Study On Segmentation And Tracking Of Moving Objects Based On Frame Difference Image

The distinct feature of MPEG-4,the second generation object-based codingstandard, is content description and coding based on video objects. So a videosequence must be segmented into a set of video objects first. Video objectsegmentation is also an opening part in the standard. The performance of thesegmentation algorithm is crucial to the final MPEG-4 coding products, so theauthor chooses this question for discussion.For the static background sequences, the author presents a segmentationalgorithm based on the combining of Canny edge detection and the fourth-ordermoment testing. Experiments show that the algorithm can separate movingobjects well. And for the moving background sequences, we should make theglobal motion estimation and compensation first, and then segment objects interms of the static one. Two methods of the global motion estimation areproposed in this paper, then using 8-parameter movement model forcompensation. After the initial segmentation, the author proposes amultiresolution algorithm based on the improved Hausdorff Distance to trackobjects. It can obtain the new objects accurately even if they are sheltered ortheir shapes have slow changes. Vectors between the last interested objects inprevious frame and the moving objects in current frame are established to followthe present interested objects by the trajectory analysis of every moving object.The hardware realization of the algorithm is a challenging issue which mustbe solved. It is also the future study direction that the author should make greatefforts on.