The main objective of this project is to develop digital methods for stabilizing shaky videos which is crucial for reliable sport video analysis.
-
The notebook 1_Introduction_to_Digital_video_stabilization.ipynb :
- justifies the need for video stabilization.
- presents examples related to sport that need a pre-processing with this technique before analysis.
- explains the steps of the stabilization process.
- presents state of the art methods that implement this process.
- defines a performance measure to assess the stabilization method.
-
The notebook 2_implementation_of_video_stabilization.ipynb :
-
generates artificially destabilized videos (zooming, translation, rotation, random_tranlsation/rotation).
-
lists state of the art methods.
-
-
The notebook 3_Performance_measure.ipynb makes a comparaison between ffmpeg based method which uses gaussian low pass filter algorithm and opencv_feature_matching_based method which trackes interest keypoints and tries to smooth their trajectory.
Metric used for video stabilization : pixel shift of interest keypoints