- KalmanFilter.py     A Kalman filter 
- MothionAnalyser.py  Analyze the motion of frames
- NCC.py              GPU accelerated NCC algorithm
- Stabliser.py        Entry of the whole toolkit
- utils.py            Supporting functions
- Videoloader.py      Load the video
- VideoProcessor.py   Output the video, perform translation

Unfortunately, there is only the GPU version of the NCC algorithm... As a result you must have an Nvidia graphic card and a properly configured CUDA toolkit to run the code. The CuPy package is also obligatory,

The script is tested on a host with Ubuntu 18.04 LTS, with an Intel i7 processor, an RTX 2060 Super graphic card and 16GB RAM.

1. Install the CuPy with instructions from Installation Guide. The CuPy package names are different depending on the CUDA version you have installed on your host. For CUDA 10.1, execute pip install cupy-cuda101.
2. Install the OpenCV by pip install opencv-python and pip install opencv-python-contrib
3. Install other necessary packages
4. Run Stabliser.py with parameters, for example

python Stabliser.py --path_to_video ./videos/IMG_1442.MOV

5. The output video is saved to ./output/ by default