This research is aimed at continuation of a study on 3-dimensional (3D) motion analysis with current emphasis on performance analysis and evaluation. Two new problems with regard to motion will be pursued. The first problem is the matching of corresponding regions in consecutive images and the generation of a 3D object model from the image sequence. The model and the matching will take into consideration occlusion, hidden faces and shape changes of object faces caused by 3D rotation. The second problem is the computation of velocity field (optical flow) using linear or nonlinear shape change constraints instead of the conventional smoothness constaints. On a more practical level, rule-based system for 3D shape recovery and orientation determination from a single perspective view will be developed. The system is expected to have possibly 100 heuristic rules dealing with parallel lines, perpendicular lines, right corners, junction interpretations, geometric configurations, skewed symmetry, etc. Certainty factors will be assigned to the rules and evidences, control strategies will be devised, and the rule-based system will be implemented and tested on fairly complex 3D objects.
