Contour motion estimation using 2D distribution of displacement vectors field

The motion information obtained from a sequence of time-varying two-dimensional images gives fundamental clues about the three-dimensional structure of the observed scene. For many vision systems, the main task is to extract that information. The authors present a quantitative and qualitative analysis of the 2D motion field obtained applying optical flow algorithms (with either global or local conditions), and pel-recursive algorithms to image sequences (using real and synthetic data). Special emphasis is placed on assessing the performance of both techniques in scene segmentation. The experiments were carried out in two stages: the first one consisted of a qualitative analysis of the results of the motion field estimation process for a known motion field; the second one was the quantitative evaluation of the applicability of the results to scene segmentation, pondering sensitivity and specificity as figures of merit. The results are presented and discussed