OpenCV Installation

The CV image operator relies on OpenCV. Before using this operator, compile OpenCV first.

Set up the OpenCV environment.

Install OpenCV dependencies. The Python version must be 3.2 or later.

yum install -y binutils cmake git pkgconfig gcc gcc-c++ lapack ffmpeg ffmpeg-devel libjpeg-* python3-devel python3-numpy

Apply the patch for the OpenCV 4.10.0 source code.
1. Obtain the OpenCV source package 4.10.0.zip.
2. Decompress the BoostKit-kp_cv_***.zip package to obtain the kcv.patch and kcv_AIOP.patch files.
3. Apply the patch files in the OpenCV source code directory based on the interfaces to be optimized.
  1
  unzip /home/opencv-4.10.0.zip && cd /home/opencv-4.10.0
  - kcv.patch is required for optimizing the cvtColor, dilate, GaussianBlur, merge, remap, and resize interfaces.
    1
    git apply /home/kcv.patch
  - kcv_AIOP.patch is required for optimizing the convertTo, threshold, warpAffine, contourArea, countNonZero, morphologyEx, copyMakeBorder, bitwise_and, bitwise_or, and t interfaces.
    1
    git apply /home/kcv_AIOP.patch
  - kcv_normalAndNCHW is optimized based on the user-defined normalAndNCHW interface. It is not an OpenCV interface function and can be called directly without applying any patch. For details, see kcv_normalAndNCHW.

In the opencv directory, run the cmake command to generate a makefile.

mkdir build
cmake -S . -B build \
    -D CMAKE_BUILD_TYPE=RELEASE \
    -D CMAKE_INSTALL_PREFIX=/usr/local \
    -D CMAKE_CXX_FLAGS="-O3  -march=armv8-a" \
    -D CMAKE_C_FLAGS="-O3  -march=armv8-a" \
    -D BUILD_opencv_python3=ON\
    -D OPENCV_GENERATE_PKGCONFIG=ON \
    -D PYTHON3_EXECUTABLE=$(which python3) 

Compile and install OpenCV.
1 2
cd build make -j && make install
The optimized OpenCV can be properly called.

Troubleshooting

Problem 1: "Error: invalid shift amount at operand 3 -- 'shll v22.4s,v19.4h,#8'" is displayed after assembly instructions such as shll are executed.

Solution: Install the FFmpeg dependency and libjpeg-related packages.
1
yum install ffmpeg libjpeg-*
Problem 2: Python cannot find cv2 after the import cv2 command is run.
Solution: cv2 is not correctly installed in the site-packages directory of Python. After you specify the Python path in the cmake command and run this command in 3, if the information shown in Figure 1 is displayed, the installation is successful. The information similar to Figure 2 is an example of incorrect installation.

Figure 1 Command output when cv2 is correctly installed

Figure 2 Command output when cv2 is not correctly installed
Problem 3: "typing_stubs_generation.nodes.type_node.TypeResolutionError: Failed to resolve alias 'GProtoArg' exposed as 'GProtoArg'" exposed as "GProtoArg"" is displayed during installation.

Solution: The ADE package cannot be downloaded from GitHub due to network issues.

The ADE package is used for automatic differentiation to enhance OpenCV functions in deep learning and machine learning, especially in scenarios where gradient calculation is required. This problem does not affect the use of KPCV.
Problem 4: The DNN module displays the message "[modules/dnn/CMakeFiles/opencv_dnn.dir/build.make: 2530: lib/libopencv_dnn.so.4.10.0] Error 1."

Solution: KPCV does not involve the DNN module. Add the compilation option -DBUILD_opencv_dnn=OFF during compilation to mask the DNN module.

Parent topic: CV Image Operators