Optical Flow with OpenCV

Optical Flow

Optical flow calculations on OpenSuse 15.2

Resources

• https://learnopencv.com/optical-flow-in-opencv/
• https://docs.opencv.org/master/d7/d9f/tutorial_linux_install.html
• https://github.com/spmallick/learnopencv.git

Install

To test python examples from the git repo directory learnopencv/Optical-Flow-in-OpenCV in a conda environment is easy

git clone https://github.com/spmallick/learnopencv.git
cd learnopencv/Optical-Flow-in-OpenCV/
conda create -n optical-flow python=3.8
conda activate optical-flow
pip install -r reqirements.txt

To run the c++ examples is a bit more involved. The Opensuse packages of opencv do not include all prerequisites and you get an error:

fatal error: opencv2/optflow.hpp: No such file or directory

To get rid of this, use a self compiled version of opencv with contrib part. It needs up to 2GB RAM per (c)make core, so I limited it to 4 cores. It chokes if it finds a python2, so I had to temporarily rename python2 stuff

for f in /usr/lib64/no-libpython2.7.so /usr/lib64/no-python2.7 /usr/bin/no-python2 /usr/bin/no-python2.7 /usr/bin/no-python2-config /usr/bin/no-python2.7-config
do 
  g=${f/no-/}; sudo mv $g $f
done

wget -O opencv.zip https://github.com/opencv/opencv/archive/master.zip
wget -O opencv_contrib.zip https://github.com/opencv/opencv_contrib/archive/master.zip
unzip opencv.zip
unzip opencv_contrib.zip
mkdir -p build && cd build
cmake -DPYTHON3_EXECUTABLE=/home/joachim/miniconda3/envs/Optical-Flow-in-OpenCV/bin/python -DOPENCV_EXTRA_MODULES_PATH=../opencv_contrib-master/modules ../opencv-master
cmake --build . --parallel 4
make install

Restore python2 stuff

for f in /usr/lib64/no-libpython2.7.so /usr/lib64/no-python2.7 /usr/bin/no-python2 /usr/bin/no-python2.7 /usr/bin/no-python2-config /usr/bin/no-python2.7-config
do 
  g=${f/no-/}; sudo mv $f $g
done

Now compile the sample code

cd algorithms
cmake .
make -j4

Execute Examples

Just paste the example commands from the git repo readme, e.g.:

python3 demo.py --algorithm lucaskanade --video_path videos/car.mp4

This will show 2 windows, one with an example video and one with the flow information

For the c++ installation, this would be the equivalent command:

./OpticalFlow ../videos/car.mp4 lucaskanade