Image sequence analysis using property coherence

The main goal in computer vision is to interpret motion of moving 3-D objects from a continuous sequence of 2-D images. Current research in computer vision is far below this goal. Nevertheless, the research in computer vision has rapidly increased in the past decade and established the basis for the practical use of computer vision technology in the real world environments. Motion analysis is currently becoming a major research focus.;This research explored the correspondence which is a low level motion analysis and is essential for any higher level motion analyses. The higher level motion analyses include motion estimation, recovery of 3-D structure and motion understanding which is one major goal in computer vision. These higher level motion analyses require the correspondence process as preprocessing.;Currently several point-token based correspondence processes have been developed for practical use. Our point-token based correspondence has very strong performance, is robust to noise (motion filtering), and shows real time response (target tracking); however, the point-token based correspondence needs a perfect algorithm for the corner point detection. Currently there is no perfect corner detector. Thus current point-token based correspondence schemes do not guarantee tracking of complex or unstructured objects.;This research is to overcome the weak point of the point-token based correspondence and to generalize the token types for more reliability and high performance of the correspondence. Thus tracking of some complex objects can be performed by the generalized-token based correspondence. The straight line correspondence has been developed for this goal through the research. The successful straight line correspondence provides the possibility of greater generalization of token-based correspondence because the straight line correspondence uses multiple parameters obtained from the line properties for the line token. The use of multiple properties reduces the probability of incorrect matching because the correspondence does not depend on any one property. This idea to use the multiple parameters for a line token can also be applied to the window-region based correspondence for more generalization of the token type which will be future work.