingbox is complaint with below Vitis Vision function:

void rectangle(Mat& img, Rect rec, const Scalar& color, int thickness=1, int lineType=8, int shift=0 )

Canny Edge Detection

The Canny edge detector finds the edges in an image or video frame. It is one of the most popular algorithms for edge detection. Canny algorithm aims to satisfy three main criteria:

  1. Low error rate: A good detection of only existent edges.

  2. Good localization: The distance between edge pixels detected and real edge pixels have to be minimized.

  3. Minimal response: Only one detector response per edge.

In this algorithm, the noise in the image is reduced first by applying a Gaussian mask. The Gaussian mask used here is the average mask of size 3x3. Thereafter, gradients along x and y directions are computed using the Sobel gradient function. The gradients are used to compute the magnitude and phase of the pixels. The phase is quantized and the pixels are binned accordingly. Non-maximal suppression is applied on the pixels to remove the weaker edges.

Edge tracing is applied on the remaining pixels to draw the edges on the image. In this algorithm, the canny up to non-maximal suppression is in one kernel and the edge linking module is in another kernel. After non-maxima suppression, the output is represented as 2-bit per pixel, Where:

  • 00 - represents the background

  • 01 - represents the weaker edge

  • 11 - represents the strong edge

The output is packed as 8-bit (four 2-bit pixels) in 1 pixel per cycle operation and packed as 16-bit (eight 2-bit pixels) in 8 pixel per cycle operation. For the edge linking module, the input is 64-bit, such 32 pixels of 2-bit are packed into a 64-bit. The edge tracing is applied on the pixels and returns the edges in the image.

API Syntax

The .. rubric:: API Syntax for Canny is:

template<int FILTER_TYPE,int NORM_TYPE,int SRC_T,int DST_T, int ROWS, int COLS,int NPC,int NPC1,bool USE_URAM=false>
void Canny(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<DST_T, ROWS, COLS, NPC1> & _dst_mat,unsigned char _lowthreshold,unsigned char _highthreshold)

The .. rubric:: API Syntax for EdgeTracing is:

template<int SRC_T, int DST_T, int ROWS, int COLS,int NPC_SRC,int NPC_DST,bool USE_URAM=false>
voidEdgeTracing(xf::cv::Mat<SRC_T, ROWS, COLS, NPC_SRC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC_DST> & _dst)

Parameter Descriptions

The following table describes the xf::cv::Canny template and function parameters:

Table 64. xf::cv::Canny Parameter Description

Parameter

Description

FILTER_TYPE

The filter window dimensions. The options are 3 and 5.

NORM_TYPE

The type of norm used. The options for norm type are L1NORM and L2NORM.

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1)

DST_T

Output pixel type. Only XF_2UC1 is supported. The output in case of NPC=XF_NPPC1 is 8-bit, where four 2-bit pixel values are packed into 8-bit. The output in case of NPC=XF_NPPC8 is 16-bit, where 8 2-bit pixel values are packed into 16-bit.

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image (must be a multiple of 8, in case of 8 pixel mode)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively. In XF_NPPC, the output image pixels are packed and precision is XF_NPPC4. In XF_NPPC8, output pixels precision is XF_NPPC8.

USE_URAM

Enable to map some storage structures to URAM

_src_mat

Input image

_dst_mat

Output image

_lowthreshold

The lower value of threshold for binary thresholding.

_highthreshold

The higher value of threshold for binary thresholding.

The following table describes the EdgeTracing template and function parameters:

Table 65. EdgeTracing Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only XF_2UC1 is supported. The input in case of NPC=XF_NPPC1 is 8-bit, where four 2-bit pixel values are packed into 8-bit. The input in case of of NPC=XF_NPPC8 is 16-bit, where 8 2-bit pixel values are packed into 16-bit.

DST_T

Output pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1)

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image (must be a multiple of 32)

NPC_SRC

Number of pixels to be processed per cycle. Fixed to XF_NPPC32.

NPC_DST

Number of pixels to be written to destination. Fixed to XF_NPPC8.

USE_URAM

Enable to map storage structures to URAM.

_src

Input image

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of xf::cv::Canny and EdgeTracing in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image for Filter size is 3.

Name

Resource Utilization

1 pixel

1 pixel

8 pixel

8 pixel

Edge Linking

Edge Linking

L1NORM,FS:3

L2NORM,FS:3

L1NORM,FS:3

L2NORM,FS:3

300 MHz

300 MHz

150 MHz

150 MHz

300 MHz

150 MHz

BRAM_18K

22

18

36

32

84

84

DSP48E

2

4

16

32

3

3

FF

3027

3507

4899

6208

17600

14356

LUT

2626

3170

6518

9560

15764

14274

CLB

606

708

1264

1871

2955

3241

The following table summarizes the resource utilization of xf::cv::Canny and EdgeTracing in different configurations, generated using Vivado HLS 2019.1 tool for the xczu7ev-ffvc1156-2-e FPGA, to process a grayscale 4K image for Filter size is 3.

Name

Resource Utilization

1 pixel

1 pixel

8 pixel

8 pixel

Edge Linking

Edge Linking

L1NORM,FS:3

L2NORM,FS:3

L1NORM,FS:3

L2NORM,FS:3

300 MHz

300 MHz

150 MHz

150 MHz

300 MHz

150 MHz

BRAM_18K

10

8

3

3

4

4

URAM

1

1

15

13

8

8

DSP48E

2

4

16

32

8

8

FF

3184

3749

5006

7174

5581

7054

LUT

2511

2950

6695

9906

4092

6380

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image for L1NORM, filter size is 3 and including the edge linking module.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

10.8

8 pixel operation (150 MHz)

8.5

Deviation from OpenCV

In OpenCV Canny function, the Gaussian blur is not applied as a pre-processing step.

Channel Combine

The merge function, merges single channel images into a multi-channel image. The number of channels to be merged should be two, three or four.

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void merge(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src1, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src2, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src3, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src4, xf::cv::Mat<DST_T, ROWS, COLS, NPC> &_dst)

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void merge(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src1, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src2, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src3, xf::cv::Mat<DST_T, ROWS, COLS, NPC> &_dst)

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void merge(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src1, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src2, xf::cv::Mat<DST_T, ROWS, COLS, NPC> &_dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table 69. merge Parameter Description

Paramete r

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1, channel is supported (XF_8UC1)

DST_T

Output pixel type. 8-bit, unsigned,2,3 and 4 channels are supported (XF_8UC2, XF_8UC3 and XF_8UC4)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 for 1 pixel operation.

_src1

Input single-channel image

_src2

Input single-channel image

_src3

Input single-channel image (only for 3 and 4 input config)

_src4

Input single-channel image (only for 4 input config)

_dst

Output multi-channel image

Resource Utilization

The following table summarizes the resource utilization of the merge function, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process 4 single-channel HD (1080x1920) images.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

8

494

386

85

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process 4 single channel HD (1080x1920) images.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.92

Channel Extract

The extractChannel function splits a multi-channel array (32-bit pixel-interleaved data) into several single-channel arrays and returns a single channel. The channel to be extracted is specified by using the channel argument.

The value of the channel argument is specified by macros defined in the xf_channel_extract_e enumerated data type. The following table summarizes the possible values for the xf_channel_extract_e enumerated data type:

Table 72. xf_channel_extract_e Enumerated Data Type Values

Channel

Enumerated Type

Unknown

XF_EXTRACT_CH_0

Unknown

XF_EXTRACT_CH_1

Unknown

XF_EXTRACT_CH_2

Unknown

XF_EXTRACT_CH_3

RED

XF_EXTRACT_CH_R

GREEN

XF_EXTRACT_CH_G

BLUE

XF_EXTRACT_CH_B

ALPHA

XF_EXTRACT_CH_A

LUMA

XF_EXTRACT_CH_Y

Cb/U

XF_EXTRACT_CH_U

Cr/V/Value

XF_EXTRACT_CH_V

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void extractChannel(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_mat, uint16_t _channel)

Parameter Descriptions

The following table describes the template and the function parameters.

Table 73. extractChannel Parameter Description

Paramete r

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 4channel is supported (XF_8UC4)

DST_T

Output pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1)

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be multiple of 8 for 8 pixel mode

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 for 1 pixel operation.

_src_ma t

Input multi-channel image

_dst_ma t

Output single channel image

_channe l

Channel to be extracted (See xf_channel_extract_e enumerated type in file xf_params.h for possible values.)

Resource Utilization

The following table summarizes the resource utilization of the extractChannel function, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4 channel HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

8

508

354

96

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a 4 channel HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.92

Color Conversion

The color conversion functions convert one image format to another image format, for the combinations listed in the following table. The rows represent the input formats and the columns represent the output formats. Supported conversions are discussed in the following sections.

I/O Formats

RGBA

NV12

NV21

IYUV

UYVY

YUYV

YUV4

RGB

BGR

RGBA

N/A

For details, see the RGBA to NV12

For details, see the RGBA to NV21

For details, see the RGBA/RGB to IYUV

For details, see the RGBA/RGB to YUV4

NV12

For details, see the NV12 to RGBA

N/A

For details, see the NV12 to NV21/NV21 to NV12

For details, see the NV12 to IYUV

For details, see the NV12/NV21 to UYVY/YUYV

For details, see the NV12/NV21 to UYVY/YUYV

For details, see the NV12 to YUV4

For details, see the NV12/NV21 to RGB/ BGR

For details, see the NV12/NV21 to RGB/ BGR

NV21

For details, see the NV21 to RGBA

For details, see the NV12 to NV21/NV21 to NV12

N/A

For details, see the NV21 to IYUV

For details, see the NV12/NV21 to UYVY/YUYV

For details, see the NV12/NV21 to UYVY/YUYV

For details, see the NV21 to YUV4

For details, see the NV12/NV21 to RGB/ BGR

For details, see the NV12/NV21 to RGB/ BGR

IYUV

For details, see the IYUV to RGBA/RGB

For details, see the IYUV to NV12

N/A

For details, see the IYUV to YUV4

For details, see the IYUV to RGBA/RGB

UYVY

For details, see the UYVY to RGBA

For details, see the UYVY to NV12

For details, see the UYVY to IYUV

N/A

YUYV

For details, see the YUYV to RGBA

For details, see the YUYV to NV12

For details, see the YUYV to IYUV

N/A

YUV4

N/A

RGB

For details see the RGB/ BGR to NV12/NV21

For details see theRGB/ BGR to NV12/NV21

For details see the RGBA/RGB to IYUV

For details see the RGB/BGR to UYVY/YUYV

For details see the RGB/BGR to UYVY/YUYV

For details see the RGBA/RGB to YUV4

For details see the BGR to RGB / RGB to BGR

BGR

For details see the RGB/ BGR to NV12/NV21

For details see theRGB/ BGR to NV12/NV21

For details see the RGB/BGR to UYVY/YUYV

For details see the RGB/BGR to UYVY/YUYV

For details see the BGR to RGB / RGB to BGR

Other conversions

Few other conversions are also added. BGR/RGB<->HSV,BGR/RGB<->HLS,BGR/RGB<->YCrCb,BGR/RGB<->XYZ and RGB<->BGR conversions are added.

RGB to YUV Conversion Matrix

Following is the formula to convert RGB data to YUV data:
image31

YUV to RGB Conversion Matrix

Following is the formula to convert YUV data to RGB data:
image32

Source: http://www.fourcc.org/fccyvrgb.php

RGBA/RGB to YUV4

The rgba2yuv4 function converts a 4-channel RGBA image to YUV444 format and the rgb2yuv4 function converts a 3-channel RGB image to YUV444 format. The function outputs Y, U, and V streams separately.

API Syntax

template <int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void rgba2yuv4(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _u_image, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _v_image)

template <int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void rgb2yuv4(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _u_image, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _v_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table (rgba/rgb)2yuv4 Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 4(RGBA) and 3(RGB)-channel are supported (XF_8UC4 and XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input Y plane of size (ROWS, COLS).

_y_image

Output Y image of size (ROWS, COLS).

_u_image

Output U image of size (ROWS, COLS).

_v_image

Output V image of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of RGBA/RGB to YUV4 for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

9

589

328

96

Performance Estimate

The following table summarizes the performance of RGBA/RGB to YUV4 for different configurations, as generated using the Vivado HLS 2019.1 version for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

1.89

RGBA/RGB to IYUV

The rgba2iyuv function converts a 4-channel RGBA image to IYUV (4:2:0) format and the rgb2iyuv function converts a 3-channel RGB image to IYUV (4:2:0) format. The function outputs Y, U, and V planes separately. IYUV holds subsampled data, Y is sampled for every RGBA/RGB pixel and U,V are sampled once for 2row and 2column(2x2) pixels. U and V planes are of (rows/2)*(columns/2) size, by cascading the consecutive rows into a single row the planes size becomes (rows/4)*columns.

API Syntax

template <int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void rgba2iyuv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<DST_T, ROWS/4, COLS, NPC> & _u_image, xf::cv::Mat<DST_T, ROWS/4, COLS, NPC> & _v_image)

template <int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void rgb2iyuv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<DST_T, ROWS/4, COLS, NPC> & _u_image, xf::cv::Mat<DST_T, ROWS/4, COLS, NPC> & _v_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table (rgba/rgb)2iyuv Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit,unsigned, 4(RGBA) and 3(RGB)-channel are supported (XF_8UC4 and XF_8UC3).

DST_T

Output pixel type. Only 8-bit,unsigned, 1-channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input Y plane of size (ROWS, COLS).

_y_image

Output Y image of size (ROWS, COLS).

_u_image

Output U image of size (ROWS/4, COLS).

_v_image

Output V image of size (ROWS/4, COLS).

Resource Utilization

The following table summarizes the resource utilization of RGBA/RGB to IYUV for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

9

816

472

149

Performance Estimate

The following table summarizes the performance of RGBA/RGB to IYUV for different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

1.8

RGBA to NV12

The rgba2nv12 function converts a 4-channel RGBA image to NV12 (4:2:0) format. The function outputs Y plane and interleaved UV plane separately. NV12 holds the subsampled data, Y is sampled for every RGBA pixel and U, V are sampled once for 2row and 2columns (2x2) pixels. UV plane is of (rows/2)*(columns/2) size as U and V values are interleaved.

API Syntax

template <int SRC_T, int Y_T, int UV_T, int ROWS, int COLS, int NPC=1>
void rgba2nv12(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC> & _uv)

Parameter Descriptions

The following table describes the template and the function parameters.

Table rgba2nv12 Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit,unsigned, 4-channel is supported (XF_8UC4).

Y_T

Output pixel type. Only 8-bit,unsigned, 1-channel is supported (XF_8UC1).

UV_T

Output pixel type. Only 8-bit,unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input RGBA image of size (ROWS, COLS).

_y

Output Y image of size (ROWS, COLS).

_uv

Output UV image of size (ROWS/2, COLS/2).

Resource Utilization

The following table summarizes the resource utilization of RGBA to NV12 for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

9

802

452

128

Performance Estimate

The following table summarizes the performance of RGBA to NV12 for different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

1.8

RGBA to NV21

The rgba2nv21 function converts a 4-channel RGBA image to NV21 (4:2:0) format. The function outputs Y plane and interleaved VU plane separately. NV21 holds subsampled data, Y is sampled for every RGBA pixel and U, V are sampled once for 2 row and 2 columns (2x2) RGBA pixels. UV plane is of (rows/2)*(columns/2) size as V and U values are interleaved.

API Syntax

template <int SRC_T, int Y_T, int UV_T, int ROWS, int COLS, int NPC=1>
void rgba2nv21(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC> & _uv)

Parameter Descriptions

The following table describes the template and the function parameters.

Table rgba2nv21 Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 4-channel is supported (XF_8UC4).

Y_T

Output pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Output pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input RGBA image of size (ROWS, COLS).

_y

Output Y image of size (ROWS, COLS).

_uv

Output UV image of size (ROWS/2, COLS/2).

Resource Utilization

The following table summarizes the resource utilization of RGBA to NV21 for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

9

802

453

131

Performance Estimate

The following table summarizes the performance of RGBA to NV21 for different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

1.89

YUYV to RGBA

The yuyv2rgba function converts a single-channel YUYV (YUV 4:2:2) image format to a 4-channel RGBA image. YUYV is a sub-sampled format, a set of YUYV value gives 2 RGBA pixel values. YUYV is represented in 16-bit values where as, RGBA is represented in 32-bit values.

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void yuyv2rgba(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table yuyv2rgba Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 16-bit, unsigned, 1-channel is supported (XF_16UC1).

DST_T

Output pixel type. Only 8-bit, unsigned, 4-channel is supported (XF_8UC4).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 incase of 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input image of size (ROWS, COLS).

_dst

Output image of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of YUYV to RGBA for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

6

765

705

165

Performance Estimate

The following table summarizes the performance of UYVY to RGBA for different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

YUYV to NV12

The yuyv2nv12 function converts a single-channel YUYV (YUV 4:2:2) image format to NV12 (YUV 4:2:0) format. YUYV is a sub-sampled format, 1 set of YUYV value gives 2 Y values and 1 U and V value each.

API Syntax

template<int SRC_T,int Y_T,int UV_T,int ROWS,int COLS,int NPC=1,int NPC_UV=1>
void yuyv2nv12(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y_image,xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & _uv_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table yuyv2nv12 Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 16-bit, unsigned, 1-channel is supported (XF_16UC1).

Y_T

Output pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Output UV image pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

NPC_UV

Number of UV image Pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input image of size (ROWS, COLS).

_y_image

Output Y plane of size (ROWS, COLS).

_uv_image

Output U plane of size (ROWS/2, COLS/2).

Resource Utilization

The following table summarizes the resource utilization of YUYV to NV12 for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

831

491

149

8 Pixel

150

0

0

1196

632

161

Performance Estimate

The following table summarizes the performance of YUYV to NV12 for different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

YUYV to IYUV

The yuyv2iyuv function converts a single-channel YUYV (YUV 4:2:2) image format to IYUV(4:2:0) format. Outputs of the function are separate Y, U, and V planes. YUYV is a sub-sampled format, 1 set of YUYV value gives 2 Y values and 1 U and V value each. U, V values of the odd rows are dropped as U, V values are sampled once for 2 rows and 2 columns in the IYUV(4:2:0) format.

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void yuyv2iyuv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<DST_T, ROWS/4, COLS, NPC> & _u_image, xf::cv::Mat<DST_T, ROWS/4, COLS, NPC> & _v_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table yuyv2iyuv Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 16-bit, unsigned,1 channel is supported (XF_16UC1).

DST_T

Output pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel modes.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input image of size (ROWS, COLS).

_y_image

Output Y plane of size (ROWS, COLS).

_u_image

Output U plane of size (ROWS/4, COLS).

_v_image

Output V plane of size (ROWS/4, COLS).

Resource Utilization

The following table summarizes the resource utilization of YUYV to IYUV for different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

835

497

149

8 Pixel

150

0

0

1428

735

210

Performance Estimate

The following table summarizes the performance of YUYV to IYUV for different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

UYVY to IYUV

The uyvy2iyuv function converts a UYVY (YUV 4:2:2) single-channel image to the IYUV format. The outputs of the functions are separate Y, U, and V planes. UYVY is sub sampled format. One set of UYVY value gives two Y values and one U and V value each.

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void uyvy2iyuv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _y_image,xf::cv::Mat<DST_T, ROWS/4, COLS, NPC> & _u_image, xf::cv::Mat<DST_T, ROWS/4, COLS, NPC> & _v_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table uyvy2iyuv Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 16-bit, unsigned, 1-channel is supported (XF_16UC1).

DST_T

Output pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input image of size (ROWS, COLS).

_y_image

Output Y plane of size (ROWS, COLS).

_u_image

Output U plane of size (ROWS/4, COLS).

_v_image

Output V plane of size (ROWS/4, COLS).

Resource Utilization

The following table summarizes the resource utilization of UYVY to IYUV for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Table uyvy2iyuv Function Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

835

494

139

8 Pixel

150

0

0

1428

740

209

Performance Estimate

The following table summarizes the performance of UYVY to IYUV for different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

UYVY to RGBA

The uyvy2rgba function converts a UYVY (YUV 4:2:2) single-channel image to a 4-channel RGBA image. UYVY is sub sampled format, 1set of UYVY value gives 2 RGBA pixel values. UYVY is represented in 16-bit values where as RGBA is represented in 32-bit values.

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void uyvy2rgba(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table uyvy2rgba Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 16-bit, unsigned, 1-channel is supported (XF_16UC1).

DST_T

Output pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input image of size (ROWS, COLS).

_dst

Output image of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of UYVY to RGBA for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

6

773

704

160

Performance Estimate

The following table summarizes the performance of UYVY to RGBA for different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.8

UYVY to NV12

The uyvy2nv12 function converts a UYVY (YUV 4:2:2) single-channel image to NV12 format. The outputs are separate Y and UV planes. UYVY is sub sampled format, 1 set of UYVY value gives 2 Y values and 1 U and V value each.

API Syntax

template<int SRC_T, int Y_T, int UV_T, int ROWS, int COLS, int NPC=1, int NPC_UV=1>
void uyvy2nv12(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y_image,xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & _uv_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table uyvy2nv12 Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 16-bit, unsigned, 1-channel is supported (XF_16UC1).

Y_T

Output pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Output UV image pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

NPC_UV

Number of UV image Pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC4 for 1 pixel and 8 pixel operations respectively.

_src

Input image of size (ROWS, COLS).

_y_image

Output Y plane of size (ROWS, COLS).

_uv_image

Output U plane of size (ROWS/2, COLS/2).

Resource Utilization

The following table summarizes the resource utilization of UYVY to NV12 for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

831

488

131

8 Pixel

150

0

0

1235

677

168

Performance Estimate

The following table summarizes the performance of UYVY to NV12 for different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

IYUV to RGBA/RGB

The iyuv2rgba function converts single channel IYUV (YUV 4:2:0) image to a 4-channel RGBA image and iyuv2rgb function converts single channel IYUV (YUV 4:2:0) image to a 3-channel RGB image . The inputs to the function are separate Y, U, and V planes. IYUV is sub sampled format, U and V values are sampled once for 2 rows and 2 columns of the RGBA/RGB pixels. The data of the consecutive rows of size (columns/2) is combined to form a single row of size (columns).

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void iyuv2rgba(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & src_u,xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & src_v, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst0)

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void iyuv2rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & src_u,xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & src_v, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst0)

Parameter Descriptions

The following table describes the template and the function parameters.

Table iyuv2(rgba/rgb) Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

DST_T

Output pixel type. Only 8-bit, unsigned, 4(RGBA) and 3(RGB)-channel are supported (XF_8UC4 and XF_8UC3).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src_y

Input Y plane of size (ROWS, COLS).

src_u

Input U plane of size (ROWS/4, COLS).

src_v

Input V plane of size (ROWS/4, COLS).

_dst0

Output RGBA image of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of IYUV to RGBA/RGB for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

2

5

1208

728

196

Performance Estimate

The following table summarizes the performance of IYUV to RGBA/RGB for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

IYUV to NV12

The iyuv2nv12 function converts single channel IYUV image to NV12 format. The inputs are separate U and V planes. There is no need of processing Y plane as both the formats have a same Y plane. U and V values are rearranged from plane interleaved to pixel interleaved.

API Syntax

template<int SRC_T, int UV_T, int ROWS, int COLS, int NPC =1, int NPC_UV=1>
void iyuv2nv12(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & src_u,xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & src_v,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & _uv_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table iyuv2nv12 Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Output pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8 for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

NPC_UV

Number of UV Pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC4 for 1 pixel and 4-pixel operations respectively.

src_y

Input Y plane of size (ROWS, COLS).

src_u

Input U plane of size (ROWS/4, COLS).

src_v

Input V plane of size (ROWS/4, COLS).

_y_image

Output V plane of size (ROWS, COLS).

_uv_image

Output UV plane of size (ROWS/2, COLS/2).

Resource Utilization

The following table summarizes the resource utilization of IYUV to NV12 for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image..

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

12

907

677

158

8 Pixel

150

0

12

1591

1022

235

Performance Estimate

The following table summarizes the performance of IYUV to NV12 for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

IYUV to YUV4

The iyuv2yuv4 function converts a single channel IYUV image to a YUV444 format. Y plane is same for both the formats. The inputs are separate U and V planes of IYUV image and the outputs are separate U and V planes of YUV4 image. IYUV stores subsampled U,V values. YUV format stores U and V values for every pixel. The same U, V values are duplicated for 2 rows and 2 columns (2x2) pixels in order to get the required data in the YUV444 format.

API Syntax

template<int SRC_T, int ROWS, int COLS, int NPC=1>
void iyuv2yuv4(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & src_u,xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & src_v,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _u_image, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _v_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table iyuv2yuv4 Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8, for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src_y

Input Y plane of size (ROWS, COLS).

src_u

Input U plane of size (ROWS/4, COLS).

src_v

Input V plane of size (ROWS/4, COLS).

_y_image

Output Y image of size (ROWS, COLS).

_u_image

Output U image of size (ROWS, COLS).

_v_image

Output V image of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of IYUV to YUV4 for different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

1398

870

232

8 Pixel

150

0

0

2134

1214

304

Performance Estimate

The following table summarizes the performance of IYUV to YUV4 for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

13.8

8 pixel operation (150 MHz)

3.4

NV12 to IYUV

The nv122iyuv function converts NV12 format to IYUV format. The function inputs the interleaved UV plane and the outputs are separate U and V planes. There is no need of processing the Y plane as both the formats have a same Y plane. U and V values are rearranged from pixel interleaved to plane interleaved.

API Syntax

template<int SRC_T, int UV_T, int ROWS, int COLS, int NPC=1, int NPC_UV=1>
void nv122iyuv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & src_uv,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _y_image,xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & _u_image,xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & _v_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table nv122iyuv Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Input pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8 pixel mode).

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

NPC_UV

Number of UV image Pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC4 for 1 pixel and 4-pixel operations respectively.

src_y

Input Y plane of size (ROWS, COLS).

src_uv

Input UV plane of size (ROWS/2, COLS/2).

_y_image

Output Y plane of size (ROWS, COLS).

_u_image

Output U plane of size (ROWS/4, COLS).

_v_image

Output V plane of size (ROWS/4, COLS).

Resource Utilization

The following table summarizes the resource utilization of NV12 to IYUV for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

1344

717

208

8 Pixel

150

0

1

1961

1000

263

Performance Estimate

The following table summarizes the performance of NV12 to IYUV for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

NV12 to RGBA

The nv122rgba function converts NV12 image format to a 4-channel RGBA image. The inputs to the function are separate Y and UV planes. NV12 holds sub sampled data, Y plane is sampled at unit rate and 1 U and 1 V value each for every 2x2 Y values. To generate the RGBA data, each U and V value is duplicated (2x2) times.

API Syntax

template<int SRC_T, int UV_T, int DST_T, int ROWS, int COLS, int NPC=1>
void nv122rgba(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y,xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC> & src_uv,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst0)

Parameter Descriptions

The following table describes the template and the function parameters.

Table nv122rgba Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Input pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

DST_T

Output pixel type. Only 8-bit,unsigned,4channel is supported (XF_8UC4).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8, for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src_y

Input Y plane of size (ROWS, COLS).

src_uv

Input UV plane of size (ROWS/2, COLS/2).

_dst0

Output RGBA image of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of NV12 to RGBA for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

2

5

1191

708

195

Performance Estimate

The following table summarizes the performance of NV12 to RGBA for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

NV12 to YUV4

The nv122yuv4 function converts a NV12 image format to a YUV444 format. The function outputs separate U and V planes. Y plane is same for both the image formats. The UV planes are duplicated 2x2 times to represent one U plane and V plane of the YUV444 image format.

API Syntax

template<int SRC_T,int UV_T, int ROWS, int COLS, int NPC=1, int NPC_UV=1>
void nv122yuv4(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & src_uv,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _u_image,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _v_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table nv122yuv4 Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Input pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8 pixel mode).

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

NPC_UV

Number of UV image Pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC4 for 1 pixel and 4-pixel operations respectively.

src_y

Input Y plane of size (ROWS, COLS).

src_uv

Input UV plane of size (ROWS/2, COLS/2).

_y_image

Output Y plane of size (ROWS, COLS).

_u_image

Output U plane of size (ROWS, COLS).

_v_image

Output V plane of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of NV12 to YUV4 for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

1383

832

230

8 Pixel

150

0

0

1772

1034

259

Performance Estimate

The following table summarizes the performance of NV12 to YUV4 for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

13.8

8 pixel operation (150 MHz)

3.4

NV21 to IYUV

The nv212iyuv function converts a NV21 image format to an IYUV image format. The input to the function is the interleaved VU plane only and the outputs are separate U and V planes. There is no need of processing Y plane as both the formats have same the Y plane. U and V values are rearranged from pixel interleaved to plane interleaved.

API Syntax

template<int SRC_T, int UV_T, int ROWS, int COLS, int NPC=1,int NPC_UV=1>
void nv212iyuv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & src_uv,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & _u_image,xf::cv::Mat<SRC_T, ROWS/4, COLS, NPC> & _v_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table nv212iyuv Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Input pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image .

COLS

Maximum width of input and output image. Must be a multiple of 8, for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

NPC_UV

Number of UV image Pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC4 for 1 pixel and 4-pixel operations respectively.

src_y

Input Y plane of size (ROWS, COLS).

src_uv

Input UV plane of size (ROWS/2, COLS/2).

_y_image

Output Y plane of size (ROWS, COLS).

_u_image

Output U plane of size (ROWS/4, COLS).

_v_image

Output V plane of size (ROWS/4, COLS).

Resource Utilization

The following table summarizes the resource utilization of NV21 to IYUV for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

1377

730

219

8 Pixel

150

0

1

1975

1012

279

Performance Estimate

The following table summarizes the performance of NV21 to IYUV for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

NV21 to RGBA

The nv212rgba function converts a NV21 image format to a 4-channel RGBA image. The inputs to the function are separate Y and VU planes. NV21 holds sub sampled data, Y plane is sampled at unit rate and one U and one V value each for every 2x2 Yvalues. To generate the RGBA data, each U and V value is duplicated (2x2) times.

API Syntax

template<int SRC_T, int UV_T, int DST_T, int ROWS, int COLS, int NPC=1>
void nv212rgba(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC> & src_uv,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst0)

Parameter Descriptions

The following table describes the template and the function parameters.

Table nv212rgba Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Input pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

DST_T

Output pixel type. Only 8-bit, unsigned, 4-channel is supported (XF_8UC4).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be a multiple of 8, incase of 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src_y

Input Y plane of size (ROWS, COLS).

src_uv

Input UV plane of size (ROWS/2, COLS/2).

_dst0

Output RGBA image of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of NV21 to RGBA for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

2

5

1170

673

183

Performance Estimate

The following table summarizes the performance of NV12 to RGBA for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

NV21 to YUV4

The nv212yuv4 function converts an image in the NV21 format to a YUV444 format. The function outputs separate U and V planes. Y plane is same for both formats. The UV planes are duplicated 2x2 times to represent one U plane and V plane of YUV444 format.

API Syntax

template<int SRC_T, int UV_T, int ROWS, int COLS, int NPC=1,int NPC_UV=1>
void nv212yuv4(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & src_uv, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _y_image, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _u_image, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _v_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table nv212yuv4 Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Input pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8 pixel mode).

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

NPC_UV

Number of UV image Pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC4 for 1 pixel and 4-pixel operations respectively.

src_y

Input Y plane of size (ROWS, COLS).

src_uv

Input UV plane of size (ROWS/2, COLS/2).

_y_image

Output Y plane of size (ROWS, COLS).

_u_image

Output U plane of size (ROWS, COLS).

_v_image

Output V plane of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of NV21 to YUV4 for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

1383

817

233

8 Pixel

150

0

0

1887

1087

287

Performance Estimate

The following table summarizes the performance of NV21 to YUV4 for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

13.8

8 pixel operation (150 MHz)

3.5

RGB to GRAY

The rgb2gray function converts a 3-channel RGB image to GRAY format.

Y= 0.299*R+0.587*G+0.114*B

Where,

  • Y = Gray pixel

  • R= Red channel

  • G= Green channel

  • B= Blue channel

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void rgb2gray(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table RGB2GRAY Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image.

NPC

Number of pixels to be processed per cycle.

_src

RGB input image

_dst

GRAY output image

Resource Utilization

The following table summarizes the resource utilization of RGB to GRAY for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

3

439

280

Performance Estimate

The following table summarizes the performance of RGB to GRAY for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Table RGB2GRAY Function Performance Estimate Summary +—————————–+——————+ | Operating Mode | Latency Estimate | + +——————+ | | Max Latency (ms) | +=============================+==================+ | 1 pixel operation (300 MHz) | 6.9 | +—————————–+——————+

BGR to GRAY

The bgr2gray function converts a 3-channel BGR image to GRAY format.

Y= 0.299*R+0.587*G+0.114*B

Where,

  • Y = Gray pixel

  • R= Red channel

  • G= Green channel

  • B= Blue channel

API Syntax

template<int SRC_T, int DST_T, int ROWS, int COLS, int NPC=1>
void bgr2gray(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table bgr2gray Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned,1-channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle.

_src

BGR input image

_dst

GRAY output image

Resource Utilization

The following table summarizes the resource utilization of BGR to GRAY for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

3

439

280

Performance Estimate

The following table summarizes the performance of BGR to GRAY for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Table bgr2gray Function Performance Estimate Summary

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

GRAY to RGB

The gray2rgb function converts a gray intensity image to RGB color format.

R<-Y, G<-Y, B<-Y

  • Y = Gray pixel

  • R= Red channel

  • G= Green channel

  • B= Blue channel

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void gray2rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table gray2rgb Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle.

_src

GRAY input image.

_dst

RGB output image.

Resource Utilization

The following table summarizes the resource utilization of gray2rgb for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

0

156

184

Performance Estimate

The following table summarizes the performance of gray2rgb for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

GRAY to BGR

The gray2bgr function converts a gray intensity image to RGB color format.

R<-Y, G<-Y, B<-Y

Where,

  • Y = Gray pixel

  • R= Red channel

  • G= Green channel

  • B= Blue channel

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>
void gray2bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table gray2bgr Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle;

_src

GRAY input image.

_dst

BGR output image.

Resource Utilization

The following table summarizes the resource utilization of gray2bgr for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

0

156

184

Performance Estimate

The following table summarizes the performance of gray2bgr for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

HLS to RGB/BGR

The hls2(rgb/bgr) function converts HLS color space to 3-channel RGB/BGR image.
image33
image34
image35
image36
image37
image38

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void hls2rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void hls2bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table HLS2RGB/BGR Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle.

_src

HLS input image.

_dst

RGB/BGR output image.

Resource Utilization

The following table summarizes the resource utilization of HLS2RGB/BGRR for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

3

4366

3096

Performance Estimate

The following table summarizes the performance of HLS2RGB/BGR for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

RGB to XYZ

The rgb2xyz function converts a 3-channel RGB image to XYZ color space.
image39

  • R= Red channel

  • G= Green channel

  • B= Blue channel

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void rgb2xyz(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table RGB2XYZ Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported. (XF_8UC3).

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle.

_src

RGB input image.

_dst

XYZ output image.

Resource Utilization

The following table summarizes the resource utilization of RGB to XYZ for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

8

644

380

Performance Estimate

The following table summarizes the performance of RGB to XYZ for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

BGR to XYZ

The bgr2xyz function converts a 3-channel BGR image to XYZ color space.
image40

  • R= Red channel

  • G= Green channel

  • B= Blue channel

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void bgr2xyz(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table RGB2XYZ Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image. Must be a multiple of 8.

COLS

Maximum width of input and output image. Must be a multiple of 8.

NPC

Number of pixels to be processed per cycle.

_src

BGR input image.

_dst

XYZ output image.

Resource Utilization

The following table summarizes the resource utilization of BGR to XYZ for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

8

644

380

Performance Estimate

The following table summarizes the performance of BGR to XYZ for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

RGB/BGR to YCrCb

The (rgb/bgr)2ycrcb function converts a 3-channel RGB image to YCrCb color space.

  • Y = 0.299*R + 0.587*G + 0.114*B

  • Cr= (R-Y)*0.713+delta

  • Cb= (B-Y)*0.564+delta


image41

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void rgb2ycrcb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void bgr2ycrcb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table RGB/BGR2YCrCb Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3)

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3)

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle

_src

RGB/BGR input image

_dst

YCrCb output image

Resource Utilization

The following table summarizes the resource utilization of RGB/BGR2YCrCb for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

5

660

500

Performance Estimate

The following table summarizes the performance of RGB/BGR2YCrCb for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

RGB/BGR to HSV

The (rgb/bgr)2hsv function converts a 3-channel RGB image to HSV color space.
image42
image43
image44
image45

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void rgb2hsv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1> void bgr2hsv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table RGB/BGR2HSV Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle

_src

RGB/BGR input image

_dst

HSV output image

Resource Utilization

The following table summarizes the resource utilization of RGB/BGR2HSV for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

6

8

1582

1274

Performance Estimate

The following table summarizes the performance of RGB/BGR2HSV for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

RGB/BGR to HLS

The (rgb/bgr)2hls function converts a 3-channel RGB image to HLS color space.
image46
image47
image48
image49
image50

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void rgb2hls(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void bgr2hls(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table RGB/BGR2HLS Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle.

_src

RGB/BGR input image.

_dst

HLS output image.

Resource Utilization

The following table summarizes the resource utilization of RGB/BGR2HLS for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

3

4366

3096

Performance Estimate

The following table summarizes the performance of RGB/BGR2HLS for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

YCrCb to RGB/BGR

The ycrcb2(rgb/bgr) function converts YCrCb color space to 3-channel RGB/BGR image.

Where,

  • R= Y+1.403*(Cr-delta)

  • G= Y-0.714*(Cr-delta)-0.344*(cb-delta)

  • B= Y+1.773+(Cb-delta)

image51

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void ycrcb2rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void ycrcb2bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table YCrCb2RGB/BGR Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image. Must be a multiple of 8.

COLS

Maximum width of input and output image. Must be a multiple of 8.

NPC

Number of pixels to be processed per cycle.

_src

YCrCb input image.

_dst

RGB/BGR output image.

Resource Utilization

The following table summarizes the resource utilization of YCrCb2RGB/BGR for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

4

538

575

Performance Estimate

The following table summarizes the performance of YCrCb2RGB/BGR for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

HSV to RGB/BGR

The hsv2(rgb/bgr) function converts HSV color space to 3-channel RGB/BGR image.
image52
image53
image54
image55
image56
image57

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void hsv2rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void hsv2bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table HSV2RGB/BGR Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3)

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3)

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle

_src

HSV input image

_dst

RGB/BGR output image

Resource Utilization

The following table summarizes the resource utilization of HSV2RGB/BGRR for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

8

1543

1006

Performance Estimate

The following table summarizes the performance of HSV2RGB/BGR for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

NV12/NV21 to RGB/ BGR

The nv122rgb/nv122bgr/nv212rgb/nv212bgr converts NV12 image format to a 3-channel RGB/BGR image. The inputs to the function are separate Y and UV planes. NV12 holds sub sampled data, Y plane is sampled at unit rate, and 1 U and 1 V value each for every 2x2 Y values. To generate the RGB data, each U and V value is duplicated (2x2) times.

API Syntax

NV122RGB:

template<int SRC_T,int UV_T,int DST_T,int ROWS,int COLS,int NPC=1,int NPC_UV=1>void nv122rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y,xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & src_uv,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst0)

NV122BGR:

template<int SRC_T,int UV_T,int DST_T,int ROWS,int COLS,int NPC=1,int NPC_UV=1>void nv122bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y,xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & src_uv,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst0)

NV212RGB:

template<int SRC_T,int UV_T,int DST_T,int ROWS,int COLS,int NPC=1,int NPC_UV=1>void nv212rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y,xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & src_uv,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst0)

NV212BGR:

template<int SRC_T,int UV_T,int DST_T,int ROWS,int COLS,int NPC=1,int NPC_UV=1>void nv212bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src_y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & src_uv, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst0)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit,unsigned, 1-channel is supported (XF_8UC1).

UV_T

Input pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be a multiple of NPC for N pixel mode.

NPC

Number of Y Pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2,XF_NPPC4 and XF_NPPC8.

NPC_UV

Number of UV Pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2 and XF_NPPC4.

src_y

Y input image of size(ROWS, COLS)

src_uv

UV output image of size (ROWS/2, COLS/2).

_dst0

Output UV image of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of NV12/NV21 to RGB/ BGR function in Normal mode (1 pixel), as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

2

5

339

289

76

Performance Estimate

The following table summarizes the performance of the kernel in single pixel configuration as generated using Vivado HLS 2018.3 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

NV12 to NV21/NV21 to NV12

The nv122nv21/nv212nv12 function converts a NV12 (YUV4:2:0) to NV21 (YUV4:2:0) or vice versa, where 8-bit Y plane followed by an interleaved U/V plane with 2x2 sub-sampling.

API Syntax

NV122NV21:

template<int SRC_Y,int SRC_UV,int ROWS,int COLS,int NPC=1,int NPC_UV=1>
void nv122nv21(xf::cv::Mat<SRC_Y, ROWS, COLS, NPC> & _y,xf::cv::Mat<SRC_UV, ROWS/2, COLS/2, NPC_UV> & _uv,xf::cv::Mat<SRC_Y, ROWS, COLS, NPC> & out_y,xf::cv::Mat<SRC_UV, ROWS/2, COLS/2, NPC_UV> & out_uv)

NV212NV12:

template<int SRC_Y, int SRC_UV, int ROWS, int COLS, int NPC=1,int NPC_UV=1>void nv212nv12(xf::cv::Mat<SRC_Y, ROWS, COLS, NPC> & _y, xf::cv::Mat<SRC_UV, ROWS/2, COLS/2, NPC_UV> & _uv, xf::cv::Mat<SRC_Y, ROWS, COLS, NPC> & out_y, xf::cv::Mat<SRC_UV, ROWS/2, COLS/2, NPC_UV> & out_uv)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Parameter Description

Parameter

Description

SRC_Y

Input Y pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1)

SRC_UV

Input UV pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2)

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be multiple of N.

NPC_Y

Number of Y pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2,XF_NPPC4 and XF_NPPC8.

NPC_UV

Number of UV Pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2 and XF_NPPC4.

_y

Y input image

_uv

UV input image

out_y

Y output image

out_uv

UV output image

Resource Utilization

The following table summarizes the resource utilization of NV122NV21/NV212NV12 function in Normal mode (1-Pixel), as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

258

161

61

Performance Estimate

The following table summarizes the performance of the kernel in single pixel configuration as generated using Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

NV12/NV21 to UYVY/YUYV

The NV12/NV21 to UYVY/YUYV function converts a NV12/NV21 (YUV4:2:0) image to a single-channel YUYV/UYVY (YUV 4:2:2) image format. YUYV is a sub-sampled format. YUYV/UYVY is represented in 16-bit values whereas, RGB is represented in 24-bit values.

API Syntax

NV122UYVY:

template<int SRC_Y, int SRC_UV, int DST_T, int ROWS, int COLS, int NPC=1,int NPC_UV=1>void nv122uyvy(xf::cv::Mat<SRC_Y, ROWS, COLS, NPC> & _y,xf::cv::Mat<SRC_UV, ROWS/2, COLS/2, NPC_UV> & _uv,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

NV122YUYV:

template<int SRC_Y, int SRC_UV, int DST_T, int ROWS, int COLS, int NPC=1,int NPC_UV=1>void nv122yuyv(xf::cv::Mat<SRC_Y, ROWS, COLS, NPC> & _y, xf::cv::Mat<SRC_UV, ROWS/2, COLS/2, NPC_UV> & _uv, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

NV212UYVY:

template<int SRC_Y, int SRC_UV, int DST_T, int ROWS, int COLS, int NPC=1,int NPC_UV=1>void nv212uyvy(xf::cv::Mat<SRC_Y, ROWS, COLS, NPC> & _y, xf::cv::Mat<SRC_UV, ROWS/2, COLS/2, NPC_UV> & _uv,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

NV212YUYV:

template<int SRC_Y, int SRC_UV, int DST_T,int ROWS, int COLS, int NPC=1,int NPC_UV=1>void nv212yuyv(xf::cv::Mat<SRC_Y, ROWS, COLS, NPC> & _y, xf::cv::Mat<SRC_UV, ROWS/2, COLS/2, NPC_UV> & _uv, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Parameter Description

Parameter

Description

SRC_Y

Input Y image pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

SRC_UV

Input UV image pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

DST_T

Output pixel type. Only 16-bit, unsigned, 1-channel is supported (XF_16UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of NPC.

NPC

Number of pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2,XF_NPPC4 and XF_NPPC8.

NPC_UV

Number of pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2 and XF_NPPC4.

_y

Y input image

_uv

UV input image

_dst

UYVY/YUYV output image

Resource Utilization

The following table summarizes the resource utilization of NV12/NV21 to UYVY/YUYV function in Normal mode(1-Pixel), as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

1

0

337

201

64

Performance Estimate

The following table summarizes the performance of the kernel in single pixel configuration as generated using Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

UYVY/YUYV to RGB/BGR

The yuyv2rgb/yuyv2bgr/uyvy2rgb/uyvy2bgr function converts a single-channel YUYV/UYVY (YUV 4:2:2) image format to a 3- channel RGB/BGR image. YUYV/UYVY is a sub-sampled format, a set of YUYV/UYVY values gives 2 RGB pixel values. YUYV/UYVY is represented in 16-bit values whereas, RGB/BGR is represented in 24-bit values.

API Syntax

YUYV2RGB:

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void yuyv2rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

YUYV2BGR:

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void yuyv2bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

UYVY2RGB

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void uyvy2rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

UYVY2BGR:

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void uyvy2bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 16-bit, unsigned,1-channel is supported (XF_16UC1).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be a multiple of NPC for N pixel mode.

NPC

Number of Y pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2,XF_NPPC4 and XF_NPPC8.

_src

Input image of size(ROWS, COLS)

_dst

Output image of size (ROWS, COLS).

Resource Utilization

The following table summarizes the resource utilization of UYVY/YUYV to RGB/BGR function in Normal mode(1-Pixel), as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

6

444

486

109

Performance Estimate

The following table summarizes the performance of the kernel in single pixel configuration as generated using Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

UYVY to YUYV/ YUYV to UYVY

The yuyv2uyvy/uyvy2yuyv function converts a YUYV (YUV4:2:2) to UYVY (YUV4:2:2) or vice versa, where 8-bit Y plane followed by an interleaved U/V plane with 2x2 sub sampling.

API Syntax

UYVY2YUYV:

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void uyvy2yuyv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & uyvy,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & yuyv)

YUYV2UYVY:

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void yuyv2uyvy(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & yuyv,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & uyvy)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Parameter Description

Parameter

Description

SRC_T

Input Y pixel type. Only 16-bit, unsigned, 1-channel is supported (XF_16UC1).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be a multiple of N.

NPC

Number of pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2,XF_NPPC4 and XF_NPPC8.

yuyv

Input image

uyvy

Output image

Resource Utilization

The following table summarizes the resource utilization of UYVY to YUYV/ YUYV to UYVY function in Normal mode (1 pixel), as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

368

176

109

Performance Estimate

The following table summarizes the performance of the kernel in single pixel configuration as generated using Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

UYVY/YUYV to NV21

The UYVY/YUYV2NV21 function converts a single-channel YUYV/UYVY (YUV 4:2:2) image format to NV21 (YUV 4:2:0) format. YUYV/UYVY is a sub-sampled format, 1 set of YUYV/UYVY value gives 2 Y values and 1 U and V value each.

API Syntax

UYVY2NV21:

template<int SRC_T,int Y_T,int UV_T,int ROWS,int COLS,int NPC=1,int NPC_UV=1>void uyvy2nv21(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y_image,xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & _uv_image)

YUYV2NV21:

template<int SRC_T,int Y_T,int UV_T,int ROWS,int COLS,int NPC=1,int NPC_UV=1>void yuyv2nv21(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y_image,xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & _uv_image)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 16-bit, unsigned,1-channel is supported (XF_16UC1).

Y_T

Output Y image pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Output UV image pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of NPC.

NPC

Number of pixels to be processed per cycle; Possible options are XF_NPPC1,XF_NPPC2,XF_NPPC4 and XF_NPPC8.

NPC_UV

Number of U, V Pixels to be processed per cycle; Possible options are XF_NPPC1,XF_NPPC2 and XF_NPPC4.

_src

Input image

_y_image

Y Output image

_uv_image

UV Output image

Resource Utilization

The following table summarizes the resource utilization of UYVY/YUYV to NV21 function in Normal mode (1 pixel), as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

215

73

42

Performance Estimate

The following table summarizes the performance of the kernel in single pixel configuration as generated using Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

RGB/ BGR to NV12/NV21

The rgb2nv12/bgr2nv12/rgb2nv21/bgr2nv21 converts a 3-channel RGB/BGR image to NV12/NV21 (4:2:0) format. The function outputs Y plane and interleaved UV/VU plane separately. NV12/NV21 holds the subsampled data, Y is sampled for every RGB/BGR pixel and U, V are sampled once for 2 rows and 2 columns (2x2) pixels. UV/VU plane is of (rows/2)*(columns/2) size as U and V values are interleaved.

API Syntax

RGB2NV12

template <int SRC_T, int Y_T, int UV_T, int ROWS, int COLS, int NPC=1,int NPC_UV=1>void rgb2nv12(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & _uv)

BGR2NV12

template <int SRC_T, int Y_T, int UV_T, int ROWS, int COLS, int NPC=1,int NPC_UV=1>void bgr2nv12(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & _uv)

RGB2NV21

template <int SRC_T, int Y_T, int UV_T, int ROWS, int COLS, int NPC=1,int NPC_UV=1>void rgb2nv21(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & _uv)

BGR2NV21

template <int SRC_T, int Y_T, int UV_T, int ROWS, int COLS, int NPC=1,int NPC_UV=1>void bgr2nv21(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<Y_T, ROWS, COLS, NPC> & _y, xf::cv::Mat<UV_T, ROWS/2, COLS/2, NPC_UV> & _uv)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

Y_T

Output pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

UV_T

Output pixel type. Only 8-bit, unsigned, 2-channel is supported (XF_8UC2).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be a multiple of NPC for N pixel mode.

NPC

Number of Pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2,XF_NPPC4 and XF_NPPC8.

NPC_UV

Number of Pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2 and XF_NPPC4

_src

RGB input image of size(ROWS,COLS)

_y

Output Y image of size (ROWS, COLS).

_uv

Output UV image of size (ROWS/2, COLS/2).

Resource Utilization

The following table summarizes the resource utilization of RGB/BGR to NV12/NV21 function in Normal mode (1-Pixel), as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

9

413

279

66

Performance Estimate

The following table summarizes the performance of the kernel in single pixel configuration as generated using Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

BGR to RGB / RGB to BGR

The bgr2rgb/rgb2bgr function converts a 3-channel BGR to RGB format or RGB to BGR format.

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void bgr2rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void rgb2bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of N.

NPC

Number of Pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2,XF_NPPC4 and XF_NPPC8.

_src

BGR/RGB input image

_dst

RGB/BGR output image

Resource Utilization

The following table summarizes the resource utilization of RGB to BGR/ BGR to RGB function in Normal mode (1-Pixel), as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

0

317

118

98

Performance Estimate

The following table summarizes the performance of the kernel in single pixel configuration as generated using Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

RGB/BGR to UYVY/YUYV

The RGB/BGR to UYVY/YUYV function converts a 3- channel RGB/BGR image to a single-channel YUYV/UYVY (YUV 4:2:2) image format. YUYV is a sub-sampled format, 2 RGBA pixel gives set of YUYV/UYVY values. YUYV/UYVY is represented in 16-bit values whereas, RGB is represented in 24-bit values

API Syntax

RGB to UYVY:

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void rgb2uyvy(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

RGB to YUYV:

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void rgb2yuyv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

BGR to UYVY:

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void bgr2uyvy(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

BGR to YUYV:

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void bgr2yuyv(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3)

DST_T

Output pixel type. Only 16-bit, unsigned, 1-channel is supported (XF_16UC1)

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be multiple of NPC.

NPC

Number of pixels to be processed per cycle. Possible options are XF_NPPC1,XF_NPPC2,XF_NPPC4 and XF_NPPC8..

_src

RGB/BGR input image

_dst

UYVY/YUYV output image

Resource Utilization

The following table summarizes the resource utilization of RGB/BGR to UYVY/YUYV function in normal mode(1-Pixel), as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

9

249

203

55

Performance Estimate

The following table summarizes the performance of the kernel in single pixel configuration as generated using Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

XYZ to RGB/BGR

The xyz2rgb function converts XYZ color space to 3-channel RGB image.
image58

API Syntax

template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void xyz2rgb(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)template<int SRC_T,int DST_T,int ROWS,int COLS,int NPC=1>void xyz2bgr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table XYZ2RGB/BGR Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 3-channel is supported (XF_8UC3).

ROWS

Maximum height of input and output image. Must be multiple of 8.

COLS

Maximum width of input and output image. Must be multiple of 8.

NPC

Number of pixels to be processed per cycle.

_src

XYZ input image.

_dst

RGB/BGR output image.

Resource Utilization

The following table summarizes the resource utilization of XYZ2RGB/BGR for different configurations, as generated in the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

300

0

8

639

401

Performance Estimate

The following table summarizes the performance of XYZ2RGB/BGR for different configurations, as generated using the Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1, to process a HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

Color correction matrix

Color correction matrix algorithm converts the input image color format to output image color format using the colorcorrection matrix provided by the user (CCM_TYPE).

API Syntax

template <int CCM_TYPE, int SRC_T, int DST_T, int ROWS, int COLS, int NPC = 1>
void colorcorrectionmatrix(xf::cv::Mat<SRC_T, ROWS, COLS, NPC>& _src_mat,
                           xf::cv::Mat<SRC_T, ROWS, COLS, NPC>& _dst_mat)

Parameter Descriptions

The following table describes template parameters and arguments of the function.

Table colorcorrectionmatrix correction Parameter Description

Parameter

Description

CCM_TYPE

colorcorrection matrix.

SRC_T

Input pixel type. 8/10/12/16-bit unsigned, 3 channel are supported (XF_8UC3, XF_10UC3, XF_12UC3, XF_16UC3).

DST_T

Output pixel type. 8/10/12/16-bit unsigned, 3 channel are supported (XF_8UC3, XF_10UC3, XF_12UC3, XF_16UC3).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options is XF_NPPC1, XF_NPPC2 AND so on

_src_mat

Input image

_dst_mat

Output image

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vitis HLS 2020.2 tool, to process a FULL HD image.

Table colorcorrectionmatrix correction Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel-8U

300

0

9

283

254

73

1 pixel-16U

300

0

9

353

239

75

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vitis HLS 2020.2 tool, to process a FULL HD image.

Table colorcorrectionmatrix correction Function Performance Estimate Summary

Operating Mode

Operating Frequency (MHz)

Latency Estimate Max (ms)

1 pixel

300

7

2 pixel

300

3.6

Color Thresholding

The colorthresholding function compares the color space values of the source image with low and high threshold values, and returns either 255 or 0 as the output.

API Syntax

template<int SRC_T,int DST_T,int MAXCOLORS, int ROWS, int COLS,int NPC>
          void colorthresholding(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_mat,unsigned char low_thresh[MAXCOLORS*3], unsigned char high_thresh[MAXCOLORS*3])

Parameter Descriptions

The table below describes the template and the function parameters.

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 3 channel is supported (XF_8UC3).

DST_T

Output pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1).

MAXCOLORS

Maximum number of color values

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be a multiple of 8, for 8 pixel mode.

NPC

Number of pixels to be processed per cycle. Only XF_NPPC1 supported.

_src_mat

Input image

_dst_mat

Thresholded image

low_thresh

Lowest threshold values for the colors

high_thresh

Highest threshold values for the colors

Compare

The Compare function performs the per element comparison of pixels in two corresponding images src1, src2 and stores the result in dst.

dst(x,y)=src1(x,y) CMP_OP src2(x,y)

CMP_OP – a flag specifies correspondence between the pixels.

  • XF_CMP_EQ : src1 is equal to src2

  • XF_CMP_GT : src1 is greater than src2

  • XF_CMP_GE : src1 is greater than or equal to src2

  • XF_CMP_LT : src1 is less than src2

  • XF_CMP_LE : src1 is less than or equal to src2

  • XF_CMP_NE : src1 is unequal to src2

If the comparison result is true, then the corresponding element of dst is set to 255; else it is set to 0.

API Syntax

template<int CMP_OP,  int SRC_T , int ROWS, int COLS, int NPC=1>
void compare(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src2, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Compare Parameter Description

Parameter

Description

CMP_OP

The flag that specify the relation between the elements needs to be checked

SRC_T

Input Pixel Type. 8-bit, unsigned, 1 channel is supported (XF_8UC1)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

First input image

_src2

Second input image

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the Compare XF_CMP_NE configuration in Resource optimized (8 pixels) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

87

60

LUT

38

84

CLB

16

20

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

CompareS

The CompareS function performs the comparison of a pixel in the input image (src1) and the given scalar value scl, and stores the result in dst.

dst(x,y)=src1(x,y) CMP_OP scalar

CMP_OP – a flag specifies correspondence between the pixel and the scalar.

  • XF_CMP_EQ : src1 is equal to scl

  • XF_CMP_GT : src1 is greater than scl

  • XF_CMP_GE : src1 is greater than or equal to scl

  • XF_CMP_LT : src1 is less than scl

  • XF_CMP_LE : src1 is less than or equal to scl

  • XF_CMP_NE : src1 is unequal to scl

If the comparison result is true, then the corresponding element of dst is set to 255, else it is set to 0.

API Syntax

template<int CMP_OP,  int SRC_T , int ROWS, int COLS, int NPC=1>
void compareS(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, unsigned char _scl[XF_CHANNELS(SRC_T,NPC)], xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table CompareS Parameter Description

Parameter

Description

CMP_OP

The flag that specifying the relation between the elements to be checked

SRC_T

Input pixel type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. In case of N-pixel parallelism, the width should be a multiple of N

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixels operations respectively.

_src1

First input image

_scl

Input scalar value, the size should be number of channels

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the CompareS function with XF_CMP_NE configuration in Resource optimized (8 pixels) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

93

93

LUT

39

68

CLB

21

28

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

convertScaleAbs

The convertScaleAbs function converts an input image src with optional linear transformation, save the result as image dst.

dst(x,y)= src1(x,y)*scale+shift

API Syntax

template< int SRC_T,int DST_T, int ROWS, int COLS, int NPC = 1>
void convertScaleAbs(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src1, xf::cv::Mat<DST_T, ROWS, COLS, NPC> & dst,float scale, float shift)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . convertScaleAbs Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1).

DST_T

Output pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

scale

Scale factor

shift

Delta/shift added to scaled value.

dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the convertScaleAbs function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

10

38

FF

949

1971

LUT

1052

1522

CLB

218

382

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image…

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Crop

The Crop function extracts the region of interest (ROI) from the input image.

P(X,Y) ≤ P(xi, yi) ≤ P(X’,Y’)

  • P(X,Y) - Top left corner of ROI

  • P(X’,Y’) - Bottom Right of ROI

API Syntax

template<int SRC_T, int ROWS, int COLS,int ARCH_TYPE=0,int NPC=1>
void crop(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<SRC_T, ROWS, COLS, NPC>  &_dst_mat,xf::cv::Rect_<unsigned int> &roi)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Crop Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of 8 for 8-pixel operation.

ARCH_TYPE

Architecture type. 0 resolves to stream implementation and 1 resolves to memory mapped implementation.

NPC

Number of pixels to be processed per cycle. NPC should be power of 2.

_src_mat

Input image

_dst_mat

Output ROI image

roi

ROI is a xf::cv::Rect object that consists of top left corner of the rectangle along with the height and width of the rectangle.

Resource Utilization

The following table summarizes the resource utilization of crop function in normal mode (NPC=1) for 3 ROIs (480x640, 100x200, 300x300) in memory mapped mode as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

300 MHz

BRAM_18K

0

0

DSP48E

2

2

FF

1163

2358

LUT

909

3140

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image for 3 ROIs (480x640, 100x200, 300x300).

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

1.7

8 pixel operation (150 MHz)

0.6

Multiple ROI Extraction

You can call the xf::cv::crop function multiple times in accel.cpp.

Multiple ROI Extraction Example

void crop_accel(xf::cv::Mat<TYPE, HEIGHT, WIDTH, NPIX> &_src,xf::cv::Mat<TYPE,HEIGHT, WIDTH, NPIX> _dst[NUM_ROI],xf::cv::Rect_<unsigned int> roi[NUM_ROI])

{xf::cv::crop<TYPE, TYPE, HEIGHT, WIDTH, NPIX>(_src, _dst[0],roi[0]); xf::cv::crop<TYPE, TYPE, HEIGHT, WIDTH, NPIX>(_src, _dst[1],roi[1]); xf::cv::crop<TYPE, TYPE, HEIGHT, WIDTH, NPIX>(_src, _dst[2],roi[2]);}

Custom Convolution

The filter2D function performs convolution over an image using a user-defined kernel.

Convolution is a mathematical operation on two functions f and g, producing a third function, The third function is typically viewed as a modified version of one of the original functions, that gives the area overlap between the two functions to an extent that one of the original functions is translated.

The filter can be unity gain filter or a non-unity gain filter. The filter must be of type XF_16SP. If the co-efficients are floating point, it must be converted into the Qm.n and provided as the input as well as the shift parameter has to be set with the ‘n’ value. Else, if the input is not of floating point, the filter is provided directly and the shift parameter is set to zero.

API Syntax

template<int BORDER_TYPE,int FILTER_WIDTH,int FILTER_HEIGHT, int SRC_T,int DST_T, int ROWS, int COLS,int NPC=1>
void filter2D(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_mat,short int filter[FILTER_HEIGHT*FILTER_WIDTH],unsigned char _shift)

Parameter Descriptions

The following table describes the template and the function parameters.

Table filter2D Parameter Description

Parameter

Description

BORDER_TYPE

Border Type supported is XF_BORDER_CONSTANT

FILTER_HEIGHT

Number of rows in the input filter

FILTER_WIDTH

Number of columns in the input filter

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

DST_T

Output pixel type. 8-bit unsigned single and 3 channels (XF_8UC1, XF_8UC3) and 16-bit signed single and 3 channels (XF_16SC1, XF_16SC3) supported.

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be multiple of 8, for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src_mat

Input image

_dst_mat

Output image

filter

The input filter of any size, provided the dimensions should be an odd number. The filter co-efficients either a 16-bit value or a 16-bit fixed point equivalent value.

_shift

The filter must be of type XF_16SP. If the co-efficients are floating point, it must be converted into the Qm.n and provided as the input as well as the shift parameter has to be set with the ‘n’ value. Else, if the input is not of floating point, the filter is provided directly and the shift parameter is set to zero.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

3x3

300

3

9

1701

1161

269

5x5

300

5

25

3115

2144

524

8 Pixel

3x3

150

6

72

2783

2768

638

5x5

150

10

216

3020

4443

1007

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K 3 Channel image.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

3x3

300

18

27

886

801

8 Pixel

5x5

300

30

75

1793

1445

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Filter Size

Latency Estimate

Max (ms)

1 pixel

300

3x3

7

300

5x5

7.1

8 pixel

150

3x3

1.86

150

5x5

1.86

Delay

In image processing pipelines, it is possible that the inputs to a function with FIFO interfaces are not synchronized. That is, the first data packet for first input might arrive a finite number of clock cycles after the first data packet of the second input. If the function has FIFOs at its interface with insufficient depth, this causes the whole design to stall on hardware. To synchronize the inputs, we provide this function to delay the input packet that arrives early, by a finite number of clock cycles.

API Syntax

template<int MAXDELAY, int SRC_T, int ROWS, int COLS,int NPC=1 >
          void delayMat(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The table below describes the template and the function parameters.

Parameter

Description

SRC_T

Input and output pixel type

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8 pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

MAXDELAY

Maximum delay that the function is to be instantiated for.

_src

Input image

_dst

Output image

Demosaicing

The Demosaicing function converts a single plane Bayer pattern output, from the digital camera sensors to a color image. This function implements an improved bi-linear interpolation technique proposed by Malvar, He, and Cutler.

The above figure shows the Bayer mosaic for color image capture in single-CCD digital cameras.

API Syntax

template<int BFORMAT, int SRC_T, int DST_T, int ROWS, int COLS, int NPC,bool USE_URAM=false>
void demosaicing(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &src_mat, xf::cv::Mat<DST_T, ROWS, COLS, NPC> &dst_mat)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Demosaicing Parameter Description

Parameter

Description

BFORMAT

Input Bayer pattern. XF_BAYER_BG, XF_BAYER_GB, XF_BAYER_GR, and XF_BAYER_RG are the supported values.

SRC_T

Input pixel type. 8-bit, unsigned,1 channel (XF_8UC1) and 16-bit, unsigned, 1 channel (XF_16UC1) are supported.

DST_T

Output pixel type. 8-bit, unsigned, 4 channel (XF_8UC4) and 16-bit, unsigned, 4 channel (XF_16UC4) are supported.

ROWS

Number of rows in the image being processed.

COLS

Number of columns in the image being processed. Must be multiple of 8, in case of 8 pixel mode.

NPC

Number of pixels to be processed per cycle; single pixel parallelism (XF_NPPC1), two-pixel parallelism (XF_NPPC2) and four-pixel parallelism (XF_NPPC4) are supported. XF_NPPC4 is not supported with XF_16UC1 pixel type.

USE_URAM

Enable to map storage structures to UltraRAM.

_src_mat

Input image

_dst_mat

Output image

. rubric:: Resource Utilization

The following table below shows the resource utilization of the Demosaicing function, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

300

8

0

1906

1915

412

2 Pixel

300

8

0

2876

3209

627

4 Pixel

300

8

0

2950

3222

660

The following table shows the resource utilization of the Demosaicing function, generated using Vivado HLS 2019.1 version tool for the xczu7ev-ffvc1156-2-e FPGA.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

URAM

DSP_48Es

FF

LUT

CLB

1 Pixel

300

0

1

0

1366

1399

412

Performance Estimate

The following table shows the performance in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 to process a 4K (3840x2160) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

27.82

2 pixel operation (300 MHz)

13.9

4 pixel operation

(300 MHz, 8-bit image only)

6.95

Dilate

During a dilation operation, the current pixel intensity is replaced by the maximum value of the intensity in a nxn neighborhood of the current pixel.


image59

API Syntax

template<int BORDER_TYPE, int TYPE, int ROWS, int COLS,int K_SHAPE,int K_ROWS,int K_COLS, int ITERATIONS, int NPC=1>
void dilate (xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _src, xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _dst,unsigned char _kernel[K_ROWS*K_COLS])

Parameter Descriptions

The following table describes the template and the function parameters.

Table dilate Parameter Description

Parameter

Description

BORDER_TYPE

Border Type supported is XF_BORDER_CONSTANT

TYPE

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be multiple of 8, for 8-pixel operation)

K_SHAPE

Shape of the kernel . The supported kernel shapes are RECT, CROSS, and ELLIPSE.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

K_ROWS

Height of the kernel.

K_COLS

Width of the kernel.

ITERATIONS

Number of times the dilation is applied. Currently supporting for Rectangular shape kernel element.

_src_mat

Input image

_dst_mat

Output image

_kernel

Dilation kernel of size K_ROWS * K_COLS.

Resource Utilization

The following table summarizes the resource utilization of the Dilation function with rectangle shape structuring element in 1 pixel operation and 8 pixel operation, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA for HD (1080X1920) image.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

3

6

DSP48E

0

0

FF

411

657

LUT

392

1249

CLB

96

255

Performance Estimate

The following table summarizes the resource utilization of the Dilation function with rectangle shape structuring element in 1 pixel operation, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA for 4K 3channel image.

Name

Resource Utilization

1 pixel per clock operation

300 MHz

BRAM_18K

18

DSP48E

0

FF

983

LUT

745

CLB

186

The following table summarizes a performance estimate of the Dilation function for Normal Operation (1 pixel) and Resource Optimized (8 pixel) configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

Latency Estimate

Min Latency (ms)

Max Latency (ms)

1 pixel operation (300 MHz)

7.0

7.0

8 pixel operation (150 MHz)

1.87

1.87

Distance Transform Feature Matcher

The distance transform is an operator normally only applied to binary images, where in this case the image must be coded as zero and non-zero pixels as a grayscale image. The result of the transform is a graylevel image that looks similar to the input image, except that the graylevel intensities of points inside foreground regions are changed to show the distance to the closest boundary from each point.

This Xilinx implementation applies 3x3 mask, of distance type DIST_L2 (Euclidean distance), with horizontal/vertical shift cost, a = 0.955, and diagonal shift cost b = 1.3693.

Computing the distance takes two passes, forward and backward. During the forward pass, forward mask is applied, and while the backward pass the backward mask is applied over the forward pass data. In this implementation, it is required to pass a cache memory for the kernel to interact (write while forward pass, read while backward pass). The cache memory must be of image dimensions and of type ap_uint<32>.

API Syntax

template <int IN_PTR, int FW_PTR, int ROWS, int COLS, int USE_URAM>
void distanceTransform(ap_uint<IN_PTR>* _src,
                       float* _dst, ap_uint<FW_PTR>* _fw_pass,
                       int rows, int cols)

Parameter Descriptions

The following table describes template paramters and arguments of the function.

Table distance-transform Parameter Description

Parameter

Description

IN_PTR

Input pointer width must be ‘8’.

FW_PTR

Forward pass data pointer width must be ‘32’.

ROWS

Maximum number of rows of the input image that the hardware kernel must be built for.

COLS

Maximum number of columns of the input image that the hardware kernel must be built for.

USE_URAM

Default is ‘0’. Can be set to ‘1’, if the device has URAM support.

_src

Grayscale input image pointer, of ap_uint<8>* type.

_dst

The distance image pointer,of type float*.

_fw_pass

Forward pass pointer, of type ap_uint<32>. This is used as an intermediary cache, between forward and backward passes.

rows

Number of rows in the input image, must be less than ROWS.

cols

Number of cols in the input image, must be less than COLS.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vitis 2020.2 tool, to process a 4K image.

Table distance-transform Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

default

300

22

0

5129

7444

1757

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vitis 2020.2 tool, to process a 4K image.

Table distance-transform Function Performance Estimate Summary

Operating Mode

Operating Frequency (MHz)

Latency Estimate Max (ms)

default

200

86.249

Duplicate

When various functions in a pipeline are implemented by a programmable logic, FIFOs are instantiated between two functions for dataflow processing. When the output from one function is consumed by two functions in a pipeline, the FIFOs need to be duplicated. This function facilitates the duplication process of the FIFOs.

API Syntax

template<int SRC_T, int ROWS, int COLS,int NPC=1>
          void duplicateMat(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst1,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst2)

Parameter Descriptions

The table below describes the template and the function parameters.

Paramete r

Description

SRC_T

Input and output pixel type

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle. Possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input image

_dst1

Duplicate output for _src

_dst2

Duplicate output for _src

Erode

The erode function finds the minimum pixel intensity in the NXN neighborhood of a pixel and replaces the pixel intensity with the minimum value.


image60

API Syntax

template<int BORDER_TYPE, int TYPE, int ROWS, int COLS,int K_SHAPE,int K_ROWS,int K_COLS, int ITERATIONS, int NPC=1>
void erode (xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _src, xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _dst,unsigned char _kernel[K_ROWS*K_COLS])

Parameter Descriptions

The following table describes the template and the function parameters.

Table erode Parameter Description

Parameter

Description

BORDER_TYPE

Border type supported is XF_BORDER_CONSTANT

TYPE

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be multiple of 8, for 8-pixel operation)

K_SHAPE

Shape of the kernel . The supported kernel shapes are RECT,CROSS and ELLIPSE.

K_ROWS

Height of the kernel.

K_COLS

Width of the kernel.

ITERATIONS

Number of times the erosion is applied.Currently supporting for Rectangular shape kernel element.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src_mat

Input image

_dst_mat

Output image

_kernel

Erosion kernel of size K_ROWS * K_COLS.

Resource Utilization

The following table summarizes the resource utilization of the Erosion function with rectangular shape structuring element generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA,for FullHD image(1080x1920).

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

3

6

DSP48E

0

0

FF

411

657

LUT

392

1249

CLB

96

255

The following table summarizes the resource utilization of the Erosion function with rectangular shape structuring element generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA,for 4K image with 3channels.

Name

Resource Utilization

1 pixel per clock operation

300 MHz

BRAM_18K

18

DSP48E

0

FF

983

LUT

3745

CLB

186

Performance Estimate

The following table summarizes a performance estimate of the Erosion function for Normal Operation (1 pixel) and Resource Optimized (8 pixel) configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

Latency Estimate

Min Latency (ms)

Max Latency (ms)

1 pixel operation (300 MHz)

7.0

7.0

8 pixel operation (150 MHz)

1.85

1.85

FAST Corner Detection

Features from accelerated segment test (FAST) is a corner detection algorithm, that is faster than most of the other feature detectors.

The fast function picks up a pixel in the image and compares the intensity of 16 pixels in its neighborhood on a circle, called the Bresenham’s circle. If the intensity of 9 contiguous pixels is found to be either more than or less than that of the candidate pixel by a given threshold, then the pixel is declared as a corner. Once the corners are detected, the non-maximal suppression is applied to remove the weaker corners.

This function can be used for both still images and videos. The corners are marked in the image. If the corner is found in a particular location, that location is marked with 255, otherwise it is zero.

API Syntax

template<int NMS,int SRC_T,int ROWS, int COLS,int NPC=1>
void fast(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst_mat,unsigned char _threshold)

Parameter Descriptions

The following table describes the template and the function parameters.

Table fast Parameter Description

Parameter

Description

NMS

If NMS == 1, non-maximum suppression is applied to detected corners (keypoints). The value should be 0 or 1.

SRC_T

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1)

ROWS

Maximum height of input image.

COLS

Maximum width of input image (must be a multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src_mat

Input image

_dst_mat

Output image. The corners are marked in the image.

_threshol d

Threshold on the intensity difference between the center pixel and its neighbors. Usually it is taken around 20.

Resource Utilization

The following table summarizes the resource utilization of the kernel for different configurations, generated using Vivado HLS 2019.1 for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image with NMS.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

10

20

DSP48E

0

0

FF

2695

7310

LUT

3792

20956

CLB

769

3519

Performance Estimate

The following table summarizes the performance of kernel for different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image with non-maximum suppression (NMS).

Operating Mode

Operating Frequency (MHz)

Filter Size

Latency Estimate

Max (ms)

1 pixel

300

3x3

7

8 pixel

150

3x3

1.86

Gaincontrol

The gain control modules improve the overall brightness of the input image. In this module, applying a multiplicative gain (weight) for red and blue channel of the input bayerized image.

API Syntax

template< int BFORMAT,int SRC_T, int ROWS, int COLS, int NPC = 1>void gaincontrol(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src1, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & dst)

The following table describes the template and the function parameters.

Parameter

Description

BFORMAT

Input Bayer pattern.

SRC_T

Input and Output Pixel Type.

ROWS

Maximum height of input and output image (Must be multiple of NPC)

COLS

Maximum width of input and output image (Must be multiple of NPC)

NPC

Number of Pixels to be processed per cycle.

Src1

Input Bayer image

dst

Output Bayer image

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.2 tool for the Xilinx xc7vx485t-ffg1157-1 FPGA, to process 4K image.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

SLICE

1 pixel

300

0

3

233

95

59

2 pixel

300

0

3

235

95

59

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vivado HLS 2019.2 tool for the Xilinx xc7vx485t-ffg1157-1, to process a 4K image.

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

Max (ms)

1 pixel

300

27.7

2 pixel

300

14.2

Gammacorrection

The gamma correction modules improves the overall brightness of image. The gamma loouptable is generated using the gamma value and with following equation.

image162

image163

API Syntax

template <int SRC_T, int DST_T, int ROWS, int COLS, int NPC = 1>void gammacorrection(xf::cv::Mat<SRC_T, ROWS, COLS, NPC>& src,xf::cv::Mat<DST_T, ROWS, COLS, NPC>& dst,float gammaval)

The following table describes the template and the function parameters.

Parameter

Description

SRC_T

Input Pixel Type.

DST_T

Output Pixel Type.

ROWS

Maximum height of input and output image (Must be multiple of NPC)

COLS

Maximum width of input and output image (Must be multiple of NPC)

NPC

Number of Pixels to be processed per cycle.

src

Input image

dst

Output image

Gammaval

Input gamma value for brightness increase.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.2 tool for the Xilinx xc7vx485t-ffg1157-1 FPGA, to process a 4K image.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

13

31

4261

3213

1295

4 pixel

300

22

32

4297

3232

1261

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vivado HLS 2019.2 tool for the Xilinx xc7vx485t-ffg1157-1 FPGA, to process a 4K image.

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

Max (ms)

1 pixel

300

27.9

4 pixel

300

7

Gaussian Filter

The GaussianBlur function applies Gaussian blur on the input image. Gaussian filtering is done by convolving each point in the input image with a Gaussian kernel.


image61

Where image62,image63 are the mean values and image64, image65 are the variances in x and y directions respectively. In the GaussianBlur function, values of image66, image67 are considered as zeroes and the values of image68, image69 are equal.

API Syntax

template<int FILTER_SIZE, int BORDER_TYPE, int SRC_T, int ROWS, int COLS, int NPC =  1>
void GaussianBlur(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & dst, float sigma)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . GaussianBlur Parameter Description

Parameter

Description

FILTER_SIZE

Filter size. Filter size of 3 (XF_FILTER_3X3), 5 (XF_FILTER_5X5) and 7 (XF_FILTER_7X7) are supported.

BORDER_TYPE

Border type supported is XF_BORDER_CONSTANT

SRC_T

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible values are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src

Input image

dst

Output image

sigma

Standard deviation of Gaussian filter

Resource Utilization

The following table summarizes the resource utilization of the Gaussian Filter in different configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to progress a grayscale HD (1080x1920) image.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

3x3

300

3

17

3641

2791

610

5x5

300

5

27

4461

3544

764

7x7

250

7

35

4770

4201

894

8 Pixel

3x3

150

6

52

3939

3784

814

5x5

150

10

111

5688

5639

1133

7x7

150

14

175

7594

7278

1518

The following table summarizes the resource utilization of the Gaussian Filter in different configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to progress a 4K 3 Channel image.

Operating Mode

Filter Size

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

LUT

CLB

1 Pixel

3x3

300

18

33

4835

3742

5x5

300

30

51

5755

3994

7x7

300

42

135

8086

5422

Performance Estimate

The following table summarizes a performance estimate of the Gaussian Filter in different configurations, as generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Filter Size

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

3x3

7.01

5x5

7.03

7x7

7.06

8 pixel operation (150 MHz)

3x3

1.6

5x5

1.7

7x7

1.74

Gradient Magnitude

The magnitude function computes the magnitude for the images. The input images are x-gradient and y-gradient images of type 16S. The output image is of same type as the input image.

For L1NORM normalization, the magnitude computed image is the pixel-wise added image of absolute of x-gradient and y-gradient, as shown below:.


image70

For L2NORM normalization, the magnitude computed image is as follows:


image71

API Syntax

template< int NORM_TYPE ,int SRC_T,int DST_T, int ROWS, int COLS,int NPC=1>
void magnitude(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_matx,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _src_maty,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_mat)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . magnitude Parameter Description

Parameter

Description

NORM_TYPE

Normalization type can be either L1 or L2 norm. Values are XF_L1NORM or XF_L2NORM

SRC_T

Input pixel type. Only 16-bit, signed, 1 channel is supported (XF_16SC1)

DST_T

Output pixel type. Only 16-bit, signed,1 channel is supported (XF_16SC1)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible values are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src_matx

First input, x-gradient image.

_src_maty

Second input, y-gradient image.

_dst_mat

Output, magnitude computed image.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image and for L2 normalization.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

2

16

FF

707

2002

LUT

774

3666

CLB

172

737

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image and for L2 normalization.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

7.2

8 pixel operation (150 MHz)

1.7

Gradient Phase

The phase function computes the polar angles of two images. The input images are x-gradient and y-gradient images of type 16S. The output image is of same type as the input image.

For radians:

image72

For degrees:

image73

API Syntax

template<int RET_TYPE ,int SRC_T,int DST_T, int ROWS, int COLS,int NPC=1 >
void phase(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_matx,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _src_maty,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_mat)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . phase Parameter Description

Parameter

Description

RET_TYPE

Output format can be either in radians or degrees. Options are XF_RADIANS or XF_DEGREES.

  • If the XF_RADIANS option is selected, phase API will return result in Q4.12 format. The output range is (0, 2 pi).

  • If the XF_DEGREES option is selected, xFphaseAPI will return result in Q10.6 degrees and output range is (0, 360).

SRC_T

Input pixel type. Only 16-bit, signed, 1 channel is supported (XF_16SC1).

DST_T

Output pixel type. Only 16-bit, signed, 1 channel is supported (XF_16SC1)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src_matx

First input, x-gradient image.

_src_maty

Second input, y-gradient image.

_dst_mat

Output, phase computed image.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

6

24

DSP48E

6

19

FF

873

2396

LUT

753

3895

CLB

187

832

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Table . phase Function Performance Estimate Summary

Operating Mode

Operating Frequency (MHz)

Latency Estimate (ms)

1 pixel

300

7.2

8 pixel

150

1.7

Deviation from OpenCV

In phase implementation, the output is returned in a fixed point format. If XF_RADIANS option is selected, phase API will return result in Q4.12 format. The output range is (0, 2 pi). If XF_DEGREES option is selected, phase API will return result in Q10.6 degrees and output range is (0, 360).

Harris Corner Detection

In order to understand Harris Corner Detection, let us consider a grayscale image. Sweep a window w(x,y) (with displacements u in the x-direction and v in the y-direction), I calculates the variation of intensity w(x,y).


image74

Where:

  • w(x,y) is the window position at (x,y)

  • I(x,y) is the intensity at (x,y)

  • I(x+u,y+v) is the intensity at the moved window (x+u,y+v).

Since we are looking for windows with corners, we are looking for windows with a large variation in intensity. Hence, we have to maximize the equation above, specifically the term:


image75

Using Taylor expansion:


image76

Expanding the equation and cancelling I(x,y) with -I(x,y):


image77

The above equation can be expressed in a matrix form as:


image78

So, our equation is now:


image79

A score is calculated for each window, to determine if it can possibly contain a corner:


image80
Where,

  • image81

  • image82

API Syntax

Non-Maximum Suppression:

In non-maximum suppression (NMS) if radius = 1, then the bounding box is 2*r+1 = 3.

In this case, consider a 3x3 neighborhood across the center pixel. If the center pixel is greater than the surrounding pixel, then it is considered a corner. The comparison is made with the surrounding pixels, which are within the radius.

Radius = 1

x-1, y-1

x-1, y

x-1, y+1

x, y-1

x, y

x, y+1

x+1, y-1

x+1, y

x+1, y+1

Threshold:

A threshold=442, 3109 and 566 is used for 3x3, 5x5, and 7x7 filters respectively. This threshold is verified over 40 sets of images. The threshold can be varied, based on the application. The corners are marked in the output image. If the corner is found in a particular location, that location is marked with 255, otherwise it is zero.

template<int FILTERSIZE,int BLOCKWIDTH, int NMSRADIUS,int SRC_T,int ROWS, int COLS,int NPC=1,bool USE_URAM=false>
void cornerHarris(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & dst,uint16_t threshold, uint16_t k)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . cornerHarris Parameter Description

Parameter

Description

FILTERSIZE

Size of the Sobel filter. 3, 5, and 7 supported.

BLOCKWIDTH

Size of the box filter. 3, 5, and 7 supported.

NMSRADIUS

Radius considered for non-maximum suppression. Values supported are 1 and 2.

TYPE

Input pixel type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

ROWS

Maximum height of input image.

COLS

Maximum width of input image (must be multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

USE_URAM

Enable to map some storage structures to URAM

src

Input image

dst

Output image.

threshold

Threshold applied to the corner measure.

k

Harris detector parameter

Resource Utilization

The following table summarizes the resource utilization of the Harris corner detection in different configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=3 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

33

66

DSP48E

10

80

FF

3254

9330

LUT

3522

13222

CLB

731

2568

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=5 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

45

90

DSP48E

10

80

FF

5455

12459

LUT

5695

24594

CLB

1132

4498

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=7 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

57

114

DSP48E

10

80

FF

8783

16593

LUT

9157

39813

CLB

1757

6809

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=3 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

35

70

DSP48E

10

80

FF

4656

11659

LUT

4681

17394

CLB

1005

3277

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=5 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

47

94

DSP48E

10

80

FF

6019

14776

LUT

6337

28795

CLB

1353

5102

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=7 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

59

118

DSP48E

10

80

FF

9388

18913

LUT

9414

43070

CLB

1947

7508

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=3 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

37

74

DSP48E

11

88

FF

6002

13880

LUT

6337

25573

CLB

1327

4868

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=5 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

49

98

DSP48E

11

88

FF

7410

17049

LUT

8076

36509

CLB

1627

6518

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=7 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

61

74

DSP48E

11

88

FF

10714

21137

LUT

11500

51331

CLB

2261

8863

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=3 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

41

82

DSP48E

10

80

FF

5519

10714

LUT

5094

16930

CLB

1076

3127

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=5 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

53

106

DSP48E

10

80

FF

6798

13844

LUT

6866

28286

CLB

1383

4965

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=7 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

65

130

DSP48E

10

80

FF

10137

17977

LUT

10366

43589

CLB

1940

7440

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=3 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

43

86

DSP48E

10

80

FF

5957

12930

LUT

5987

21187

CLB

1244

3922

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=5 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

55

110

DSP48E

10

80

FF

5442

16053

LUT

6561

32377

CLB

1374

5871

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=7 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

67

134

DSP48E

10

80

FF

10673

20190

LUT

10793

46785

CLB

2260

8013

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=3 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

45

90

DSP48E

11

88

FF

7341

15161

LUT

7631

29185

CLB

1557

5425

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=5 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

57

114

DSP48E

11

88

FF

8763

18330

LUT

9368

40116

CLB

1857

7362

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=7 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

69

138

DSP48E

11

88

FF

12078

22414

LUT

12831

54652

CLB

2499

9628

Resource Utilization with URAM enable

This section summarizes the resource utilization of the Harris corner detection in different configurations, generated using Vivado HLS 2019.1 version tool for the xczu7ev-ffvc1156-2-e FPGA, to process a grayscale 4K (3840X2160) image.

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=3 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

12

12

URAM

4

21

DSP48E

10

80

FF

5306

11846

LUT

3696

13846

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=5 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

12

12

URAM

7

30

DSP48E

10

80

FF

7625

13899

LUT

5596

27136

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=7 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

12

12

URAM

7

42

DSP48E

10

80

FF

12563

19919

LUT

8816

39087

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=3 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

12

12

URAM

4

23

DSP48E

10

80

FF

6689

15022

LUT

4506

18719

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=5 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

12

12

URAM

7

32

DSP48E

10

80

FF

9050

17063

LUT

6405

31992

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=7 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

12

12

URAM

7

44

DSP48E

10

80

FF

13946

23116

LUT

9626

44738

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=3 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

12

12

URAM

4

25

DSP48E

11

88

FF

8338

17378

LUT

6151

24844

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=5 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

12

12

URAM

7

34

DSP48E

11

88

FF

10497

19457

LUT

7858

39762

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=7 and NMS_RADIUS =1.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

12

12

URAM

7

46

DSP48E

11

88

FF

15393

25450

LUT

11080

50662

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=3 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

20

20

URAM

4

21

DSP48E

10

80

FF

6286

13441

LUT

4704

18072

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=5 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

20

20

URAM

7

30

DSP48E

10

80

FF

8626

15498

LUT

6606

31371

The following table summarizes the resource utilization for Sobel Filter = 3, Box filter=7 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

20

20

URAM

7

42

DSP48E

10

80

FF

13543

21522

LUT

9853

43301

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=3 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

20

20

URAM

4

23

DSP48E

10

80

FF

7670

16750

LUT

5513

22854

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=5 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

20

20

URAM

7

32

DSP48E

10

80

FF

9712

18793

LUT

7338

36136

The following table summarizes the resource utilization for Sobel Filter = 5, Box filter=7 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

20

20

URAM

7

44

DSP48E

10

80

FF

14650

24846

LUT

10558

48866

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=3 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

20

20

URAM

4

25

DSP48E

11

88

FF

9562

19101

LUT

7405

29986

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=5 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

20

20

URAM

7

34

DSP48E

11

88

FF

11751

21180

LUT

9254

44024

The following table summarizes the resource utilization for Sobel Filter = 7, Box filter=7 and NMS_RADIUS =2.

Name

Resource Utilization

1 pixel

8 pixel

300 MHz

150 MHz

BRAM_18K

20

20

URAM

7

46

DSP48E

11

88

FF

16723

27156

LUT

12474

54858

Performance Estimate

The following table summarizes a performance estimate of the Harris corner detection in different configurations, as generated using Vivado HLS 2019.1 tool for Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency

(MHz)

Configuration

Latency Estimate

Sobel

Box

NMS Radius

Latency(In ms)

1 pixel

300 MHz

3

3

1

7

1 pixel

300 MHz

3

5

1

7.1

1 pixel

300 MHz

3

7

1

7.1

1 pixel

300 MHz

5

3

1

7.2

1 pixel

300 MHz

5

5

1

7.2

1 pixel

300 MHz

5

7

1

7.2

1 pixel

300 MHz

7

3

1

7.22

1 pixel

300 MHz

7

5

1

7.22

1 pixel

300 MHz

7

7

1

7.22

8 pixel

150 MHz

3

3

1

1.7

8 pixel

150 MHz

3

5

1

1.7

8 pixel

150 MHz

3

7

1

1.7

8 pixel

150 MHz

5

3

1

1.71

8 pixel

150 MHz

5

5

1

1.71

8 pixel

150 MHz

5

7

1

1.71

8 pixel

150 MHz

7

3

1

1.8

8 pixel

150 MHz

7

5

1

1.8

8 pixel

150 MHz

7

7

1

1.8

1 pixel

300 MHz

3

3

2

7.1

1 pixel

300 MHz

3

5

2

7.1

1 pixel

300 MHz

3

7

2

7.1

1 pixel

300 MHz

5

3

2

7.21

1 pixel

300 MHz

5

5

2

7.21

1 pixel

300 MHz

5

7

2

7.21

1 pixel

300 MHz

7

3

2

7.22

1 pixel

300 MHz

7

5

2

7.22

1 pixel

300 MHz

7

7

2

7.22

8 pixel

150 MHz

3

3

2

1.8

8 pixel

150 MHz

3

5

2

1.8

8 pixel

150 MHz

3

7

2

1.8

8 pixel

150 MHz

5

3

2

1.81

8 pixel

150 MHz

5

5

2

1.81

8 pixel

150 MHz

5

7

2

1.81

8 pixel

150 MHz

7

3

2

1.9

8 pixel

150 MHz

7

5

2

1.91

8 pixel

150 MHz

7

7

2

1.92

Deviation from OpenCV

In Vitis Vision, thresholding and NMS are included, but in OpenCV they are not included. In Vitis Vision, all the blocks are implemented in fixed point. Whereas, in OpenCV, all the blocks are implemented in floating point.

Histogram Computation

The calcHist function computes the histogram of given input image.
image83
Where, H is the array of 256 elements.

API Syntax

template<int SRC_T,int ROWS, int COLS,int NPC=1>
void calcHist(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, uint32_t *histogram)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . calcHist Parameter Description

Parameter

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle

_src

Input image

histogram

Output array of 256 elements

Resource Utilization

The following table summarizes the resource utilization of the calcHist function for Normal Operation (1 pixel) and Resource Optimized (8 pixel) configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz for 1 pixel case and at 150 MHz for 8 pixel mode.

Name

Resource Utilization

Normal Operation (1 pixel)

Resource Optimized (8 pixel)

BRAM_18K

2

16

DSP48E

0

0

FF

196

274

LUT

240

912

CLB

57

231

The following table summarizes the resource utilization of the calcHist function for Normal Operation (1 pixel), generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz for 1 pixel case for 4K image 3 channel.

Name

Resource Utilization

Normal Operation (1 pixel)

BRAM_18K

8

DSP48E

0

FF

381

LUT

614

CLB

134

Performance Estimate

The following table summarizes a performance estimate of the calcHist function for Normal Operation (1 pixel) and Resource Optimized (8 pixel) configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz for 1 pixel and 150 MHz for 8 pixel mode.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Histogram Equalization

The equalizeHist function performs histogram equalization on input image or video. It improves the contrast in the image, to stretch out the intensity range. This function maps one distribution (histogram) to another distribution (a wider and more uniform distribution of intensity values), so the intensities are spread over the whole range.

For histogram H[i], the cumulative distribution H’[i] is given as:


image84

The intensities in the equalized image are computed as:


image85

API Syntax

template<int SRC_T, int ROWS, int COLS, int NPC = 1>
void equalizeHist(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . equalizeHist Parameter Description

Parameter

Description

SRC_T

Input and output pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle

_src

Input image

_src1

Input image

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the equalizeHist function for Normal Operation (1 pixel) and Resource Optimized (8 pixel) configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz for 1 pixel and 150 MHz for 8 pixel mode.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

4

5

3492

1807

666

8 pixel

150

25

5

3526

2645

835

Performance Estimate

The following table summarizes a performance estimate of the equalizeHist function for Normal Operation (1 pixel) and Resource Optimized (8 pixel) configurations, generated using Vivado HLS 2019.1version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz for 1 pixel and 150 MHz for 8 pixel mode.

Operating Mode

Latency Estimate

Max (ms)

1 pixel per clock operation

13.8

8 pixel per clock operation

3.4

HOG

The histogram of oriented gradients (HOG) is a feature descriptor used in computer vision for the purpose of object detection. The feature descriptors produced from this approach is widely used in the pedestrian detection.

The technique counts the occurrences of gradient orientation in localized portions of an image. HOG is computed over a dense grid of uniformly spaced cells and normalized over overlapping blocks, for improved accuracy. The concept behind HOG is that the object appearance and shape within an image can be described by the distribution of intensity gradients or edge direction.

Both RGB and gray inputs are accepted to the function. In the RGB mode, gradients are computed for each plane separately, but the one with the higher magnitude is selected. With the configurations provided, the window dimensions are 64x128, block dimensions are 16x16.

API Syntax

template<int WIN_HEIGHT, int WIN_WIDTH, int WIN_STRIDE, int BLOCK_HEIGHT, int BLOCK_WIDTH, int CELL_HEIGHT, int CELL_WIDTH, int NOB, int DESC_SIZE, int IMG_COLOR, int OUTPUT_VARIANT, int SRC_T, int DST_T, int ROWS, int COLS, int NPC = XF_NPPC1,bool USE_URAM=false>
void HOGDescriptor(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_in_mat, xf::cv::Mat<DST_T, 1, DESC_SIZE, NPC> &_desc_mat);

Parameter Descriptions

The following table describes the template parameters.

Table . HOGDescriptor Template .. rubric:: Parameter Descriptions

Parameters

Description

WIN_HEIGHT

The number of pixel rows in the window. This must be a multiple of 8 and should not exceed the number of image rows.

WIN_WIDTH

The number of pixel cols in the window. This must be a multiple of 8 and should not exceed the number of image columns.

WIN_STRIDE

The pixel stride between two adjacent windows. It is fixed at 8.

BLOCK_HEIGHT

Height of the block. It is fixed at 16.

BLOCK_WIDTH

Width of the block. It is fixed at 16.

CELL_HEIGHT

Number of rows in a cell. It is fixed at 8.

CELL_WIDTH

Number of cols in a cell. It is fixed at 8.

NOB

Number of histogram bins for a cell. It is fixed at 9

DESC_SIZE

The size of the output descriptor.

IMG_COLOR

The type of the image, set as either XF_GRAY or XF_RGB

OUTPUT_VARIE NT

Must be either XF_HOG_RB or XF_HOG_NRB

SRC_T

Input pixel type. Must be either XF_8UC1 or XF_8UC4, for gray and color respectively.

DST_T

Output descriptor type. Must be XF_32UC1.

ROWS

Number of rows in the image being processed.

COLS

Number of columns in the image being processed.

NPC

Number of pixels to be processed per cycle; this function supports only XF_NPPC1 or 1 pixel per cycle operations.

USE_URAM

Enable to map UltraRAM instead of BRAM for some storage structures.

The following table describes the function parameters.

Where,

  • NO is normal operation (single pixel processing)

  • RB is repetitive blocks (descriptor data are written window wise)

  • NRB is non-repetitive blocks (descriptor data are written block wise, in order to reduce the number of writes).

Note: In the RB mode, the block data is written to the memory taking the overlap windows into consideration. In the NRB mode, the block data is written directly to the output stream without consideration of the window overlap. In the host side, the overlap must be taken care.

Resource Utilization

The following table shows the resource utilization of HOGDescriptor function for normal operation (1 pixel) mode as generated in Vivado HLS 2019.1 version tool for the part Xczu9eg-ffvb1156-1-i-es1 at 300 MHz to process an image of 1920x1080 resolution.

Resource

Utilization (at 300 MHz) of 1 pixel operation

NRB

RB

Gray

RGB

Gray

RGB

BRAM_18K

43

49

171

177

DSP48E

34

46

36

48

FF

15365

15823

15205

15663

LUT

12868

13267

13443

13848

The following table shows the resource utilization of HOGDescriptor function for normal operation (1 pixel) mode as generated in Vivado HLS 2019.1 version tool for the part xczu7ev-ffvc1156-2-e at 300 MHz to process an image of 1920x1080 resolution with UltraRAM enabled.

UltraRAM enabled

Resource

Utilization (at 300 MHz) of 1 pixel operation

NRB

RB

Gray

RGB

Gray

RGB

BRAM_18K

10

12

18

20

URAM

15

15

15

17

DSP48E

34

46

36

48

FF

17285

17917

18270

18871

LUT

12409

12861

12793

13961

Performance Estimate

The following table shows the performance estimates of HOGDescriptor() function for different configurations as generated in Vivado HLS 2019.1 version tool for the part Xczu9eg-ffvb1156-1-i-es1 to process an image of 1920x1080p resolution.

Operating Mode

Operating Frequency (MHz)

Latency Estimate

Min (ms)

Max (ms)

NRB-Gray

300

6.98

8.83

NRB-RGBA

300

6.98

8.83

RB-Gray

300

176.81

177

RB-RGBA

300

176.81

177

Deviations from OpenCV

Listed below are the deviations from the OpenCV:

  1. Border care

    The border care that OpenCV has taken in the gradient computation is BORDER_REFLECT_101, in which the border padding will be the neighboring pixels’ reflection. Whereas, in the Xilinx implementation, BORDER_CONSTANT (zero padding) was used for the border care.

  2. Gaussian weighing

    The Gaussian weights are multiplied on the pixels over the block, that is a block has 256 pixels, and each position of the block are multiplied with its corresponding Gaussian weights. Whereas, in the HLS implementation, gaussian weighing was not performed.

  3. Cell-wise interpolation The magnitude values of the pixels are distributed across different cells in the blocks but on the corresponding bins. image86 Pixels in the region 1 belong only to its corresponding cells, but the pixels in region 2 and 3 are interpolated to the adjacent 2 cells and 4 cells respectively. This operation was not performed in the HLS implementation.

  4. Output handling

    The output of the OpenCV will be in the column major form. In the HLS implementation, output will be in the row major form. Also, the feature vector will be in the fixed point type Q0.16 in the HLS implementation, while in the OpenCV it will be in floating point.

Limitations

  1. The configurations are limited to Dalal’s implementation

  2. Image height and image width must be a multiple of cell height and cell width respectively.

HoughLines

The HoughLines function here is equivalent to HoughLines Standard in OpenCV. The HoughLines function is used to detect straight lines in a binary image. To apply the Hough transform, edge detection preprocessing is required. The input to the Hough transform is an edge detected binary image. For each point (xi,yi) in a binary image, we define a family of lines that go through the point as:

rho= xi cos(theta) + yi sin(theta)

Each pair of (rho,theta) represents a line that passes through the point (xi,yi). These (rho,theta) pairs of this family of lines passing through the point form a sinusoidal curve in (rho,theta) plane. If the sinusoids of N different points intersect in the (rho,theta) plane, then that intersection (rho1, theta1) represents the line that passes through these N points. In the HoughLines function, an accumulator is used to keep the count (also called voting) of all the intersection points in the (rho,theta) plane. After voting, the function filters spurious lines by performing thinning, that is, checking if the center vote value is greater than the neighborhood votes and threshold, then making that center vote as valid and other wise making it zero. Finally, the function returns the desired maximum number of lines (LINESMAX) in (rho,theta) form as output.

The design assumes the origin at the center of the image i.e at (Floor(COLS/2), Floor(ROWS/2)). The ranges of rho and theta are:

theta = [0, pi)

rho=[-DIAG/2, DIAG/2), where DIAG = cvRound{SquareRoot( (COLS*COLS) + (ROWS*ROWS))}

For ease of use, the input angles THETA, MINTHETA and MAXTHETA are taken in degrees, while the output theta is in radians. The angle resolution THETA is declared as an integer, but treated as a value in Q6.1 format (that is, THETA=3 signifies that the resolution used in the function is 1.5 degrees). When the output (rho, ? theta) is used for drawing lines, you should be aware of the fact that origin is at the center of the image.

API Syntax

template<unsigned int RHO,unsigned int THETA,int MAXLINES,int DIAG,int MINTHETA,int MAXTHETA,int SRC_T, int ROWS, int COLS,int NPC>

void HoughLines(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,float outputrho[MAXLINES],float outputtheta[MAXLINES],short threshold,short linesmax)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . HoughLines Parameter Description

Parameter

Description

RHO

Distance resolution of the accumulator in pixels.

THETA

Angle resolution of the accumulator in degrees and Q6.1 format.

MAXLINES

Maximum number of lines to be detected

MINTHETA

Minimum angle in degrees to check lines.

MAXTHETA

Maximum angle in degrees to check lines

DIAG

Diagonal of the image. It should be cvRound(sqrt(rows*rows + cols*cols)/RHO).

SRC_T

Input Pixel Type. Only 8-bit, unsigned, 1-channel is supported (XF_8UC1).

ROWS

Maximum height of input image

COLS

Maximum width of input image

NPC

Number of Pixels to be processed per cycle; Only single pixel supported XF_NPPC1.

_src_mat

Input image should be 8-bit, single-channel binary image.

outputrho

Output array of rho values. rho is the distance from the coordinate origin (center of the image).

outputthe ta

Output array of theta values. Theta is the line rotation angle in radians.

threshold

Accumulator threshold parameter. Only those lines are returned that get enough votes (>threshold).

linesmax

Maximum number of lines.

Resource Utilization

The table below shows the resource utilization of the kernel for different configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 to process a grayscale HD (1080x1920) image for 512 lines.

Name

Resource Utilization

THETA=1, RHO=1

BRAM_18K

542

DSP48E

10

FF

60648

LUT

56131

Performance Estimate

The following table shows the performance of kernel for different configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 to process a grayscale HD (1080x1920) image for 512 lines.

Operating Mode

Operating Frequency (MHz)

Latency Estimate

Max (ms)

THETA=1, RHO=1

300

12.5

Preprocessing for Deep Neural Networks

The input image are typically pre-processed before being fed for inference of different deep neural networks (DNNs). The preProcess function provides various modes to perform various preprocessing operations. The preprocessing function\(\ f(x\)) can be described using below equations.

image164

The preProcess function supports operating modes presented in the below table:

Op Code

Operation

Description

0

image165

Mean subtraction

1

image166

Scale and clip

2

image167

Clipping

3

image168

Scale and bias

4

image169

Scale and bias with mean subtraction

5

image170

Complete operation

API Syntax

template <int INPUT_PTR_WIDTH_T,int OUTPUT_PTR_WIDTH_T, int T_CHANNELS_T, int CPW_T, int ROWS_T, int COLS_T, int NPC_T, bool PACK_MODE_T, int WX_T, int WA_T, int WB_T, int WY_T, int WO_T, int FX_T, int FA_T, int FB_T, int FY_T,int FO_T, bool SIGNED_IN_T, int OPMODE_T>

void preProcess(hls::stream<ap_uint<INPUT_PTR_WIDTH_T> > &srcStrm, ap_uint<OUTPUT_PTR_WIDTH_T> \*out, float params[3*T_CHANNELS_T], int rows, int cols, int th1, int th2)

The following table describes the template and the function parameters.

Parameter

Description

srcStrm

Input image stream

out

Output pointer

params

Array containing α, β and γ values

rows

Input image height

cols

Input image width

th1

Upper threshold

th2

Lower threshold

INPUT_PTR_WIDTH_T

Width of input pointer

OUTPUT_PTR_WIDTH_T

Width of output pointer

T_CHANNELS_T

Total Channels

CPW_T

Channels Packed per DDR Word

ROWS_T

Max Height of Image

COLS_T

Max Width of Image

NPC_T

Number of pixels processed per clock

PACK_MODE_T

data format (pixel packed or channel packed)

WX_T

x bit width

WA_T

alpha bit width

WB_T

beta bit width

WY_T

Gamma bit width

WO_T

Output bit width

FX_T

Number of integer bits for x

FA_T

Number of integer bits for alpha

FB_T

Number of integer bits for beta

FY_T

Number of integer bits for gamma

FO_T

Number of integer bits for output

SIGNED_IN_T

Signed input flag

OPMODE_T

Operating mode

Resource Utilization

The following table summarizes the resource utilization of preProcess for NPC_T =8, CPW_T=3 and OPMODE=0, for a maximum input image size of 1280x720 pixels. The results are after synthesis in Vitis 2019.2 for the Xilinx xcu200-fsgd2104-2-e FPGA at 300 MHz. Latency for this configuration is 0.7 ms.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

SLICE

8 pixel

300

0

2

7554

11127

2155

Pyramid Up

The pyrUp function is an image up-sampling algorithm. It first inserts zero rows and zero columns after every input row and column making up to the size of the output image. The output image size is always image87. The zero padded image is then smoothened using Gaussian image filter. Gaussian filter for the pyramid-up function uses a fixed filter kernel as given below:
image88

However, to make up for the pixel intensity that is reduced due to zero padding, each output pixel is multiplied by 4.

API Syntax

template<int TYPE, int ROWS, int COLS, int NPC>
void pyrUp (xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _src, xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . pyrUp Parameter Description

Parameter

Description

TYPE

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Maximum Height or number of output rows to build the hardware for this kernel

COLS

Maximum Width or number of output columns to build the hardware for this kernel

NPC

Number of pixels to process per cycle. Currently, the kernel supports only 1 pixel per cycle processing (XF_NPPC1).

_src

Input image stream

_dst

Output image stream

Resource Utilization

The following table summarizes the resource utilization of pyrUp for 1 pixel per cycle implementation, for a maximum input image size of 1920x1080 pixels. The results are after synthesis in Vivado HLS 2019.1 for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

1 Pixel

300

1124

1199

0

10

The following table summarizes the resource utilization of pyrUp for 1 pixel per cycle implementation, for a maximum input image size of 4K with BGR. The results are after synthesis in Vivado HLS 2019.1 for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

1 Pixel

300

2074

2176

0

59

Performance Estimate

The following table summarizes performance estimates of pyrUp function on Vivado HLS 2019.1 for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

Operating Frequency

(MHz)

Input Image Size

Latency Estimate

Max (ms)

1 pixel

300

1920x1080

27.82

Pyramid Down

The pyrDown function is an image down-sampling algorithm which smoothens the image before down-scaling it. The image is smoothened using a Gaussian filter with the following kernel:
image89

Down-scaling is performed by dropping pixels in the even rows and the even columns. The resulting image size is image90.

API Syntax

template<int TYPE, int ROWS, int COLS, int NPC,bool USE_URAM=false>
void pyrDown (xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _src, xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . pyrDown Parameter Description

Parameter

Description

TYPE

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Maximum Height or number of input rows to build the hardware for this kernel

COLS

Maximum Width or number of input columns to build the hardware for this kernel

NPC

Number of pixels to process per cycle. Currently, the kernel supports only 1 pixel per cycle processing (XF_NPPC1).

USE_URAM

Enable to map storage structures to UltraRAM

_src

Input image stream

_dst

Output image stream

Resource Utilization

The following table summarizes the resource utilization of pyrDown for 1 pixel per cycle implementation, for a maximum input image size of 1920x1080 pixels. The results are after synthesis in Vivado HLS 2019.1 for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

1 Pixel

300

1171

1238

1

5

The following table summarizes the resource utilization of pyrDown for 1 pixel per cycle implementation, for a maximum input image size of 4K with BGR image. The results are after synthesis in Vivado HLS 2019.1 for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

1 Pixel

300

2158

1983

2

30

The following table summarizes the resource utilization of pyrDown for 1 pixel per cycle implementation, for a maximum input image size of 3840x2160 pixels. The results are after synthesis in Vivado HLS 2019.1 for the Xilinx xczu7eg-ffvb1156-1 FPGA at 300 MHz with UltraRAM enabled.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

URAM

1 Pixel

300

1171

1243

0

0

1

Performance Estimate

The following table summarizes performance estimates of pyrDown function in Vivado HLS 2019.1 for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

Operating Frequency

(MHz)

Input Image Size

Latency Estimate

Max (ms)

1 pixel

300

1920x1080

6.99

InitUndistortRectifyMapInverse

The InitUndistortRectifyMapInverse function generates mapx and mapy, based on a set of camera parameters, where mapx and mapy are inputs for the xf::cv::remap function. That is, for each pixel in the location (u, v) in the destination (corrected and rectified) image, the function computes the corresponding coordinates in the source image (the original image from camera). The InitUndistortRectifyMapInverse module is optimized for hardware, so the inverse of rotation matrix is computed outside the synthesizable logic. Note that the inputs are fixed point, so the floating point camera parameters must be type casted to Q12.20 format.

API Syntax

template< int CM_SIZE, int DC_SIZE, int MAP_T, int ROWS, int COLS, int NPC >
void InitUndistortRectifyMapInverse ( ap_fixed<32,12> *cameraMatrix, ap_fixed<32,12> *distCoeffs, ap_fixed<32,12> *ir, xf::cv::Mat<MAP_T, ROWS, COLS, NPC> &_mapx_mat, xf::cv::Mat<MAP_T, ROWS, COLS, NPC> &_mapy_mat, int _cm_size, int _dc_size)

Parameter Descriptions

The following table describes the template and the function parameters.

Descriptions

Parameter

Description

CM_SIZE

It must be set at the compile time, 9 for 3x3 matrix

DC_SIZE

It must be set at the compile time, must be 4,5 or 8

MAP_T

It is the type of output maps, and must be XF_32FC1

ROWS

Maximum image height, necessary to generate the output maps

COLS

Maximum image width, necessary to generate the output maps

NPC

Number of pixels per cycle. This function supports only one pixel per cycle, so set to XF_NPPC1

cameraMatrix

The input matrix representing the camera in the old coordinate system

distCoeffs

The input distortion coefficients (k1,k2,p1,p2[,k3[,k4,k5,k6]])

ir

The input transformation matrix is equal to Invert(newCameraMatrix*R), where newCameraMatrix represents the camera in the new coordinate system and R is the rotation matrix.. This processing will be done outside the synthesizable block

_mapx_mat

Output mat objects containing the mapx

_mapy_mat

Output mat objects containing the mapy

_cm_size

9 for 3x3 matrix

_dc_size

4, 5 or 8. If this is 0, then it means there is no distortion

InRange

The InRange function checks if pixels in the image src lie between the given boundaries. dst(x,y) is set to 255, if src(x,y) is within the specified thresholds and otherwise 0.

Dst(I)= lowerb ≤ src(I) ≤ upperb

Where (x,y) is the spatial coordinate of the pixel.

API Syntax

template<int SRC_T, int ROWS, int COLS,int NPC=1>
void inRange(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src,unsigned char lower_thresh,unsigned char upper_thresh,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . InRange Parameter Description

Parameter

Description

SRC_T

Input Pixel Type. 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src

Input image

dst

Output image

lower_thresh

Lower threshold value

upper_thresh

Upper threshold value

Resource Utilization

The following table summarizes the resource utilization of the InRange function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

86

154

LUT

60

148

CLB

15

37

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Integral Image

The integral function computes an integral image of the input. Each output pixel is the sum of all pixels above and to the left of itself.
image91

API Syntax

template<int SRC_TYPE,int DST_TYPE, int ROWS, int COLS, int NPC=1>
void integral(xf::cv::Mat<SRC_TYPE, ROWS, COLS, NPC> & _src_mat, xf::cv::Mat<DST_TYPE, ROWS, COLS, NPC> & _dst_mat)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . integral Parameter Description

Parameter

Description

SRC_TYPE

Input pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1)

DST_TYPE

Output pixel type. Only 32-bit,unsigned,1 channel is supported(XF_32UC1)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image

NPC

Number of pixels to be processed per cycle; this function supports only XF_NPPC1 or 1 pixel per cycle operations.

_src_mat

Input image

_dst_mat

Output image

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Name

Resource Utilization

1 pixel per clock operation

300 MHz

BRAM_18K

4

DSP48E

0

FF

613

LUT

378

CLB

102

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

7.2

Dense Pyramidal LK Optical Flow

Optical flow is the pattern of apparent motion of image objects between two consecutive frames, caused by the movement of object or camera. It is a 2D vector field, where each vector is a displacement vector showing the movement of points from first frame to second.

Optical Flow works on the following assumptions:

  • Pixel intensities of an object do not have too many variations in consecutive frames

  • Neighboring pixels have similar motion

Consider a pixel I(x, y, t) in first frame. (Note that a new dimension, time, is added here. When working with images only, there is no need of time). The pixel moves by distance (dx, dy) in the next frame taken after time dt. Thus, since those pixels are the same and the intensity does not change, the following is true:
image92

Taking the Taylor series approximation on the right-hand side, removing common terms, and dividing by dt gives the following equation:

image93

Where image94, image95, image96 and image97.

The above equation is called the Optical Flow equation, where, fx and fy are the image gradientsand ft is the gradient along time. However, (u, v) is unknown. It is not possible to solve this equation with two unknown variables. Thus, several methods are provided to solve this problem. One method is Lucas-Kanade. Previously it was assumed that all neighboring pixels have similar motion. The Lucas-Kanade method takes a patch around the point, whose size can be defined through the ‘WINDOW_SIZE’ template parameter. Thus, all the points in that patch have the same motion. It is possible to find (fx, fy, ft ) for these points. Thus, the problem now becomes solving ‘WINDOW_SIZE * WINDOW_SIZE’ equations with two unknown variables,which is over-determined. A better solution is obtained with the “least square fit” method. Below is the final solution, which is a problem with two equations and two unknowns:

image98

This solution fails when a large motion is involved and so pyramids are

used. Going up in the pyramid, small motions are removed and large motions become small motions and so by applying Lucas-Kanade, the optical flow along with the scale is obtained.

API Syntax

template< int NUM_PYR_LEVELS, int NUM_LINES, int WINSIZE, int FLOW_WIDTH, int FLOW_INT, int TYPE, int ROWS, int COLS, int NPC,bool USE_URAM=false>
void densePyrOpticalFlow(
xf::cv::Mat<TYPE,ROWS,COLS,NPC> & _current_img,
xf::cv::Mat<TYPE,ROWS,COLS,NPC> & _next_image,
xf::cv::Mat<XF_32UC1,ROWS,COLS,NPC> & _streamFlowin,
xf::cv::Mat<XF_32UC1,ROWS,COLS,NPC> & _streamFlowout,
const int level, const unsigned char scale_up_flag, float scale_in, ap_uint<1> init_flag)

Parameter Descriptions

The following table describes the template and the function parameters.

Descriptions

Parameter

Description

NUM_PYR_LEVE LS

Number of Image Pyramid levels used for the optical flow computation

NUM_LINES

Number of lines to buffer for the remap algorithm – used to find the temporal gradient

WINSIZE

Window Size over which Optical Flow is computed

FLOW_WIDTH, FLOW_INT

Data width and number of integer bits to define the signed flow vector data type. Integer bit includes the signed bit.

The default type is 16-bit signed word with 10 integer bits and 6 decimal bits.

TYPE

Pixel type of the input image. XF_8UC1 is only the supported value.

ROWS

Maximum height or number of rows to build the hardware for this kernel

COLS

Maximum width or number of columns to build the hardware for this kernel

NPC

Number of pixels the hardware kernel must process per clock cycle. Only XF_NPPC1, 1 pixel per cycle, is supported.

USE_URAM

Enable to map some storage structures to UltraRAM

_curr_img

First input image stream

_next_img

Second input image to which the optical flow is computed with respect to the first image

_streamFlow in

32-bit Packed U and V flow vectors input for optical flow. The bits from 31-16 represent the flow vector U while the bits from 15-0 represent the flow vector V.

_streamFlow out

32-bit Packed U and V flow vectors output after optical flow computation. The bits from 31-16 represent the flow vector U while the bits from 15-0 represent the flow vector V.

level

Image pyramid level at which the algorithm is currently computing the optical flow.

scale_up_fla g

Flag to enable the scaling-up of the flow vectors. This flag is set at the host when switching from one image pyramid level to the other.

scale_in

Floating point scale up factor for the scaling-up the flow vectors. The value is (previous_rows-1)/(current_rows-1). This is not 1 when switching from one image pyramid level to the other.

init_flag

Flag to initialize flow vectors to 0 in the first iteration of the highest pyramid level. This flag must be set in the first iteration of the highest pyramid level (smallest image in the pyramid). The flag must be unset for all the other iterations.

Resource Utilization

The following table summarizes the resource utilization of densePyrOpticalFlow for 1 pixel per cycle implementation, with the optical flow computed for a window size of 11 over an image size of 1920x1080 pixels. The results are after implementation in Vivado HLS 2019.1 for the Xilinx xczu9eg-ffvb1156-2L-e FPGA at 300 MHz.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

1 Pixel

300

32231

16596

52

215

Resource Utilization with UltraRAM Enable

The following table summarizes the resource utilization of densePyrOpticalFlow for 1 pixel per cycle implementation, with the optical flow computed for a window size of 11 over an image size of 3840X2160 pixels. The results are after implementation in Vivado HLS 2019.1 for the Xilinx xczu7ev-ffvc1156-2 FPGA at 300 MHz with UltraRAM enabled.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

URAM

1 Pixel

300

31164

42320

81

34

23

Performance Estimate

The following table summarizes performance figures on hardware for the densePyrOpticalFlow function for 5 iterations over 5 pyramid levels scaled down by a factor of two at each level. This has been tested on the zcu102 evaluation board.

Operating Mode

Operating Frequency

(MHz)

Image Size

Latency Estimate

Max (ms)

1 pixel

300

1920x1080

49.7

1 pixel

300

1280x720

22.9

1 pixel

300

1226x370

12.02

Dense Non-Pyramidal LK Optical Flow

Optical flow is the pattern of apparent motion of image objects between two consecutive frames, caused by the movement of object or camera. It is a 2D vector field, where each vector is a displacement vector showing the movement of points from first frame to second.

Optical Flow works on the following assumptions:

  • Pixel intensities of an object do not have too many variations in consecutive frames

  • Neighboring pixels have similar motion

Consider a pixel I(x, y, t) in first frame. (Note that a new dimension, time, is added here. When working with images only, there is no need of time). The pixel moves by distance (dx, dy) in the next frame taken after time dt. Thus, since those pixels are the same and the intensity does not change, the following is true:
image99

Taking the Taylor series approximation on the right-hand side, removing common terms, and dividing by dt gives the following equation:

image100

Where image101, image102, image103 and image104.

The above equation is called the Optical Flow equation, where, fx and fy are the image gradientsand ft is the gradient along time. However, (u, v) is unknown. It is not possible to solve this equation with two unknown variables. Thus, several methods are provided to solve this problem. One method is Lucas-Kanade. Previously it was assumed that all neighboring pixels have similar motion. The Lucas-Kanade method takes a patch around the point, whose size can be defined through the ‘WINDOW_SIZE’ template parameter. Thus, all the points in that patch have the same motion. It is possible to find (fx, fy, ft ) for these points. Thus, the problem now becomes solving ‘WINDOW_SIZE * WINDOW_SIZE’ equations with two unknown variables,which is over-determined. A better solution is obtained with the “least square fit” method. Below is the final solution, which is a problem with two equations and two unknowns:

image105

API Syntax

template<int TYPE, int ROWS, int COLS, int NPC, int WINDOW_SIZE,bool USE_URAM=false>
void DenseNonPyrLKOpticalFlow (xf::cv::Mat<TYPE, ROWS, COLS, NPC> & frame0, xf::cv::Mat<TYPE, ROWS, COLS, NPC> & frame1, xf::cv::Mat<XF_32FC1, ROWS, COLS, NPC> & flowx, xf::cv::Mat<XF_32FC1, ROWS, COLS, NPC> & flowy)

Parameter Descriptions

The following table describes the template and the function parameters.

Descriptions

Parameter

Description

Type

pixel type. The current supported pixel value is XF_8UC1, unsigned 8 bit.

ROWS

Maximum number of rows of the input image that the hardware kernel must be built for.

COLS

Maximum number of columns of the input image that the hardware kernel must be built for.

NPC

Number of pixels to process per cycle. Supported values are XF_NPPC1 (=1) and XF_NPPC2(=2).

WINDOW_SIZE

Window size over which optical flow will be computed. This can be any odd positive integer.

USE_URAM

Enable to map storage structures to UltraRAM.

frame0

First input images.

frame1

Second input image. Optical flow is computed between frame0 and frame1.

flowx

Horizontal component of the flow vectors. The format of the flow vectors is XF_32FC1 or single precision.

flowy

Vertical component of the flow vectors. The format of the flow vectors is XF_32FC1 or single precision.

Resource Utilization

The following table summarizes the resource utilization of DenseNonPyrLKOpticalFlow for a 4K image, as generated in the Vivado HLS 2019.1 version tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA at 300 MHz.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUTs

1 pixel

300

178

42

11984

7730

2 pixel

300

258

82

22747

15126

The following table summarizes the resource utilization of DenseNonPyrLKOpticalFlow for a 4K image, as generated in the Vivado HLS version tool for the Xilinx Xczu7eg-ffvb1156-1 FPGA at 300 MHz with UltraRAM enabled.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

URAM

DSP_48Es

FF

LUTs

1 pixel

300

0

12

42

11803

7469

2 pixel

300

0

23

80

22124

13800

Performance Estimate

The following table summarizes performance estimates of the DenseNonPyrLKOpticalFlow function for a 4K image, generated using Vivado HLS 2019.1 version tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

Max (ms)

1 pixel

300

28.01

2 pixel

300

14.01

Kalman Filter

The classic Kalman Filter is proposed for linear system. The state-space description of a linear system assumed to be:

where xk is the state vector at kth time instant, constant (known) Ak is an nxn state transition matrix, constant (known) Bk is an nxm control input matrix, constant (known) Γk is an nxp system noise input matrix, constant (known) Hk is a qxn measurement matrix, constant (known) with 1≤ m, p, q ≤ n, {u:sub:k} a (known) sequence of m vectors (called a deterministic input sequence), and image106 and image107 are respectively, (unknown) system and observation noise sequences, with known statistical information such as mean, variance, and covariance.

The Kalman filter assumes the following:

  1. image108 and image109 are assumed to be sequences of zero-mean Gaussian (or normal) white noise. That is, image110 and image111, where δkl is a Kronecker Delta function, and Qk and Rk are positive definite matrices, E(u) is an expectation of random variable u.

  2. image112

  3. The initial state x0 is also assumed to be independent of image113 and image114, that is image115.

The representation image116 means the estimate of x at time instant k using all the data measured till the time instant j.

The Kalman filter algorithm can be summarized as shown in the below equations:

Initialization

Time Update / Predict

Measurement Update/Correction

Where Pk,j is an estimate error covariance nxn matrix, Gk is Kalman gain nxq matrix, and k=1, 2,..

Computation Strategy

The numerical accuracy of the Kalman filter covariance measurement update is a concern for implementation, since it differentiates two positive definite arrays. This is a potential problem if finite precision is used for computation. This design uses UDU factorization of P to address the numerical accuracy/stability problems.

image117



image118



image119


Example for Kalman Filter

//Control Flag
    INIT_EN      = 1; TIMEUPDATE_EN = 2; MEASUPDATE_EN = 4;
    XOUT_EN_TU  = 8; UDOUT_EN_TU    = 16; XOUT_EN_MU    = 32;
    UDOUT_EN_MU = 64; EKF_MEM_OPT   = 128;
    //Load A_mat,B_mat,Uq_mat,Dq_mat,H_mat,X0_mat,U0_mat,D0_mat,R_mat
    //Initialization
KalmanFilter(A_mat, B_mat, Uq_mat, Dq_mat,  H_mat, X0_mat, U0_mat, D0_mat, R_mat, u_mat, y_mat, Xout_mat, Uout_mat, Dout_mat, INIT_EN);

for(int iteration=0; iteration< count; iteration++)
{
    //Load u_mat (control input)
    for(int index=0; index <C_CTRL; index ++)
        u_mat.write_float(index, control_input[index]);

//Time Update
KalmanFilter(A_mat, B_mat, Uq_mat, Dq_mat,  H_mat, X0_mat, U0_mat, D0_mat, R_mat, u_mat, y_mat, Xout_mat, Uout_mat, Dout_mat, TIMEUPDATE_EN + XOUT_EN_TU + UDOUT_EN_TU);

//Load y_mat (measurement vector)
    for(int index =0; index <M_MEAS; index ++)
        y_mat.write_float(index, control_input[index]);

//Measurement Update
KalmanFilter(A_mat, B_mat, Uq_mat, Dq_mat,  H_mat, X0_mat, U0_mat, D0_mat, R_mat, u_mat, y_mat, Xout_mat, Uout_mat, Dout_mat, MEASUPDATE_EN + XOUT_EN_MU + UDOUT_EN_MU);
}

API Syntax

template<int N_STATE, int M_MEAS, int C_CTRL, Int  MTU, int MMU, bool USE_URAM=0, bool EKF_EN=0, int TYPE, int NPC >
void KalmanFilter ( xf::cv::Mat<TYPE, N_STATE, N_STATE, NPC>  &A_mat,
#if KF_C!=0
xf::cv::Mat<TYPE, N_STATE, C_CTRL, NPC>   &B_mat,
#endif
xf::cv::Mat<TYPE, N_STATE, N_STATE, NPC>  &Uq_mat,
xf::cv::Mat<TYPE, N_STATE, 1, NPC>        &Dq_mat,
xf::cv::Mat<TYPE, M_MEAS, N_STATE, NPC>   &H_mat,
xf::cv::Mat<TYPE, N_STATE, 1, NPC>        &X0_mat,
xf::cv::Mat<TYPE, N_STATE, N_STATE, NPC>  &U0_mat,
xf::cv::Mat<TYPE, N_STATE, 1, NPC>        &D0_mat,
xf::cv::Mat<TYPE, M_MEAS, 1, NPC>         &R_mat,
#if KF_C!=0
xf::cv::Mat<TYPE, C_CTRL, 1, NPC>         &u_mat,
#endif
xf::cv::Mat<TYPE, M_MEAS, 1, NPC>         &y_mat,
xf::cv::Mat<TYPE, N_STATE, 1, NPC>        &Xout_mat,
xf::cv::Mat<TYPE, N_STATE, N_STATE, NPC>  &Uout_mat,
xf::cv::Mat<TYPE, N_STATE, 1, NPC>        &Dout_mat,
unsigned char flag)

Parameter Descriptions

Table . Kalman Filter Parameter Description

Parameter

Used (?) or Unused (X)

Description

Initialization

Time Update

Measurement Update

N_STATE

?

?

?

Number of state variable; possible options are 1 to 128

M_MEAS

?

?

?

Number of measurement variable; possible options are 1 to 128; M_MEAS must be less than or equal to N_STATE. In case of Extended Kalman Filter(EKF), M_MEAS should be 1.

C_CTRL

?

?

?

Number of control variable; possible options are 0 to 128; C_CTRL must be less than or equal to N_STATE. In case of EKF, C_CTRL should be 1.

MTU

?

?

?

Number of multipliers used in time update; possible options are 1 to 128; MTU must be less than or equal to N_STATE

MMU

?

?

?

Number of multipliers used in Measurement update; possible options are 1 to 128; MMU must be less than or equal to N_STATE

USE_URAM

?

?

?

URAM enable; possible options are 0 and 1

EKF_EN

?

?

?

Extended Kalman Filter Enable; possible options are 0 and 1

TYPE

?

?

?

Type of input pixel. Currently, only XF_32FC1 is supported.

NPC

?

?

?

Number of pixels to be processed per cycle; possible option is XF_NPPC1 (NOT relevant for this function)

A_mat

?

X

X

Transition matrix A. In case of EKF, Jacobian Matrix F is mapped to A_mat.

B_mat

?

X

X

Control matrix B. In case of KF, B_mat argument is not required when C_CTRL=0. And in case of EKF, Dummy matrix with size (N_STATE x 1) is mapped to B_mat.

Uq_mat

?

X

X

U matrix for Process noise covariance matrix Q

Dq_mat

?

X

X

D matrix for Process noise covariance matrix Q(only diagonal elements)

H_mat

?

X

X

Measurement Matrix H. In case of EKF, Jacobian Matrix H is mapped to H_mat.

X0_mat

?

X

X

Initial state matrix. In case of EKF, state transition function f is mapped to X0_mat.

U0_mat

?

X

X

U matrix for initial error estimate covariance matrix P

D0_mat

?

X

X

D matrix for initial error estimate covariance matrix P(only diagonal elements)

R_mat

?

X

X

Measurement noise covariance matrix R(only diagonal elements). In case of EKF, input only one value of R since M_MEAS=1.

u_mat

X

?

X

Control input vector. In case of KF, u_mat argument is not required when C_CTRL=0. And in case of EKF, observation function h is mapped to u_mat.

y_mat

X

X

?

Measurement vector. In case of EKF, input only one measurement since M_MEAS=1.

Xout_mat

X

?

?

Output state matrix

Uout_mat

X

?

?

U matrix for output error estimate covariance matrix P

Dout_mat

X

?

?

D matrix for output error estimate covariance matrix P(only diagonal elements)

flag

?

?

?

Control flag register

All U, D counterparts of all initialized matrices (Q and P) are obtained using U-D factorization
Table . Control Flag Registers

Flag bit

Description

0

Initialization enable

1

Time update enable

2

Measurement update enable

3

Xout enable for time update

4

Uout/D:sub:out enable for time update

5

Xout enables for measurement update

6

Uout/D:sub:out enable for measurement update

7

Read optimization (Uq_mat, Dq_mat, U0_mat, D0_mat and R_mat) for Extended Kalman Filter

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1 FPGA.

Name

Resource Utilization

N_STATE=128; C_CTRL=128; M_MEAS=128; MTU=24; MMU=24

N_STATE=64; C_CTRL=64; M_MEAS=12;MTU=16;MMU=16

N_STATE=5; C_CTRL=1; M_MEAS=3;MTU=2;MMU=2

300 MHz

300 MHz

300 MHz

BRAM_18K

387

142

24

DSP48E

896

548

87

FF

208084

128262

34887

LUT

113556

70942

18141

The following table shows the resource utilization of the kernel for a configuration with USE_URAM enable, generated using Vivado HLS 2019.1 for the Xilinx xczu7ev-ffvc1156-2-e FPGA.

Table . Resource Utilization with UltraRAM Enabled

Resource

Resource Utilization (N_STATE=64; C_CTRL=64; M_MEAS=12; MTU=4; MMU=4) (300 MHz) (ms)

BRAM_18K

30

DSP48E

284

FF

99210

LUT

53939

URAM

11

Performance Estimate

The following table shows the performance of kernel for different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx® Xczu9eg-ffvb1156-1, for one iteration. Latency estimate is calculated by taking average latency of 100 iteration.

Operating Mode

Latency Estimate

Operating Frequency (MHz)

Latency (ms)

N_STATE=128; C_CTRL=128; M_MEAS=128; MTU=24; MMU=24

300

0.7

N_STATE=64; C_CTRL=64; M_MEAS=12; MTU=16; MMU=16

300

0.12

N_STATE=5; C_CTRL=1; M_MEAS=3; MTU=2; MMU=2

300

0.04

The following table shows the performance of kernel for a configuration with UltraRAM enable, as generated using Vivado HLS 2019.1 tool for the Xilinx xczu7ev-ffvc1156-2-e, for one iteration. Latency estimate is calculated by taking average latency of 100 iteration.

Table . Performance Estimate with UltraRAM

Operating Mode

Operating Frequency (MHz)

Latency (ms)

N_STATE=64; C_CTRL=64; M_MEAS=12;MTU=4;MMU= 4

300

0.25

Extended Kalman Filter

The Kalman filter estimates the state vector in a linear model. If the model is nonlinear, then a linearization procedure is performed to obtain the filtering equations. The Kalman filter so obtained will be called the Extended Kalman filter. A state-space description of non-linear system can have a non-linear model of the form:



image120



image121


Where fk and hk are valued functions with ranges in Rn and Rq, respectively. 1≤q≤n, and Tk a matrix-valued function with range in RnxRq such that for each k the first order partial derivatives of fk (x:sub:k) and hk (x:sub:k) with respect to all the components of xk are continuous. We consider zero-mean Gaussian white noise sequences image122 and image123 with ranges in Rp and Rq respectively, 1≤p, q≤n.

The real-time linearization process is carried out as shown in the following equations. In the lines of the linear model, the initial estimate and predicted position are chosen to be:

image124


Then, image125, consecutively, for k=1,2,…, use the predicted positions.

image126


  1. image127, where image128, k is a time index and superscript is row index and image129

  2. image130 is a space of column vectors image131

The equation for time update computations is as follows:

image132


image133
The equation for measurement update computations is as follows:

image134



image135



image136


Example for Extended Kalman Filter

//Load F/B_mat/Uq_mat/Dq_mat/X0_mat/U0_mat/D0_mat

for(int iteration=0; iteration< count; iteration++)
{
    if(iteration ==0)
        model_fx(X0_mat, fx);// update fx using X0_mat
    else
model_fx(Xout_mat, fx); // update fx using Xout_mat

        unsigned char initFlag;
        if(iteration ==0)
            initFlag = INIT_EN;
        else
            initFlag = EKF_MEM_OPT+INIT_EN;

        //Initialization
KalmanFilter (F, B_mat, Uq_mat, Dq_mat, H, fx, U0_mat, D0_mat, R_mat, hx, y_mat, Xout_mat, Uout_mat, Dout_mat, initFlag);

        //Time Update
KalmanFilter (F, B_mat, Uq_mat, Dq_mat, H, fx, U0_mat, D0_mat, R_mat, hx, y_mat, Xout_mat, Uout_mat, Dout_mat, TIMEUPDATE_EN + XOUT_EN_TU + UDOUT_EN_TU);
        for(int index=0; index< M_MEAS; index++)
        {
if(iteration ==0)
// update hx/H using X0_mat for one measurement at a time
            model_hxH(X0_mat, hx, H, index);
        else
        //update hx/H using Xout_mat for one measurement at a time
model_hxH(Xout_mat, hx, H, index);

//Load R_mat
            R_mat.write_float(0,R_matrix[index][index]);

            //Load y_mat
            Y_mat.write_float(0,measurement_vector[index]);

//Measurement Update
KalmanFilter (F, B_mat, Uq_mat, Dq_mat, H, fx, U0_mat, D0_mat, R_mat, hx, y_mat, Xout_mat, Uout_mat, Dout_mat, MEASUPDATE_EN + XOUT_EN_MU + UDOUT_EN_MU);
        }
}

Laplacian Operator

This function calculates the Laplacian of the input image. This function internally uses the filter2D kernel to compute the Laplacian. The filter coefficients are calculated using cv::getDerivKernels OpenCV function on the host side.

API Syntax

template<int BORDER_TYPE,int FILTER_WIDTH,int FILTER_HEIGHT, int SRC_T,int DST_T, int ROWS, int COLS,int NPC=1>
void filter2D(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_mat,short int filter[FILTER_HEIGHT*FILTER_WIDTH],unsigned char _shift)

Parameter Descriptions

The following table describes the template and the function parameters.

Table filter2D Parameter Description

Parameter

Description

BORDER_TYPE

Border Type supported is XF_BORDER_CONSTANT

FILTER_HEIGHT

Number of rows in the input filter

FILTER_WIDTH

Number of columns in the input filter

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

DST_T

Output pixel type. 8-bit unsigned single and 3 channels (XF_8UC1, XF_8UC3) and 16-bit signed single and 3 channels (XF_16SC1, XF_16SC3) supported.

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be multiple of 8, for 8 pixel mode.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src_mat

Input image

_dst_mat

Output image

filter

The input filter of any size, provided the dimensions should be an odd number. The filter co-efficients either a 16-bit value or a 16-bit fixed point equivalent value.

_shift

The filter must be of type XF_16SP. If the co-efficients are floating point, it must be converted into the Qm.n and provided as the input as well as the shift parameter has to be set with the ‘n’ value. Else, if the input is not of floating point, the filter is provided directly and the shift parameter is set to zero.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 Pixel

3x3

300

3

9

1701

1161

269

5x5

300

5

25

3115

2144

524

8 Pixel

3x3

150

6

72

2783

2768

638

5x5

150

10

216

3020

4443

1007

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K 3 Channel image.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

3x3

300

18

27

886

801

8 Pixel

5x5

300

30

75

1793

1445

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Filter Size

Latency Estimate

Max (ms)

1 pixel

300

3x3

7

300

5x5

7.1

8 pixel

150

3x3

1.86

150

5x5

1.86

Lens shading correction

Vignetting/Lensshading refers to the fall-off pixel intensity from the centre towards the edges of the image. In this algorithm, vignette is corrected by considering the distance between the centre pixel and actual image pixel position. This distance is used to calculate intensity gain per pixel per channel which is used for the correction.

API Syntax

template <int SRC_T, int DST_T, int ROWS, int COLS, int NPC = 1>
void Lscdistancebased(xf::cv::Mat<SRC_T, ROWS, COLS, NPC>& src, xf::cv::Mat<DST_T, ROWS, COLS, NPC>& dst) {

Parameter Descriptions

The following table describes template parameters and arguments to the function.

Table Lensshading correction Parameter Description

Parameter

Description

SRC_T

Input pixel type. 8/10/12/16-bit, unsigned, 3 channel is supported (XF_8UC3, XF_10UC3, XF_12UC3, XF_16UC3).

DST_T

Output pixel type. 8/10/12/16-bit, unsigned, 3 channel is supported (XF_8UC3, XF_10UC3, XF_12UC3, XF_16UC3).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1, XF_NPPC2 AND so on

src

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vitis HLS 2020.2 tool, to process a FULL HD image.

Table Lensshading correction Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel-8U

300

0

26

4198

3628

889

1 pixel-16U

300

0

26

4253

3602

770

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vitis HLS 2020.2 tool, to process a FULL HD image.

Table Lensshading correction Function Performance Estimate Summary

Operating Mode

Operating Frequency (MHz)

Latency Estimate Max (ms)

1 pixel

300

7

2 pixel

300

3.6

Local Tone Mapping

Most of the display devices have limited dynamic range. Hence images with wide dynamic range cannot be seen natively on such devices. To see wide dynamic range images on devices with low dynamic range, we need to compress the wide dynamic range of image to a low dynamic range. This process is called as tone-mapping.

Local tone mapping takes pixel neighbor statistics into account, and produces images with more contrast and brightness.

This implementaion is based on the algorithm proposed by J. Yang, A. Hore and O. Yadid-Pecht.

API Syntax

LTM Class API:

template <int IN_TYPE, int OUT_TYPE, int BLOCK_HEIGHT, int BLOCK_WIDTH, int ROWS, int COLS, int NPC>
class LTM {}

Processing member function:

xf::cv::LTM<IN_TYPE, OUT_TYPE, BLOCK_HEIGHT, BLOCK_WIDTH, ROWS, COLS, NPC>::process(xf::cv::Mat<IN_TYPE, ROWS, COLS, NPC>& in,
                     int block_rows,
                     int block_cols,
                     XF_CTUNAME(IN_TYPE, NPC) omin_r[MinMaxVArrSize][MinMaxHArrSize],
                     XF_CTUNAME(IN_TYPE, NPC) omax_r[MinMaxVArrSize][MinMaxHArrSize],
                     XF_CTUNAME(IN_TYPE, NPC) omin_w[MinMaxVArrSize][MinMaxHArrSize],
                     XF_CTUNAME(IN_TYPE, NPC) omax_w[MinMaxVArrSize][MinMaxHArrSize],
                     xf::cv::Mat<OUT_TYPE, ROWS, COLS, NPC>& out)

Overlaoded processing member function:

xf::cv::LTM<IN_TYPE, OUT_TYPE, BLOCK_HEIGHT, BLOCK_WIDTH, ROWS, COLS, NPC>::process(xf::cv::Mat<IN_TYPE, ROWS, COLS, NPC>& in,
                     XF_CTUNAME(IN_TYPE, NPC) omin_r[MinMaxVArrSize][MinMaxHArrSize],
                     XF_CTUNAME(IN_TYPE, NPC) omax_r[MinMaxVArrSize][MinMaxHArrSize],
                     XF_CTUNAME(IN_TYPE, NPC) omin_w[MinMaxVArrSize][MinMaxHArrSize],
                     XF_CTUNAME(IN_TYPE, NPC) omax_w[MinMaxVArrSize][MinMaxHArrSize],
                     xf::cv::Mat<OUT_TYPE, ROWS, COLS, NPC>& out)

Parameter Descriptions

The following table describes the template and the function parameters.

Descriptions

Parameter

Description

IN_TYPE

Input pixel type. The current supported pixel value is XF_8UC3, XF_10UC3, XF_12UC3, XF_16UC3, XF_32FC3

OUT_TYPE

Input pixel type. The current supported pixel value is XF_8UC3, XF_10UC3, XF_12UC3, XF_16UC3

BLOCK_WIDTH

Max block width the image is divided into. This can be any positive integer greater than or equal to 32 and less than input image width.

BLOCK_HEIGHT

Max block height the image is divided into. This can be any positive integer greater than or equal to 32 and less than input image height.

ROWS

Maximum number of rows of the input image that the hardware kernel must be built for.

COLS

Maximum number of columns of the input image that the hardware kernel must be built for.

NPC

Number of pixels to process per cycle. Supported values are XF_NPPC1, XF_NPPC2, XF_NPPC4, XF_NPPC8.

in

Input HDR image

block_rows

Actual block height

block_cols

Actual block width

omin_r

Array of min values to be read by the next frame.

omax_r

Array of max values to be read by the next frame.

omin_w

Array of min values computed in the current frame.

omax_w

Array of max values computed in the current frame.

out

Output HDR image

Resource Utilization

The following table summarizes the resource utilization of LocalToneMapping for a 4K image, as generated in the Vitis HLS 2020.2 version tool for the Xilinx xcu200-fsgd2104-2-e FPGA at 300MHz.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUTs

1 pixel

300

0

123

35216

20246

4 pixel

300

0

330

67457

40391

Performance Estimate

The following table summarizes performance estimates of the LocalToneMapping function for a 4K image, generated using Vitis HLS 2020.2 version tool for the Xilinx xcu200-fsgd2104-2-e FPGA.

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

Max (ms)

1 pixel

300

7.2

4 pixel

300

1.9

Look Up Table

The LUT function performs the table lookup operation. Transforms the source image into the destination image using the given look-up table. The input image must be of depth XF_8UP and the output image of same type as input image.

Iout(x, y) = LUT [I:sub:in1(x, y)]

Where:

  • Iout(x, y) is the intensity of output image at (x, y) position

  • Iin(x, y) is the intensity of first input image at (x, y) position

  • LUT is the lookup table of size 256 and type unsigned char.

API Syntax

template <int SRC_T, int ROWS, int COLS,int NPC=1>
void LUT(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst,unsigned char* _lut)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . LUT Parameter Description

Parameter

Description

SRC_T

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Number of rows in the image being processed.

COLS

Number of columns in the image being processed. Must be a multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed in parallel. Possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input image of size (ROWS, COLS) and type 8U.

_dst

Output image of size (ROWS, COLS) and same type as input.

_lut

Input lookup Table of size 256 and type unsigned char.

Resource Utilization

The following table summarizes the resource utilization of the LUT function, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

1

0

937

565

137

8 pixel

150

9

0

1109

679

162

The following table summarizes the resource utilization of the LUT function, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process 4K 3Channel image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

4

0

1160

648

175

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.92

8 pixel operation (150 MHz)

1.66

Mean and Standard Deviation

The meanStdDev function computes the mean and standard deviation of input image. The output Mean value is in fixed point Q8.8 format, and the Standard Deviation value is in Q8.8 format. Mean and standard deviation are calculated as follows:


image137
image138

API Syntax

template<int SRC_T,int ROWS, int COLS,int NPC=1>
void meanStdDev(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,unsigned short* _mean,unsigned short* _stddev)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . meanStdDev Parameter Description

Parameter

Description

SRC_T

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Number of rows in the image being processed.

COLS

Number of columns in the image being processed. Must be a multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input image

_mean

16-bit data pointer through which the computed mean of the image is returned.

_stddev

16-bit data pointer through which the computed standard deviation of the image is returned.

Resource Utilization

The following table summarizes the resource utilization of the meanStdDev function, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

0

6

896

461

121

8 pixel

150

0

13

1180

985

208

The following table summarizes the resource utilization of the meanStdDev function, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K 3Channel image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

0

7

5075

3324

725

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency

1 pixel operation (300 MHz)

6.9 ms

8 pixel operation (150 MHz)

1.69 ms

Max

The Max function calculates the per-element maximum of two corresponding images src1, src2 and stores the result in dst.

dst(x,y)=max( src1(x,y) ,src2(x,y) )

API Syntax

template< int SRC_T , int ROWS, int COLS, int NPC=1>
void max(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src2, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Max Parameter Description

Parameter

Description

SRC_T

Input Pixel Type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

First input image

_src2

Second input image

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the Max function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

103

153

LUT

44

102

CLB

21

38

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

MaxS

The MaxS function calculates the maximum elements between src and given scalar value scl and stores the result in dst.

dst(I)=maxS( src(I) ,scl )

API Syntax

template< int SRC_T , int ROWS, int COLS, int NPC=1>
void maxS(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, unsigned char _scl[XF_CHANNELS(SRC_T,NPC)], xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . MaxS Parameter Description

Parameter

Description

SRC_T

Input Pixel Type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

First Input image

_scl

Input scalar value, the size should be number of channels

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the MaxS function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

162

43

LUT

103

104

CLB

32

20

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Median Blur Filter

The function medianBlur performs a median filter operation on the input image. The median filter acts as a non-linear digital filter which improves noise reduction. A filter size of N would output the median value of the NxN neighborhood pixel values, for each pixel.

API Syntax

template<int FILTER_SIZE, int BORDER_TYPE, int TYPE, int ROWS, int COLS, int NPC>
void medianBlur (xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _src, xf::cv::Mat<TYPE, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . medianBlur Parameter Description

Parameter

Description

FILTER_SIZE

Window size of the hardware filter for which the hardware kernel will be built. This can be any odd positive integer greater than 1.

BORDER_TYPE

The way in which borders will be processed in the hardware kernel. Currently, only XF_BORDER_REPLICATE is supported.

TYPE

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Number of rows in the image being processed.

COLS

Number of columns in the image being processed. Must be a multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed in parallel. Options are XF_NPPC1 (for 1 pixel processing per clock), XF_NPPC8 (for 8 pixel processing per clock

_src

Input image.

_dst

Output image.

Resource Utilization

The following table summarizes the resource utilization of the medianBlur function for XF_NPPC1 and XF_NPPC8 configurations, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

FILTER_SIZE

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

1 pixel

3

300

1197

771

0

3

8 pixel

3

150

6559

1595

0

6

1 pixel

5

300

5860

1886

0

5

The following table summarizes the resource utilization of the medianBlur function for XF_NPPC1 with 3channel image as input, generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

FILTER_SIZE

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

1 pixel

3

300

2100

1971

0

9

1 pixel

5

300

13541

9720

0

15

Performance Estimate

The following table summarizes performance estimates of medianBlur function on Vivado HLS 2019.1 version tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Mode

FILTER_SIZE

Operating Frequency

(MHz)

Input Image Size

Latency Estimate

Max (ms)

1 pixel

3

300

1920x1080

6.99

8 pixel

3

150

1920x1080

1.75

1 pixel

5

300

1920x1080

7.00

Min

The Min function calculates the per element minimum of two corresponding images src1, src2 and stores the result in dst.

dst(I)=min( src1(I) ,src2(I) )

API Syntax

template< int SRC_T , int ROWS, int COLS, int NPC=1>
void min(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src2, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Min Parameter Description

Parameter

Description

SRC_T

Input pixel type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed per cycle, possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

First input image

_src2

Second input image

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the Min function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

103

153

LUT

44

102

CLB

23

34

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

MinS

The MinS function calculates the minimum elements between src and given scalar value scl and stores the result in dst.

dst(x,y)=minS( src(x,y) ,scl )

API Syntax

template< int SRC_T , int ROWS, int COLS, int NPC=1>
void minS(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, unsigned char _scl[XF_CHANNELS(SRC_T,NPC)], xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . MinS Parameter Description

Parameter

Description

SRC_T

Input Pixel Type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

First Input image

_scl

Input scalar value, the size should be the number of channels.

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the MinS function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

104

159

LUT

43

103

CLB

23

36

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

MinMax Location

The minMaxLoc function finds the minimum and maximum values in an image and location of those values.


image139
image140

API Syntax

template<int SRC_T,int ROWS,int COLS,int NPC>
void minMaxLoc(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src,int32_t *max_value, int32_t *min_value,uint16_t *_minlocx, uint16_t *_minlocy, uint16_t *_maxlocx, uint16_t *_maxlocy )

Parameter Descriptions

The following table describes the template and the function parameters.

Table . minMaxLoc Parameter Description

Parameter

Description

SRC_T

Input pixel type. 8-bit, unsigned, 1 channel (XF_8UC1), 16-bit, unsigned, 1 channel (XF_16UC1), 16-bit, signed, 1 channel (XF_16SC1), 32-bit, signed, 1 channel (XF_32SC1) are supported.

ROWS

Number of rows in the image being processed.

COLS

Number of columns in the image being processed.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src

Input image

max_val

Maximum value in the image, of type int.

min_val

Minimum value in the image, of type int.

_minlocx

x-coordinate location of the first minimum value.

_minlocy

y-coordinate location of the first minimum value.

_maxlocx

x-coordinate location of the first maximum value.

_maxlocy

y-coordinate location of the first maximum value.

Resource Utilization

The following table summarizes the resource utilization of the minMaxLoc function, generated using Vivado HLS 2019.1 tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

0

3

451

398

86

8 pixel

150

0

3

1049

1025

220

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.69

Mean Shift Tracking

Mean shift tracking is one of the basic object tracking algorithms. Mean-shift tracking tries to find the area of a video frame that is locally most similar to a previously initialized model. The object to be tracked is represented by a histogram. In object tracking algorithms target representation is mainly rectangular or elliptical region. It contains target model and target candidate. Color histogram is used to characterize the object. Target model is generally represented by its probability density function (pdf). Weighted RGB histogram is used to give more importance to object pixels.

Mean-shift algorithm is an iterative technique for locating the maxima of a density function. For object tracking, the density function used is the weight image formed using color histograms of the object to be tracked and the frame to be tested. By using the weighted histogram we are taking spatial position into consideration unlike the normal histogram calculation. This function will take input image pointer, top left and bottom right coordinates of the rectangular object, frame number and tracking status as inputs and returns the centroid using recursive mean shift approach.

API Syntax

template <int MAXOBJ, int MAXITERS, int OBJ_ROWS, int OBJ_COLS, int SRC_T, int ROWS, int COLS, int NPC>
void MeanShift(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_in_mat, uint16_t* x1, uint16_t* y1, uint16_t* obj_height, uint16_t* obj_width, uint16_t* dx, uint16_t* dy, uint16_t* status, uint8_t frame_status, uint8_t no_objects, uint8_t no_iters );

Template Parameter Descriptions

The following table describes the template parameters.

Function Parameter Description

The following table describes the function parameters.

Table . MeanShift Function Parameters

Parameter

Description

_in_mat

Input xF Mat

x1

Top Left corner x-coordinate of all the objects

y1

Top Left corner y-coordinate of all the objects

obj_height

Height of all the objects

obj_width

Width of all the objects

dx

Centers x-coordinate of all the objects returned by the kernel function

dy

Centers y-coordinate of all the objects returned by the kernel function

status

Track the object only if the status of the object is true, that is if the object goes out of the frame, status is made zero

frame_status

Set as zero for the first frame and one for other frames

no_objects

Number of objects racked

no_iters

Number of iterations for convergence

Resource Utilization and Performance Estimate

The following table summarizes the resource utilization of the MeanShift function for normal (1 pixel) configuration as generated in Vivado HLS 2019.1 release tool for the part xczu9eg-ffvb1156-i-es1 at 300 MHz to process a RGB image of resolution,1920x1080, and for 10 objects of size of 250x250 and 4 iterations.

Performance Estimate Summary

Configuration

Max. Latency (ms)

BRAMs

DSPs

FFs

LUTs

1 pixel

19.28

76

14

13198

10064

Limitations

The maximum number of objects that can be tracked is 10.

Mode filter

Mode filter is a non-linear digital filter which improves noise reduction. This implements filter operation with given size of N by computing mode for all the pixels in an NxN window.

API Syntax

template <int FILTER_SIZE, int BORDER_TYPE, int TYPE, int ROWS, int COLS, int NPC = 1>
void modefilter(xf::cv::Mat<TYPE, ROWS, COLS, NPC>& _src, xf::cv::Mat<TYPE, ROWS, COLS, NPC>& _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Mode filter Parameter Description

Parameter

Description

FILTER_SIZE

Window size of the hardware filter for which the hardware kernel will be built. This can be any odd positive integer greater than 1.

BORDER_TYPE

The way in which borders will be processed in the hardware kernel. Currently, only XF_BORDER_REPLICATE is supported.

TYPE

Input and Output pixel type. Only 8-bit, unsigned, 1 or 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Number of rows in the image being processed.

COLS

Number of columns in the image being processed. Must be a multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed in parallel. Options are XF_NPPC1 (for 1 pixel processing per clock), XF_NPPC8 (for 8 pixel processing per clock

_src

Input image.

_dst

Output image.

Resource Utilization

The following table summarizes the resource utilization of the Mode filter function for XF_NPPC1 configurations, generated using Vitis HLS 2020.2 version tool.

Table Mode filter Function Resource Utilization Summary

Operating Mode

FILTER_SIZE

Operating Frequency (MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

1 pixel

3

300

628

900

0

3

1 pixel

5

300

2579

4070

0

5

1 pixel

7

300

7852

14065

0

7

Performance Estimate

The following table summarizes performance estimates of Mode filter function on Vitis HLS 2020.2 version tool.

Table Mode filter Function Performance Estimate Summary

Operating Mode

FILTER_SIZE

Operating Frequency (MHz)

Input Image Size

Latency Estimate Max (ms)

1 pixel

3

300

1920x1080

6.99

1 pixel

5

300

1920x1080

7.00

1 pixel

7

300

1920x1080

7.15

Otsu Threshold

Otsu threshold is used to automatically perform clustering-based image thresholding or the reduction of a gray-level image to a binary image. The algorithm assumes that the image contains two classes of pixels following bi-modal histogram (foreground pixels and background pixels), it then calculates the optimum threshold separating the two classes.

Otsu method is used to find the threshold which can minimize the intra class variance which separates two classes defined by weighted sum of variances of two classes.

image141

Where, w_1is the class probability computed from the histogram.

image142

Otsu shows that minimizing the intra-class variance is the same as maximizing inter-class variance

image143 image144

Where,image145 is the class mean.

API Syntax

template<int SRC_T, int ROWS, int COLS,int NPC=1> void OtsuThreshold(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat, uint8_t &_thresh)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . OtsuThreshold Parameter Description

Paramete r

Description

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src_ma t

Input image

_thresh

Output threshold value after the computation

Resource Utilization

The following table summarizes the resource utilization of the OtsuThreshold function, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

8

49

2239

3353

653

8 pixel

150

22

49

1106

3615

704

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.92

8 pixel operation (150 MHz)

1.76

Paintmask

The Paintmask function replace the pixel intensity value with given color value when mask is not zero or the corresponding pixel from the input image.

API Syntax

template< int SRC_T,int MASK_T, int ROWS, int COLS,int NPC=1>
void paintmask(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat, xf::cv::Mat<MASK_T, ROWS, COLS, NPC> & in_mask, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst_mat, unsigned char _color[XF_CHANNELS(SRC_T,NPC)])

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Paintmask Parameter Description

Parameter

Description

SRC_T

Input pixel type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

MASK_T

Mask value type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src_mat

Input image

_in_mask

Input mask image

_dst_mat

Output image

_color

Color value to be filled when mask is not zero

Resource Utilization

The following table summarizes the resource utilization of the Paintmask Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

95

163

LUT

57

121

CLB

14

33

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Operating Frequency (MHz)

Latency (ms)

1 pixel

300

6.9

8 pixel

150

1.7

Pixel-Wise Addition

The add function performs the pixel-wise addition between two input images and returns the output image.

Iout(x, y) = Iin1(x, y) + Iin2(x, y)

Where:

  • Iout(x, y) is the intensity of the output image at (x, y) position

  • Iin1(x, y) is the intensity of the first input image at (x, y) position

  • Iin2(x, y) is the intensity of the second input image at (x, y) position.

XF_CONVERT_POLICY_TRUNCATE: Results are the least significant bits of the output operand, as if stored in two’s complement binary format in the size of its bit-depth.

XF_CONVERT_POLICY_SATURATE: Results are saturated to the bit depth of the output operand.

API Syntax

template<int POLICY_TYPE, int SRC_T, int ROWS, int COLS, int NPC=1>
void add (
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> dst )

Parameter Descriptions

The following table describes the template and the function parameters.

Table . add Parameter Description

Parameter

Description

POLICY_TY PE

Type of overflow handling. It can be either, XF_CONVERT_POLICY_SATURATE or XF_CONVERT_POLICY_TRUNCATE.

SRC_T

Pixel type. Options are XF_8UC1, XF_8UC3, XF_16SC3, and XF_16SC1.

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

src2

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

0

0

62

55

11

8 pixel

150

0

0

65

138

24

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process 4K image with 3 channels.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

0

0

113

77

24

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Pixel-Wise Multiplication

The multiply function performs the pixel-wise multiplication between two input images and returns the output image.

Iout(x, y) = Iin1(x, y) * Iin2(x, y) * scale_val

Where:

  • Iout(x, y) is the intensity of the output image at (x, y) position

  • Iin1(x, y) is the intensity of the first input image at (x, y) position

  • Iin2(x, y) is the intensity of the second input image at (x, y) position

  • scale_val is the scale value.

XF_CONVERT_POLICY_TRUNCATE: Results are the least significant bits of the output operand, as if stored in two’s complement binary format in the size of its bit-depth.

XF_CONVERT_POLICY_SATURATE: Results are saturated to the bit depth of the output operand.

API Syntax

template<int POLICY_TYPE, int SRC_T,int ROWS, int COLS, int NPC=1>
void multiply (
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int SRC_T int ROWS, int COLS, int NPC> dst,
float scale)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . multiply Parameter Description

Parameter

Description

POLICY_TY PE

Type of overflow handling. It can be either, XF_CONVERT_POLICY_SATURATE or XF_CONVERT_POLICY_TRUNCATE.

SRC_T

pixel type. Options are XF_8UC1,XF_8UC3,XF_16SC1 and XF_16SC3.

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

src2

Input image

dst

Output image

scale_val

Weighing factor within the range of 0 and 1

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

0

2

124

59

18

8 pixel

150

0

16

285

108

43

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K image with 3 channels.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

0

9

312

211

62

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Pixel-Wise Subtraction

The subtract function performs the pixel-wise subtraction between two input images and returns the output image.

Iout(x, y) = Iin1(x, y) - Iin2(x, y)

Where:

  • Iout(x, y) is the intensity of the output image at (x, y) position

  • Iin1(x, y) is the intensity of the first input image at (x, y) position

  • Iin2(x, y) is the intensity of the second input image at (x, y) position.

XF_CONVERT_POLICY_TRUNCATE: Results are the least significant bits of the output operand, as if stored in two’s complement binary format in the size of its bit-depth.

XF_CONVERT_POLICY_SATURATE: Results are saturated to the bit depth of the output operand.

API Syntax

template<int POLICY_TYPE int SRC_T, int ROWS, int COLS, int NPC=1>
void subtract (
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src1,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> src2,
xf::cv::Mat<int SRC_T, int ROWS, int COLS, int NPC> dst )

Parameter Descriptions

The following table describes the template and the function parameters.

Table . subtract Parameter Description

Parameter

Description

POLICY_TYPE

Type of overflow handling. It can be either, XF_CONVERT_POLICY_SATURATE or XF_CONVERT_POLICY_TRUNCATE.

SRC_T

pixel type. Options are XF_8UC1,XF_8UC3,XF_16SC3 and_16SC1.

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be a multiple of 8, for 8-pixel operation)

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

src1

Input image

src2

Input image

dst

Output image

Resource Utilization

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

0

0

62

53

11

8 pixel

150

0

0

59

13

21

The following table summarizes the resource utilization in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K image with 3 channels.

Operating Mode

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

0

0

110

64

28

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Quantization & Dithering

This algorithm dithers input image using Floyd-Steinberg dithering method. It is commonly used by image manipulation software, for example when an image is converted into GIF format each pixel intensity value is quantized to 8 bit i.e. 256 colors.

API Syntax

template <int IN_TYPE, int OUT_TYPE, int ROWS, int COLS, int SCALE_FACTOR, int MAX_REPRESENTED_VALUE, int NPC>
void xf_QuatizationDithering(xf::cv::Mat<IN_TYPE, ROWS, COLS, NPC>& stream_in,
                             xf::cv::Mat<OUT_TYPE, ROWS, COLS, NPC>& stream_out)

Parameter Descriptions

The following table describes the template and the function parameters.

Table Quantization & Dithering Parameter Description

Parameter

Description

IN_TYPE

Input pixel type. It should be XF_8UC1, XF_8UC3, XF_10UC1, XF_10UC3, XF_12UC1, XF_12UC3, XF_16UC1, or XF_16UC3. Note XF_<PIXEL_BITWIDTH>UC<NUM_CHANNELS>

OUT_TYPE

Output pixel type. It should be XF_8UC1, XF_8UC3, XF_10UC1, XF_10UC3, XF_12UC1, XF_12UC3, XF_16UC1, or XF_16UC3. Output PIXEL_WIDTH should less than or equal to input PIXEL_BITWIDTH

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

SCALE_FACTOR

The SCALE_FACTOR must be power of 2 & less than or equal to 2^(output PIXEL_BITWIDTH)

MAX_REPRESENTED_VALUE

The MAX_REPRESENTED_VALUE must be equal to equal to 2^(input PIXEL_BITWIDTH)

NPC

Number of pixels to be processed per cycle; possible options is XF_NPPC1 or XF_NPPC2

stream_in

Input image

stream_out

Output image

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vitis HLS 2020.2 tool for the Xilinx xcu200-fsgd2104-2-e FPGA, to process a RGB image with a resolution of 1024x676 & pixel width 16 bit and quantized it to 8 bit pixel width.

Table Quantization & Dithering Resource Utilization Summary

Operating Mode

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

300

11

10

7592

5765

1582

2 pixel

300

14

12

8150

6945

1749

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, as generated using Vitis HLS 2020.2 tool for the Xilinx xcu200-fsgd2104-2-e FPGA, to process a RGB image with a resolution of 1024x676 & pixel width 16 bit and quantized it to 8 bit pixel width.

Table Quantization & Dithering Function Performance Estimate Summary

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

Max (ms)

1 pixel

300

2.8

2 pixel

300

1.58

Reduce

The Reduce function reduces the matrix to a vector by treating rows/cols as set of 1-D vectors and performing specified operation on vectors until a single row/col is obtained.

Reduction operation could be one of the following:

  • REDUCE_SUM : The output is the sum of all of the matrix’s rows/columns.

  • REDUCE_AVG : The output is the mean vector of all of the matrix’s rows/columns.

  • REDUCE_MAX : The output is the maximum (column/row-wise) of all of the matrix’s rows/columns.

  • REDUCE_MIN : The output is the minimum (column/row-wise) of all of the matrix’s rows/columns.

API Syntax

template< int REDUCE_OP, int SRC_T , int DST_T,  int ROWS, int COLS, int ONE_D_HEIGHT, int ONE_D_WIDTH,int NPC=1> void reduce(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat, xf::cv::Mat<DST_T, ONE_D_HEIGHT, ONE_D_WIDTH, 1> & _dst_mat, unsigned char dim)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Reduce Parameter Description

Parameter

Description

REDUCE_OP

The flag specifies the type of reduction operation to be applied.

SRC_T

Input pixel type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

DST_T

Output pixel type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

ONE_D_HEIGHT

Height of output 1-D vector or reduced matrix

ONE_D_WIDTH

Width of output 1-D vector or reduced matrix

NPC

Number of pixels to be processed per cycle; possible option is XF_NPPC1 (1 pixel per cycle).

_src_mat

Input image

_dst_mat

1-D vector

dim

Dimension index along which the matrix is reduced. 0 means that the matrix is reduced to a single row. 1 means that the matrix is reduced to a single column.

Resource Utilization

The following table summarizes the resource utilization of the Reduce function Normal mode(1 pixel) as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

300 MHz

BRAM_18K

2

DSP48E

0

FF

288

LUT

172

CLB

54

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

Remap

The remap function takes pixels from one place in the image and relocates them to another position in another image. Two types of interpolation methods are used here for mapping the image from source to destination image.
image146

API Syntax

template<int WIN_ROWS,int INTERPOLATION_TYPE, int SRC_T, int MAP_T, int DST_T, int ROWS, int COLS, int NPC = 1,bool USE_URAM=false>

void remap (xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_src_mat,
         xf::cv::Mat<DST_T, ROWS, COLS, NPC> &_remapped_mat,
         xf::cv::Mat<MAP_T, ROWS, COLS, NPC> &_mapx_mat,
         xf::cv::Mat<MAP_T, ROWS, COLS, NPC> &_mapy_mat);

Parameter Descriptions

The following table describes the template parameters.

Table . remap template .. rubric:: Parameter Descriptions

Parameter

Description

WIN_ROWS

Number of input image rows to be buffered inside. Must be set based on the map data. For instance, for left right flip, 2 rows are sufficient.

INTERPOLATIO N_TYPE

Type of interpolation, either XF_INTERPOLATION_NN (nearest neighbor) or XF_INTERPOLATION_BILINEAR (linear interpolation)

SRC_T

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

MAP_T

Map type. Single channel float type. XF_32FC1.

DST_T

Output image type. Grayscale image of type 8-bits and single channel. XF_8UC1.

ROWS

Height of input and output images

COLS

Width of input and output images

NPC

Number of pixels to be processed per cycle; this function supports only XF_NPPC1 or 1 pixel per cycle operations.

USE_URAM

Enable to map some structures to UltraRAM instead of BRAM.

The following table describes the function parameters.

Table . remap Parameter Description

PARAMETERS

DESCRIPTION

_src_mat

Input xF Mat

_remapped_m at

Output xF Mat

_mapx_mat

mapX Mat of float type

_mapy_mat

mapY Mat of float type

Resource Utilization

The following table summarizes the resource utilization of remap, for HD (1080x1920) images generated in the Vivado HLS 2019.1 version tool for the Xilinx xczu9eg-ffvb1156-i-es1 FPGA at 300 MHz, with WIN_ROWS as 64 for the XF_INTERPOLATION_BILINEAR mode.

Table . remap Function Resource Utilization Summary

Name

Resource Utilization

BRAM_18K

64

DSP48E

17

FF

1738

LUT

1593

CLB

360

The following table summarizes the resource utilization of remap, for 4K (3840x2160) images generated in the Vivado HLS 2019.1 version tool for the Xilinx xczu7ev-ffvc1156 FPGA at 300 MHz, with WIN_ROWS as 100 for the XF_INTERPOLATION_BILINEAR mode using UltraRAM .

UltraRAM Enabled

Name

Resource Utilization

BRAM_18K

3

DSP48E

10

URAM

24

FF

3196

LUT

3705

Performance Estimate

The following table summarizes the performance of remap(), for HD (1080x1920) images generated in the Vivado HLS 2019.1 version tool for the Xilinx xczu9eg-ffvb1156-i-es1 FPGA at 300 MHz, with WIN_ROWS as 64 for XF_INTERPOLATION_BILINEAR mode.

Table . remap Function Performance Estimate Summary

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

Max latency (ms)

1 pixel mode

300

7.2

Resolution Conversion (Resize)

Resolution Conversion is the method used to resize the source image to the size of the destination image. Different types of interpolation techniques can be used in resize function, namely: Nearest-neighbor, Bilinear, and Area interpolation. The type of interpolation can be passed as a template parameter to the API. The following enumeration types can be used to specify the interpolation type:

  • XF_INTERPOLATION_NN - For Nearest-neighbor interpolation

  • XF_INTERPOLATION_BILINEAR - For Bilinear interpolation

  • XF_INTERPOLATION_AREA - For Area interpolation

Note: Scaling factors greater than or equal to 0.25 are supported in down-scaling and values less than or equal to 8 are supported for up-scaling.

API Syntax

template<int INTERPOLATION_TYPE, int TYPE, int SRC_ROWS, int SRC_COLS, int DST_ROWS, int DST_COLS, int NPC,int MAX_DOWN_SCALE>
void resize (xf::cv::Mat<TYPE, SRC_ROWS, SRC_COLS, NPC> & _src, xf::cv::Mat<TYPE, DST_ROWS, DST_COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . resize Parameter Description

Parameter

Description

INTERPOLATIO N_TYPE

Interpolation type. The different options possible are

  • XF_INTERPOLATION_NN – Nearest Neighbor Interpolation

  • XF_INTERPOLATION_BILINEAR – Bilinear interpolation

  • XF_INTERPOLATION_AREA – Area Interpolation

TYPE

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

SRC_ROWS

Maximum Height of input image for which the hardware kernel would be built.

SRC_COLS

Maximum Width of input image for which the hardware kernel would be built (must be a multiple of 8).

DST_ROWS

Maximum Height of output image for which the hardware kernel would be built.

DST_COLS

Maximum Width of output image for which the hardware kernel would be built (must be a multiple of 8).

NPC

Number of pixels to be processed per cycle. Possible options are XF_NPPC1 (1 pixel per cycle) and XF_NPPC8 (8 pixel per cycle).

MAX_DOWN_SCA LE

Set to 2 for all 1 pixel modes, and for upscale in x direction. When down scaling in x direction in 8-pixel mode, please set this parameter to the next highest integer value of the down scale factor i.e., if downscaling from 1920 columns to 1280 columns, set to 2. For 1920 to 640, set to 3.

_src

Input Image

_dst

Output Image

Resource Utilization

The following table summarizes the resource utilization of Resize function in Resource Optimized (8 pixel) mode and Normal mode, as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA.

Operating Mode

Utilization Estimate

1 Pixel (at 300 MHz)

8 Pixel (at 150MHz)

IMAGESIZE

LUTs

FFs

DSPs

BRAMs

IMAGESIZE

LUTs

FFs

DSPs

BRAMs

Downscale Nearest Neighbor

1920X1080 TO 960X1620

1089

1593

4

2

3840X2160 TO 1920X1080

2545

2250

4

12

Downscale Bilinear

1920X1080 TO 960X1080

1340

1846

8

2

3840X2160 TO 1920X1080

5159

3092

36

12

Downscale Area

3840X2160 TO 1920X1080

6146

8338

19

10

3840X2160 TO 1920X1080

17892

19758

162

16

Upscale Nearest Neighbor

1920X1080 TO 3840X540

1089

1593

4

2

1920X1080 TO 3840X2160

1818

1686

4

6

Upscale Bilinear

1920X1080 TO 3840X540

1340

1846

8

2

1920X1080 TO 3840X2160

3697

2739

36

6

Upscale Area

1920X1080 TO 3840X2160

5811

8773

28

32

1920X1080 TO 3840X2160

12214

14003

98

24

The following table summarizes the resource utilization of Resize function in Normal mode, as generated in the Vivado HLS 2019.1 tool for the Xilinx xczu9eg-ffvb1156-2-i-es2 FPGA for 3channel image as input.

Operating Mode

Utilization Estimate

1 Pixel (at 300 MHz)

IMAGESIZE

LUTs

FFs

DSPs

BRAMs

Downscale Nearest Neighbor

3840X2160 TO 1920X108

1184

168

4

18

Downscale Bilinear

3840X2160 TO 1920X1080

1592

2058

14

18

Downscale Area

3840X2160 TO 1920X1080

3212

4777

104

72

Upscale Nearest Neighbor

1920X1080 TO 3840X2160

1166

1697

4

9

Upscale Bilinear

1920X1080 TO 3840X2160

1574

2053

14

9

Upscale Area

1920X1080 TO 3840X2160

1731

2733

36

31

Performance Estimate

The following table summarizes the performance estimation of Resize for various configurations, as generated in the Vivado HLS 2019.1 tool for the xczu9eg-ffvb1156-2-i-es2 FPGA at 300 MHz to resize a grayscale image from 1080x1920 to 480x640 (downscale); and to resize a grayscale image from 1080x1920 to 2160x3840 (upscale). This table also shows the latencies obtained for different interpolation types.

Operating Mode

Operating Frequency

(MHz)

Latency Estimate (ms)

Downscale

NN

Downscale

Bilinear

Downscale

Area

Upscale

NN

Upscale

Bilinear

Upscale

Area

1 pixel

300

6.94

6.97

7.09

27.71

27.75

27.74

BGR2HSV

The BGR2HSV function converts the input image color space to HSV color space and returns the HSV image as the output.

API Syntax

template<int SRC_T, int ROWS, int COLS,int NPC=1>
          void BGR2HSV(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst_mat)

Parameter Descriptions

The table below describes the template and the function parameters.

Parameter

Description

SRC_T

Input pixel type should be XF_8UC3

DST_T

Output pixel type should be XF_8UC3

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed per cycle. Only XF_NPPC1 is supported.

_src_mat

Input image

_dst_mat

Output image

Scharr Filter

The Scharr function computes the gradients of input image in both x and y direction by convolving the kernel with input image being processed.

For Kernel size 3x3:

  • GradientX: image147

  • GradientY: image148

API Syntax

template<int BORDER_TYPE, int SRC_T,int DST_T, int ROWS, int COLS,int NPC=1>
void Scharr(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_matx,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_maty)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Scharr Parameter Description

Parameter

Description

BORDER_TYPE

Border type supported is XF_BORDER_CONSTANT

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

DST_T

Output pixel type. Only 8-bit unsigned, 16-bit signed,1 and 3 channels are supported (XF_8UC1, XF_16SC1,XF_8UC3 and XF_16SC3)

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src_mat

Input image

_dst_matx

X gradient output image.

_dst_maty

Y gradient output image.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

3

6

DSP48E

0

0

FF

728

1434

LUT

812

2481

CLB

171

461

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K 3 channel image.

Name

Resource Utilization

1 pixel per clock operation

300 MHz

BRAM_18K

18

DSP48E

0

FF

1911

LUT

1392

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Table . Scharr Function Performance Estimate Summary

Operating Mode

Operating Frequency

(MHz)

Latency

(ms)

1 pixel

300

7.2

8 pixel

150

1.7

Set

The Set function sets the each pixel in input image to a given scalar value and stores the result in dst.

API Syntax

template< int SRC_T , int ROWS, int COLS, int NPC=1>
void set(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, unsigned char _scl[XF_CHANNELS(SRC_T,NPC)], xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Set Parameter Description

Parameter

Description

SRC_T

Input pixel type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. Must be multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed per cycle.

_src1

First input image

_scl

Scalar value

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the Set function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

87

87

LUT

43

42

CLB

17

18

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Sobel Filter

The Sobel function Computes the gradients of input image in both x and y direction by convolving the kernel with input image being processed.

  • For Kernel size 3x3

    • GradientX: image149

    • GradientY: image150

  • For Kernel size 5x5

    • GradientX: image151

    • GradientY: image152

  • For Kernel size 7x7

    • GradientX: image153

    • GradientY: image154

API Syntax

template<int BORDER_TYPE,int FILTER_TYPE, int SRC_T,int DST_T, int ROWS, int COLS,int NPC=1,bool USE_URAM=false>
void Sobel(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_matx,xf::cv::Mat<DST_T, ROWS, COLS, NPC> & _dst_maty)

Parameter Descriptions

The following table describes the template and the function parameters.

Parameter

Description

FILTER_TYPE

Filter size. Filter size of 3 (XF_FILTER_3X3), 5 (XF_FILTER_5X5) and 7 (XF_FILTER_7X7) are supported.

BORDER_TYPE

Border Type supported is XF_BORDER_CONSTANT

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

DST_T

Output pixel type. Only 8-bit unsigned, 16-bit signed,1 and 3 channels are supported (XF_8UC1, XF_16SC1,XF_8UC3 and XF_16SC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

USE_URAM

Enable to map storage structures to UltraRAM

_src_mat

Input image

_dst_matx

X gradient output image.

_dst_maty

Y gradient output image.

  1. Sobel 7x7 8-pixel is not supported.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Operating Mode

Filter Size

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

1 pixel

3x3

300

3

0

609

616

135

5x5

300

5

0

1133

1499

308

7x7

300

7

0

2658

3334

632

8 pixel

3x3

150

6

0

1159

1892

341

5x5

150

10

0

3024

5801

999

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a 4K 3 Channel image.

Operating Mode

Filter Size

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 pixel

3x3

300

18

0

1047

1107

5x5

300

30

0

5370

3312

7x7

300

42

0

6100

5496

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx xczu7ev-ffvc1156-2-e FPGA, to process a grayscale 4K (3840x2160) image with UltraRAM enable.

Operating Mode

Filter Size

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

URAM

DSP_48Es

FF

LUT

1 pixel

3x3

300

0

1

0

919

707

5x5

300

0

1

0

2440

1557

7x7

300

0

1

0

4066

3495

8 pixel

3x3

150

0

3

0

1803

2050

5x5

150

0

5

0

4159

6817

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1, to process a grayscale HD (1080x1920) image.

Operating Mode

Operating Frequency

(MHz)

Filter Size

Latency Estimate

(ms)

1 pixel

300

3x3

7.5

300

5x5

7.5

300

7x7

7.5

8 pixel

150

3x3

1.7

150

5x5

1.71

Semi Global Method for Stereo Disparity Estimation

Stereo matching algorithms are used for finding relative depth from a pair of rectified stereo images. The resultant disparity information can be used for 3D reconstruction by triangulation, using the known intrinsic and extrinsic parameters of the stereo camera. The Semi global method for stereo disparity estimation aggregates the cost in terms of dissimilarity across multiple paths leading to a smoother estimate of the disparity map.

For the semi-global method in Vitis Vision, census transform in conjunction with Hamming distance is used for cost computation. The semiglobal optimization block is based on the implementation by Hirschmuller, but approximates the cost aggregation by considering only four directions.

Parallelism is achieved by computing and aggregating cost for multiple disparities in parallel, and this parameter is included as a compile-time input.

API Syntax

template<int BORDER_TYPE, int WINDOW_SIZE, int NDISP, int PU, int R, int SRC_T, int DST_T, int ROWS, int COLS, int NPC>

void SemiGlobalBM(xf::cv::Mat<SRC_T,ROWS,COLS,NPC> & _src_mat_l, xf::cv::Mat<SRC_T,ROWS,COLS,NPC> & _src_mat_r, xf::cv::Mat<DST_T,ROWS,COLS,NPC> & _dst_mat, uint8_t p1, uint8_t p2)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . SemiGlobalBM Parameter Description

Parameter

Description

BORDER_TYPE

The border pixels are processed in Census transform function based on this parameter. Only XF_BORDER_CONSTANT is supported.

WINDOW_SIZE

Size of the window used for Census transform computation. Only ‘5’ (5x5) is supported.

NDISP

Number of disparities

PU

Number of disparity units to be computed in parallel

R

Number of directions for cost aggregation. It must be 2, 3, or 4.

SRC_T

Type of input image Mat object. It must be XF_8UC1.

DST_T

Type of output disparity image Mat object. It must be XF_8UC1.

ROWS

Maximum height of the input image.

COLS

Maximum width of the input image.

NPC

Number of pixels to be computed in parallel. It must be XF_NPPC1.

_src_mat_l

Left input image Mat

_src_mat_r

Right input image Mat

_dst_mat

Output disparity image Mat

p1

Small penalty for cost aggregation

p2

Large penalty for cost aggregation. The maximum value is 100.

Resource Utilization

The following table summarizes the resource utilization for a 1920 x 1080 image, with 64 number of disparities, and 32 parallel units.

Operating Mode

Filter Size

Operating Frequency (MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

1 Pixel

5x5

200

205

141

11856

19102

Performance Estimate

The following table summarizes a performance estimate for a 1920x1080 image.

Table . SemiGlobalBM Function Performance Estimate Summary

Operating Mode

Operating Frequency

Number of Disparities

Parallel Units

Latency

1 pixel/clock

200 MHz

64

32

42 ms

Stereo Local Block Matching

Stereo block matching is a method to estimate the motion of the blocks between the consecutive frames, called stereo pair. The postulate behind this idea is that, considering a stereo pair, the foreground objects will have disparities higher than the background. Local block matching uses the information in the neighboring patch based on the window size, for identifying the conjugate point in its stereo pair. While, the techniques under global method, used the information from the whole image for computing the matching pixel, providing much better accuracy than local methods. But, the efficiency in the global methods are obtained with the cost of resources, which is where local methods stands out.

Local block matching algorithm consists of pre-processing and disparity estimation stages. The pre-processing consists of Sobel gradient computation followed by image clipping. And the disparity estimation consists of SAD (Sum of Absolute Difference) computation and obtaining the disparity using winner takes all method (least SAD will be the disparity). Invalidity of the pixel relies upon its uniqueness from the other possible disparities. And the invalid pixels are indicated with the disparity value of zero.

API Syntax

template <int WSIZE, int NDISP, int NDISP_UNIT, int SRC_T, int DST_T, int ROWS, int COLS, int NPC = XF_NPPC1,bool USE_URAM=false>
void StereoBM(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_left_mat, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> &_right_mat, xf::cv::Mat<DST_T, ROWS, COLS, NPC> &_disp_mat, xf::cv::xFSBMState<WSIZE,NDISP,NDISP_UNIT> &sbmstate);

Parameter Descriptions

The following table describes the template and the function parameters.

Table . StereoBM Parameter Description

Parameter

Description

WSIZE

Size of the window used for disparity computation

NDISP

Number of disparities

NDISP_UNITS

Number of disparities to be computed in parallel.

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1)

DST_T

Output type. This is XF_16UC1, where the disparities are arranged in Q12.4 format.

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 only.

USE_URAM

Enable to map some storage structures to UltraRAM

left_image

Image from the left camera

right_image

Image from the right camera

disparity_im age

Disparities output in the form of an image.

sbmstate

Class object consisting of various parameters regarding the stereo block matching algorithm.

  1. preFilterCap: Default value is 31, can be altered by the user, value ranges from 1 to 63.

  2. minDisparity: Default value is 0, can be altered by the user, value ranges from 0 to (imgWidth-NDISP).

  3. uniquenessRatio: Default set to 15, but can be altered to any non-negative integer.

  4. textureThreshold: Default set to 10, but can be modified to any non-negative integer.

Resource Utilization

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 version tool for the Xilinx® Xczu9eg-ffvb1156-1-i-es1 FPGA, to progress a grayscale HD (1080x1920) image.

The configurations are in the format: imageSize_WSIZE_NDisp_NDispUnits.

Configurations

Frequency

(MHz)

Resource Utilization

BRAM_18k

DSP48E

FF

LUT

HD_5_16_2

300

37

20

6856

7181

HD_9_32_4

300

45

20

9700

10396

HD_11_32_32

300

49

20

34519

31978

HD_15_128_32

300

57

20

41017

35176

HD_21_64_16

300

69

20

29853

30706

The following table summarizes the resource utilization of the kernel in different configurations, generated using Vivado HLS 2019.1 version tool for the Xilinx xczu7ev-ffvc1156-2-e FPGA, to progress a grayscale HD (1080x1920) image with UltraRAM enable.

The configurations are in the format: imageSize_WSIZE_NDisp_NDispUnits.

Configurations

Frequency

(MHz)

Resource Utilization

BRAM_18k

URAM

DSP48E

FF

LUT

HD_5_16_2

300

0

12

20

7220

6529

HD_9_32_4

300

0

12

20

10186

9302

HD_11_32_32

300

0

14

20

44046

30966

HD_15_128_32

300

0

14

20

50556

38132

HD_21_64_16

300

0

16

20

35991

28464

Performance Estimate

The following table summarizes a performance estimate of the Stereo local block matching in different configurations, as generated using Vivado HLS 2019.1 tool for Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

The configurations are in the format: imageSize_WSIZE_NDisp_NDispUnits.

Configurations

Frequency

(MHz)

Latency (ms)

Min

Max

HD_5_16_2

300

55.296

55.296

HD_9_32_4

300

55.296

55.296

HD_11_32_32

300

6.912

6.912

HD_15_48_16

300

20.736

20.736

HD_15_128_32

300

27.648

27.648

HD_21_64_16

300

27.648

27.648

SubRS

The SubRS function subtracts the intensity of the source image from a scalar image and stores it in the destination image.

dst(I)= scl - src(I)

API Syntax

template<int POLICY_TYPE, int SRC_T, int ROWS, int COLS, int NPC =1>
void subRS(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, unsigned char _scl[XF_CHANNELS(SRC_T,NPC)],xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . SubRS Parameter Description

Parameter

Description

SRC_T

Input Pixel Type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

First Input image

_scl

Input scalar value,the size should be number of channels

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the SubRS function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

103

104

LUT

44

133

CLB

23

43

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

SubS

The SubS function subtracts a scalar value from the intensity of source image and stores it in the destination image.

dst(I)= src(I) - scl

API Syntax

template<int POLICY_TYPE, int SRC_T, int ROWS, int COLS, int NPC =1>
void subS(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1, unsigned char _scl[XF_CHANNELS(SRC_T,NPC)],xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . SubS Parameter Description

Parameter

Description

SRC_T

Input Pixel Type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

First Input image

_scl

Input scalar value, the size should be the number of channels.

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the SubS function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

103

104

LUT

44

133

CLB

23

43

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

Sum

The sum function calculates the sum of all pixels in input image.

API Syntax

template< int SRC_T , int ROWS, int COLS, int NPC=1>
void sum(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src1,double sum[XF_CHANNELS(SRC_T,NPC)])

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Sum Parameter Description

Parameter

Description

SRC_T

Input pixel type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image (must be multiple of 8).

NPC

Number of pixels to be processed per cycle.

_src1

Input image.

sum

Array to store sum of all pixels in the image.

Resource Utilization

The following table summarizes the resource utilization of the Sum function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

341

408

LUT

304

338

CLB

71

87

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

8 pixel operation (150 MHz)

SVM

The SVM function is the SVM core operation, which performs dot product between the input arrays. The function returns the resultant dot product value with its fixed point type.

API Syntax

template<int SRC1_T, int SRC2_T, int DST_T, int ROWS1, int COLS1, int ROWS2, int COLS2, int NPC=1, int N>
void SVM(xf::cv::Mat<SRC1_T, ROWS1, COLS1, NPC> &in_1, xf::cv::Mat<SRC2_T, ROWS2, COLS2, NPC> &in_2, uint16_t idx1, uint16_t idx2, uchar_t frac1, uchar_t frac2, uint16_t n, uchar_t *out_frac, ap_int<XF_PIXELDEPTH(DST_T)> *result)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . SVM Parameter Description

Parameters

Description

SRC1_T

Input pixel type. 16-bit, signed, 1 channel (XF_16SC1) is supported.

SRC2_T

Input pixel type. 16-bit, signed, 1 channel (XF_16SC1) is supported.

DST_T

Output data Type. 32-bit, signed, 1 channel (XF_32SC1) is supported.

ROWS1

Number of rows in the first image being processed.

COLS1

Number of columns in the first image being processed.

ROWS2

Number of rows in the second image being processed.

COLS2

Number of columns in the second image being processed.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1.

N

Max number of kernel operations

in_1

First Input Array.

in_2

Second Input Array.

idx1

Starting index of the first array.

idx2

Starting index of the second array.

frac1

Number of fractional bits in the first array data.

frac2

Number of fractional bits in the second array data.

n

Number of kernel operations.

out_frac

Number of fractional bits in the resultant value.

result

Resultant value

Resource Utilization

The following table summarizes the resource utilization of the SVM function, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Frequency (MHz)

Utilization Estimate (ms)

BRAM_18K

DSP_48Es

FF

LUT

CLB

300

0

1

27

34

12

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Frequency (MHz)

Latency Estimate

Min (cycles)

Max (cycles)

300

204

204

Thresholding

The Threshold function performs thresholding operation on the input image. There are several types of thresholding supported by the function.

API Syntax

template<int THRESHOLD_TYPE, int SRC_T, int ROWS, int COLS,int NPC=1>
void Threshold(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src_mat,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst_mat,short int thresh,short int maxval )

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Threshold Parameter Description

Parameter

Description

THRESHOLD_TY PE

Type of thresholding.

SRC_T

Input pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. Must be multiple of 8, for 8-pixel operation.

NPC

Number of pixels to be processed per cycle.

_src_mat

Input image

_dst_mat

Output image

thresh

Threshold value.

maxval

Maximum value to use with the THRESH_BINARY and THRESH_BINARY_INV thresholding types.

Resource Utilization

The following table summarizes the resource utilization of the kernel with binary thresholding in different configurations, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1 FPGA, to process a grayscale HD (1080x1920) image.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

110

154

LUT

61

139

CLB

16

37

Performance Estimate

The following table summarizes the performance of the kernel in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1, to process a grayscale HD (1080x1920) image.

Table . Threshold Function Performance Estimate Summary

Operating Mode

Operating Frequency

(MHz)

Latency Estimate

(ms)

1 pixel

300

7.2

8 pixel

150

1.7

Atan2

The Atan2LookupFP function finds the arctangent of y/x. It returns the angle made by the vector image155 with respect to origin. The angle returned by atan2 will also contain the quadrant information.

Atan2LookupFP is a fixed point version of the standard atan2 function. This function implements the atan2 using a lookup table approach. The values in the look up table are represented in Q4.12 format and so the values returned by this function are in Q4.12. A maximum error of 0.2 degrees is present in the range of 89 to 90 degrees when compared to the standard atan2 function available in glibc. For the other angles (0 to 89) the maximum error is in the order of 10-3. This function returns 0 when both xs and ys are zeroes.

API Syntax

short Atan2LookupFP(short xs, short ys, int M1,int N1,int M2, int N2)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Atan2LookupFP Parameter Description

Parameter

Description

xs

16-bit signed value x in fixed point format of QM1.N1

ys

16-bit signed value y in fixed point format of QM2.N2

M1

Number of bits to represent integer part of x.

N1

Number of bits to represent fractional part of y. Must be equal to 16-M1.

M2

Number of bits to represent integer part of y.

N2

Number of bits to represent fractional part of y. Must be equal to 16-N1.

Return

Return value is in radians. Its range varies from -pi to +pi in fixed point format of Q4.12

..rubric:: Resource Utilization

The following table summarizes the resource utilization of the Atan2LookupFP function , generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

300

4

2

275

75

139

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Frequency

(MHz)

Latency Estimate

Min (cycles)

Max (cycles)

300

1

15

Inverse (Reciprocal)

The Inverse function computes the reciprocal of a number x. The values of 1/x are stored in a look up table of 2048 size. The index for picking the 1/x value is computed using the fixed point format of x. Once this index is computed, the corresponding 1/x value is fetched from the look up table and returned along with the number of fractional bits needed to represent this value in fixed point format.

API Syntax

unsigned int Inverse(unsigned short x,int M,char *N)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Inverse Parameter Description

Parameter

Description

x

16-bit unsigned value x in fixed point format of QM.(16-M)

M

Number of bits to represent integer part of x.

N

Pointer to a char variable which stores the number of bits to represent fractional part of 1/x. This value is returned from the function.

Return

1/x value is returned in 32-bit format represented by a fixed point format of Q(32-N).N

Resource Utilization

The following table summarizes the resource utilization of the Inverse function, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

300

4

0

68

128

22

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Frequency

(MHz)

Latency Estimate

Min (cycles)

Max (cycles)

300

1

15

Square Root

The Sqrt function computes the square root of a 16-bit fixed point number using the non-restoring square root algorithm. The non-restoring square root algorithm uses the two’s complement representation for the square root result. At each iteration the algorithm can generate exact result value even in the last bit.

Input argument D must be 16-bit number, though it is declared as 32-bit. The output sqrt(D) is 16-bit type. If format of D is QM.N (where M+N = 16) then format of output is Q(M/2).N

To get a precision of ‘n’ bits in fractional part, you can simply left shift the radicand (D) by ‘2n’ before the function call and shift the solution right by ‘n’ to get the correct answer. For example, to find the square root of 35 (01100011:sub:2) with one bit after the decimal point, that is, N=1:

  1. Shift the number (0110001100:sub:2) left by 2

  2. Shift the answer (1011:sub:2) right by 1. The correct answer is 101.1, which is 5.5.

API Syntax

int Sqrt(unsigned int D)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Sqrt Parameter Description

Parameter

Description

D

Input data in a 16-bit fixed-point format.

Return

Output value in short int format.

Resource Utilization

The following table summarizes the resource utilization of the Sqrt function, generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Frequency

(MHz)

Utilization Estimate

BRAM_18K

DSP_48Es

FF

LUT

CLB

300

0

0

8

6

1

Performance Estimate

The following table summarizes the performance in different configurations, as generated using Vivado HLS 2019.1 tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA.

Operating Frequency

(MHz)

Latency Estimate

Min (cycles)

Max (cycles)

300

18

18

WarpTransform

The warpTransform function is designed to perform the perspective and affine geometric transformations on an image. The type of transform is a compile time parameter to the function.

The function uses a streaming interface to perform the transformation. Due to this and due to the fact that geometric transformations need access to many different rows of input data to compute one output row, the function stores some rows of the input data in block RAMs/UltraRAMs. The number of rows the function stores can be configured by the user by modifying a template parameter. Based on the transformation matrix, you can decide on the number of rows to be stored. You can also choose when to start transforming the input image in terms of the number of rows of stored image.

Affine Transformation

The transformation matrix consists of size parameters, and is as shown:

image156

Affine transformation is applied in the warpTransform function following the equation:

image157

Perspective Transformation

The transformation matrix is a 3x3 matrix as shown below:

image158

Perspective transformation is applied in warpTransform following the equation:

image159

The destination pixel is then computed by dividing the first two dimensions of the dst1 by the third dimension

image160

API Syntax

template<int STORE_LINES, int START_ROW, int TRANSFORMATION_TYPE, int INTERPOLATION_TYPE, int SRC_T, int ROWS, int COLS, int NPC=1,bool USE_URAM=false>
void warpTransform(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & src, xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & dst, float *transformation_matrix)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . warpTransform Parameter Description

Parameter

Description

STORE_LINES

Number of lines to store an input to process a given transformation.

START_ROW

Number of the input rows to store before starting the image transformation. This must be less than or equal to STORE_LINES.

TRANSFORMATI ON_TYPE

Affine and perspective transformations are supported. Set this flag to ‘0’ for affine and ‘1’ for perspective transformation.

INTERPOLATIO N_TYPE

Set flag to ‘1’ for bilinear interpolation and ‘0’ for nearest neighbor interpolation.

SRC_T

Input and Output pixel type. Only 8-bit, unsigned, 1 and 3 channels are supported (XF_8UC1 and XF_8UC3)

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image.

NPC

Number of pixels to be processed per cycle; only one-pixel operation supported (XF_NPPC1).

USE_URAM

Enable to map some storage structures to UltraRAM

src

Input image

dst

Output image

transformati on_matrix

Transformation matrix that is applied to the input image.

Resource Utilization

The following table summarizes the resource utilization of the Warp transform, generated using Vivado HLS 2019.1 version tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Transformation

INTERPOLATION _TYPE

STORE _LINES

START _ROW

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

Perspective

Bilinear

100

50

300

7468

9804

61

112

Perspective

Nearest Neighbor

100

50

300

4514

6761

35

104

Affine

Bilinear

100

50

300

6139

5606

40

124

Affine

Nearest Neighbor

100

50

300

4611

4589

18

112

The following table summarizes the resource utilization of the Warp transform, generated using Vivado HLS 2019.1 version tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a BGR 4K image.

Transformation

INTERPOLATION _TYPE

STORE _LINES

START _ROW

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

Perspective

Bilinear

100

50

300

9192

7910

48

616

Perspective

Nearest Neighbor

100

50

300

10533

12055

69

604

Affine

Bilinear

100

50

300

6397

8415

35

604

The following table summarizes the resource utilization of the Warp transform, generated using Vivado HLS 2019.1 version tool for the Xilinx xczu7ev-ffvc1156-2-e FPGA, to progress a grayscale 4K image with UltraRAM enabled.

Transformation

INTERPOLATION _TYPE

STORE _LINES

START _ROW

Operating Frequency

(MHz)

Utilization Estimate

LUTs

FFs

DSPs

BRAMs

URAM

Perspective

Bilinear

100

50

300

7820

12458

61

7

12

Perspective

Nearest Neighbor

100

50

300

4880

8323

35

2

6

Affine

Bilinear

100

50

300

6850

9516

40

13

12

Affine

Nearest Neighbor

100

50

300

4651

6548

18

6

6

Performance Estimate

The following table summarizes a performance estimate of the Warp transform, as generated using Vivado HLS 2019.1 tool for Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Summary

Transforma tion

INTERPOLATI ON _TYPE

STORE _LIN ES

START _ROW

Operatin g Frequenc y

(MHz)

Latency Estimate

Max (ms)

Perspectiv e

Bilinear

100

50

300

7.46

Perspectiv e

Nearest Neighbor

100

50

300

7.31

Affine

Bilinear

100

50

300

7.31

Affine

Nearest Neighbor

100

50

300

7.24

Zero

The Zero function sets the each pixel in input image to zero and stores the result in dst.

API Syntax

template< int SRC_T , int ROWS, int COLS, int NPC=1>
void zero(xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _src1,xf::cv::Mat<SRC_T, ROWS, COLS, NPC> & _dst)

Parameter Descriptions

The following table describes the template and the function parameters.

Table . Zero Parameter Description

Parameter

Description

SRC_T

Input Pixel Type. 8-bit, unsigned, 1 channel is supported (XF_8UC1).

ROWS

Maximum height of input and output image.

COLS

Maximum width of input and output image. In case of N-pixel parallelism, width should be multiple of N.

NPC

Number of pixels to be processed per cycle; possible options are XF_NPPC1 and XF_NPPC8 for 1 pixel and 8 pixel operations respectively.

_src1

Input image

_dst

Output image

Resource Utilization

The following table summarizes the resource utilization of the Zero function in Resource optimized (8 pixel) mode and normal mode as generated using Vivado HLS 2019.1 version tool for the Xczu9eg-ffvb1156-1-i-es1 FPGA.

Name

Resource Utilization

1 pixel per clock operation

8 pixel per clock operation

300 MHz

150 MHz

BRAM_18K

0

0

DSP48E

0

0

FF

78

78

LUT

42

41

CLB

15

14

Performance Estimate

The following table summarizes a performance estimate of the kernel in different configurations, generated using Vivado HLS 2019.1 tool for Xczu9eg-ffvb1156-1-i-es1 FPGA to process a grayscale HD (1080x1920) image.

Operating Mode

Latency Estimate

Max Latency (ms)

1 pixel operation (300 MHz)

6.9

8 pixel operation (150 MHz)

1.7

1 N. Dalal, B. Triggs: Histograms of oriented gradients for human detection, IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2005.