Refinement of noisy correspondence using feedback from 3D motion

In an automated feature-based motion analysis of multiple frames, correspondence da ta are usually noisy and fragmented. We developed a technique that gradually refines the initial noisy correspondence d a t a and links fragments of a single trajectory using feedback from 3-0 motion estimation. First, 3-0 motion parameters are estimated using the initial correspondence data. Then, each noisy trajectory is partitioned into subsets of points each of which conforms t o the estimated motion. The best set is used as the input t o the next motion estimation. This process is repeated and the gaps in the refined correspondence data are fi l led b y guidance from the predicted motion. Test results for a standard real image sequence are presented.