Estimating optical flow from clustered trajectories invelocity-time

摘要

Presents a new algorithm for the estimation of optical flow from amonocular sequence of images using clustered trajectories invelocity-time. In the new algorithm, the objects in the scene mayexhibit rotation and translation in all three dimensions. In addition,interframe displacement may be large-of the order of many pixels. It isassumed that there is a known upper bound on the magnitudes of the x andy components of interframe displacement. The authors conducted tests tocompare the performance of the algorithm with that of two prioralgorithms for optical flow estimation. They present the results ofthese tests. The results suggest that the algorithm is an improvementover prior algorithms in its ability to compute the optical flow fieldaccurately under several commonly encountered scene conditions that haveposed problems to earlier algorithms for optical flow estimation