The prebuilt OpenCV releases does not contain the tracker module, which lives in the contrib code repository. Here is how to build OpenCV with contributions and with python bindings and java bindings.

Prerequisite

• Download and install Visual Studio 2017 Community Edition
• Download and install cmake
• Download and install Git
• Download and install Python 3.7.2 - Windows x86-64 executable installer. Add Python it to the path when asked.

You can verify the success calling python --version from a command shell and getting back Python 3.7.2.

Common steps

From a Git bash console, go to a freshly created working directory and run the following command to download OpenCV sources:

git clone -b 4.0.1 https://github.com/opencv/opencv.git
git clone -b 4.0.1 https://github.com/opencv/opencv_contrib.git

Build with python binding

Go to a freshly created working directory and create a virtual python environment where to install the python dependencies.

python -m venv venv

This should have created a subdirectory named venv, that contains packages and scripts. Now activate the virtual environment with:

. venv/Scripts/activate

Download and install numpy in the environment:

pip install numpy

Now numpy lives in ./venv/Lib/site-packages/numpy

It is time to generate Visual Studio project for the build definition. Create a build directory:

mkdir build
cd build

cmake \
  -G "Visual Studio 15 2017 Win64" \
  -DBUILD_opencv_python3=ON \
  -DINSTALL_CREATE_DISTRIB=ON \
  -DPYTHON3_NUMPY_INCLUDE_DIRS=$PWD/../venv/lib/site-packages/numpy/core/include \
  -DOPENCV_EXTRA_MODULES_PATH=../opencv_contrib/modules ../opencv

• -G "Visual Studio 15 2017 Win64" specifies which version of visual studio project to generate.
• -DBUILD_opencv_python3=ON requires to generate python module files.
• -DPYTHON3_NUMPY_INCLUDE_DIRS=[...] indicates where to find NumPy headers
• -DOPENCV_EXTRA_MODULES_PATH=[...] specifies to build OpenCV with the contribution repository

Finally build the binaries:

cmake --build . --config Release -- -maxcpucount:2
cmake --build . --config Release --target install

Note that Python release does not contain debug binaries, and consequently it is not possible to build OpenCV with debug symbols. Once the build is over ls python_loader/cv2 should return __init__.py config.py load_config_py2.py load_config_py3.py.

Build java bindings

Clear the build directory, and install java jdk (e.g. jdk-11.0.2) from Oracle. Select the jdk you intent to use for you project with the following command:

export JAVA_HOME='C:\Program Files\Java\jdk-11.0.2

Generate Visual Studio project files:

cmake \
  -G "Visual Studio 15 2017 Win64" \
  -DOPENCV_EXTRA_MODULES_PATH=../opencv_contrib/modules \
  -DBUILD_SHARED_LIBS=OFF \
  -DBUILD_FAT_JAVA_LIB=ON \
  -DBUILD_opencv_java=ON \
  -DANT_EXECUTABLE=$PWD/../apache-ant-1.10.5/bin/ant.bat \
  -DJNI_INCLUDE_DIRS='C:\Program Files\Java\jdk-11.0.2\include' \
  ../opencv

Verify that java and java_bindings_generator can be found on the line ` To be built: `. Finally run the build command:

cmake --build . --config Release -- -maxcpucount:2
cmake --build . --config Release --target install

Build C++

Even though the Java version is convenient to build a GUI thanks to Swing, and Python is very nice to transform lists thanks to comprehension, I had to C++ to debug OpenCV. Indeed, I wasn’t able to use my GPU with OpenCL and I needed to step into OpenCV code & profile it. Here is how to build OpenCV for C++, with TBB enabled:

Download TBB from https://www.threadingbuildingblocks.org/download and unzip in you working directory. It creates a subdirectory with the release name (e.g. tbb2019_20190206oss)

Set the environment variable calling the script:

tbb2019_20190206oss\bin\tbbvars.bat

Now generate the projects:

mkdir build
cd build
cmake \
  -G "Visual Studio 15 2017 Win64" \
  -DINSTALL_CREATE_DISTRIB=ON \
  -DBUILD_SHARED_LIBS=ON \
  -DOPENCV_EXTRA_MODULES_PATH=../opencv_contrib/modules \
  -DWITH_TBB=ON \
  ../opencv

Finally run the build command:

cmake --build . --config Release -- -maxcpucount:2
cmake --build . --config Release --target install
cmake --build . --config Debug -- -maxcpucount:2
cmake --build . --config Debug --target install

In Visual Studio is created a solution directory next to tbb2019_20190206oss, and build. Edit the project properties to add the following:

• Configuration Properties
  • Debugging
    • Environment: PATH=$(ProjectDir)\..\..\tbb2019_20190206oss\bin\intel64\vc14;$(ProjectDir)\..\..\build\bin\Debug;%PATH%
  • C/C++
    • Additional Include Directories: ..\..\opencv_build_release\install\include
  • Linker
    • General
      • Additional Library Directories: ..\..\opencv_build_release\lib\Debug for debug and ..\..\opencv_build_release\lib\Release for Release
    • Input
      • AdditionalDependencies : opencv_world401d.lib for Debug and opencv_world401.lib for Release