I am running optical flow on a grid of fixed points(with Lucas Kanade) rather than running on every point (not with farnback). Is this still a dense optical flow. Is it the idea way to do dense. we are thinking of doing it this way to reduce the computation time. Can I get some better advise. We are looking to get the motion data from the video. Any dominant motion in the scene, we want to capture, with low computation time with moderate accuracy.
Thanks