Hi,
I did start a Business based on OpenCV.
But now I am struggling to get precise basic movement data out of the functions
goodFeaturesToTrack(prev_gray, prev_pts, RdOpts.MaxTrackingPoints);
calcOpticalFlowPyrLK(prev_gray, curr_gray, prev_pts, curr_pts, status, err);
It looks obvious that based on the video content or the used lens aperture the result can vary between very good and clearly faulty. I did try to find a guide how to setup the stability parameters e.g. a fullframe camera with a 600/F4 lens (usually does not work) or a video record with an outdated noisy camera with high depth of field (usually works). For me it looks like when the stabilization functions have been developed the standard parameters have been set to the outdated video cameras and stay there even they do not fit to modern cameras anymore.
On modern shallow depth of field lenses FFMPG GPL’ed videostab library works while my OpenCV based stabilization fails.
I also did have some success by denoising, blurring and sharpening the frames before processing the stabilization for the payed version which is currently under development.
The old version with the standard parameters can be downloaded at
I would be pleased If you can provide additional information how to increase the stabilization precision how and what adverse effects these optional parameters do have.
Thank’s in advance,
Alexander Lindert