Install MinGW and CMake
Install OpenCV
Download Vitis Vision Library and Link to Vitis HLS
Run Simulation and Synthesis
Result
Reference

Published September 13, 2024 © GPL3+

Build and Install OpenCV in Vitis HLS 2022.2

This article introduces how to install OpenCV in Vitis HLS 2022.2 and successfully run examples from the Vitis Vision Library.

BeginnerFull instructions provided2 hours851

Build and Install OpenCV in Vitis HLS 2022.2

Things used in this project

Software apps and online services

OpenCV – Open Source Computer Vision Library OpenCV

MinGW

CMake

Vitis Vision Library

AMD-Xilinx - Vitis HLS

Story

First, I’d like to apologize for the content I shared previously. For those who have read my earlier post, the issue was that the content did not directly use OpenCV nor explain why certain commands were used or entered. Below is the revised version.

Recently, I've been researching how to write image processing IPs using Vitis HLS, which requires integration with OpenCV. However, the setup for OpenCV in Vitis HLS is somewhat complicated, which is why I'm sharing this process guide for everyone's reference.

The general process involves using CMake with MinGW to build and install OpenCV, then setting up the Linker Flags for OpenCV and Vitis Vision Library within Vitis HLS. Therefore, I will explain the process step by step in detail.

Install MinGW and CMake

MinGW

MinGW can be downloaded through this link.

Extract MinGW files somewhere. It looks something like this.

Add MinGW to your Environment variables. The path to the MinGW bin should be added to the Path variables.

CMake

CMake can be downloaded through this link.

Just download it and install through.msi file, and you will see the CMake GUI like the following image after you execute it.

Install OpenCV

OpenCV can be downloaded through this link. The version I use is 3416.

Execute the.exe and choose the location you want to extract.

Inside the OpenCV folder, create a new folder called “HLS_Build”. It could be any name you wish.

Now, it’s time to make OpenCV.

Run Cmake-gui through the start menu. In the sources field, choose the OpenCV source path. The source path should be something like I showed.

In the second field, it asks where you want to build the binaries. Here, you select the “HLS_Build” directory that we created previously. If all goes well, you should be able to click the “Configure” button.

1 / 3

We need to check and uncheck several options.

Check WITH_OPENGL
Check ENABLE_CXX11
Uncheck WITH_IPP
Uncheck ENABLE_PRECOMPILED_HEADERS

After making your selections, click the configure button again. If there are still red items after execution, continue clicking configure until all items turn white. Then click Generate.

When you see "Configuring done" and "Generating done", it indicates that the process is complete.

Open cmd in “HLS_Build” folder.

Enter mingw32-make to start the compilation. If there are no errors and the progress reaches 100%, congratulations, the compilation was successful.

Enter mingw32-make install. After the process is complete, an install directory will be generated in the “HLS_Build” folder.

Notice: You must properly set the LD_LIBRARY_PATH environment variable for dynamic library search paths and the OpenCV path information in your Environment variables for this script and the Vitis Vision example design to function correctly. Additionally, the paths for OpenCV's include libraries and binaries must be included in the system's environment variables. Otherwise, errors related to library inclusion will occur during simulation.

LD_LIBRARY_PATH  D:\opencv\HLS_Build\install\x64\mingw\lib
OPENCV_INCLUDE   D:\opencv\HLS_Build\install\include
OPENCV_LIB       D:\opencv\HLS_Build\install\x64\mingw\lib
Path             D:\opencv\HLS_Build\install\x64\mingw\lib;D:\opencv\HLS_Build\install\x64\mingw\bin

The path is based on the location you set yourself.

Download Vitis Vision Library and Link to Vitis HLS

Vitis Vision Library can be downloaded through this link.

For us, we are concerned with the “vision” libraries. Make sure you download the library because without this our projects won’t work.

Next, we can use the Sobel filter example to demonstrate how to use the Vitis library along with the newly installed OpenCV.

Create a new folder in the Vitis HLS workspace directory. Copy hls_sobel_axi_stream.cpp, hls_sobel_axi_stream.hpp and hls_sobel_axi_stream_tb.cpp from the above Sobel filter link.

Then, add the respective files to the Source and Test Bench sections.

At Project -> Project Settings, you need to add the necessary flags to the CFLAG and CSIMFLAG in the test file of simulation settings.

Note: All the paths below are user-defined. Please enter the paths according to your actual locations.

-ID:/xxx/Vitis_Libraries-2022.2/vision/L1/include -ID:\opencv\HLS_build\install\include -std=c++14 -Wno-unknown-pragmas

In the Linker Flags, enter the following code:

-L D:/opencv/HLS_Build/install/x64/mingw/lib -lopencv_imgcodecs3416  -lopencv_imgproc3416  -lopencv_core3416 -lopencv_highgui3416 -lopencv_flann3416 -lopencv_features2d3416

3416 refers to the OpenCV version. If you have installed a different version, you need to make the corresponding modifications.

The paths are set as mentioned above mainly because when calling functions from the Vitis Library L1, it will also link to the OpenCV libraries.

Here, the software version of OpenCV is directly included instead of the FPGA version, mainly because the testbench uses the software version of OpenCV to read and write images more conveniently.

Note: The testbench will not be directly involved in the subsequent synthesis and implementation process, because the software version of OpenCV cannot be used in synthesis and implementation; only the OpenCV within the Vitis library can be used for that purpose.

In the synthesis settings, you need to add the following to the CFLAGS in the module file:

-ID:/xxx/Vitis_Libraries-2022.2/vision/L1/include -std=c++14 -Wno-unknown-pragmas

Run Simulation and Synthesis

C Simulation

1 / 2

C Synthesis

1 / 2

Result

1 / 2

This shows that we are able to use the vision libraries in Vitis HLS with OpenCV in the Windows Environment. Now, the testbench takes and image and outputs and image. It also prints some text that we didn’t see when running it. The fix is simple and you should be able to figure that out.

Subsequently, the design will undergo implementation and be exported as an IP to Vivado for block design.

Reference

#include "hls_sobel_axi_stream.hpp"
#include "imgproc/xf_sobel.hpp"
#include "imgproc/xf_cvt_color.hpp"
#include "imgproc/xf_add_weighted.hpp"

void hls_sobel_axi_stream_top(
                hls::stream<ap_axiu<AXIS_W,1,1,1> >& _src,
                hls::stream<ap_axiu<AXIS_W,1,1,1> >& _dst,
                int rows,
                int cols) {

    #pragma HLS INTERFACE axis port=_src
    #pragma HLS INTERFACE axis port=_dst 
    #pragma HLS INTERFACE s_axilite port=rows           bundle=control
    #pragma HLS INTERFACE s_axilite port=cols           bundle=control
    #pragma HLS INTERFACE s_axilite port=return         bundle=control

    xf::cv::Mat<XF_8UC3, XF_HEIGHT, XF_WIDTH, XF_NPPC1> img_buf_0(rows, cols);
    xf::cv::Mat<XF_8UC1, XF_HEIGHT, XF_WIDTH, XF_NPPC1> img_buf_1(rows, cols);
    xf::cv::Mat<XF_8UC1, XF_HEIGHT, XF_WIDTH, XF_NPPC1> img_buf_1a(rows, cols);
    xf::cv::Mat<XF_8UC1, XF_HEIGHT, XF_WIDTH, XF_NPPC1> img_buf_1b(rows, cols);
    xf::cv::Mat<XF_8UC1, XF_HEIGHT, XF_WIDTH, XF_NPPC1> img_buf_2(rows, cols);
    xf::cv::Mat<XF_8UC3, XF_HEIGHT, XF_WIDTH, XF_NPPC1> img_buf_3(rows, cols);

    #pragma HLS dataflow

    xf::cv::AXIvideo2xfMat(_src, img_buf_0);
    xf::cv::bgr2gray<XF_8UC3, XF_8UC1, XF_HEIGHT, XF_WIDTH, XF_NPPC1>(img_buf_0, img_buf_1);
    xf::cv::Sobel<0, 3, XF_8UC1, XF_8UC1, XF_HEIGHT, XF_WIDTH, XF_NPPC1, false>(img_buf_1, img_buf_1a, img_buf_1b);
    xf::cv::addWeighted<XF_8UC1, XF_8UC1, XF_HEIGHT, XF_WIDTH, XF_NPPC1>(img_buf_1a, 0.5, img_buf_1b, 0.5, 0, img_buf_2);
    xf::cv::gray2bgr<XF_8UC1, XF_8UC3, XF_HEIGHT, XF_WIDTH, XF_NPPC1>(img_buf_2, img_buf_3);
    xf::cv::xfMat2AXIvideo(img_buf_3, _dst);

    return;
}

#include "hls_sobel_axi_stream.hpp"
#include "common/xf_headers.hpp"
#include "common/xf_axi.hpp"

using namespace std;

int main(int argc, char** argv) {
    if (argc != 2) {
        fprintf(stderr, "Invalid Number of Arguments!\nUsage: <executable> <image>\n");
        return -1;
    }

    cv::Mat img, diff;
    img = cv::imread(argv[1], 1);
    if (img.data == NULL) {
        fprintf(stderr, "Cannot open image at %s\n", argv[1]);
        return 0;
    }

    int rows = img.rows;
    int cols = img.cols;
    cv::Mat out_img(rows, cols, CV_8UC3);

    // convert input to axiStream
    hls::stream<ap_axiu<AXIS_W, 1, 1, 1> > _src;
    xf::cv::cvMat2AXIvideoxf<XF_NPPC1, AXIS_W>(img, _src);

    // output axiStream
    hls::stream<ap_axiu<AXIS_W, 1, 1, 1> > _dst;

    // Launch the kernel
    hls_sobel_axi_stream_top(_src, _dst, rows, cols);

    xf::cv::AXIvideo2cvMatxf<XF_NPPC1>(_dst, out_img);

    // Write output image
    cv::imwrite("output.png", out_img);

    return 0;
}

Credits

Norris Lin

6 projects • 5 followers

An R&D professional passionate about FPGA development, skilled in system and IP integration.

Contact

Comments

Please log in or sign up to comment.

Embed the widget on your own site

Build and Install OpenCV in Vitis HLS 2022.2

Build and Install OpenCV in Vitis HLS 2022.2

Things used in this project

Software apps and online services

Story

Install MinGW and CMake

Install OpenCV

Download Vitis Vision Library and Link to Vitis HLS

Run Simulation and Synthesis

Result

Reference

Code

hls_sobel_axi_stream.cpp

hls_sobel_axi_stream.hpp

hls_sobel_axi_stream_tb.cpp

Credits

Norris Lin

Comments

Related channels and tags