Small project used to stabilize images/videos using homography computation.
Currently works with already extracted png-images. So use a tool such as ffmpeg to extract the images from a video you want stabilized. Then start the container using
docker run -t --rm -v <path/to/input/data>:/project/input -v <path/to/output/directory>:/project/output nickll/video-stabilizer
The script will store the stabilized images into the specified directory. Again, if you'd like to have a video from that, use ffmpeg.
The script will not crop the images but rather find the required size to pad the transformed images with the respective number of black pixels. Therefore, if there is a lot of movement, the images will be rather large.
Since the script is based an homographic transformations, it assumes that all pixels lie on a flat plane. For large 3-dimensional movements the results are likely going to be bad.