Research On Segmentation Of Moving Objects In Dynamic Scene Video Sequences

With the rapid development of computer science and technology, the motion analysis based on video has become a hot topic in the computer vision field. As a key technology in the vision analysis of moving object, extracting motion objects from a dynamic scene plays a significant role for video analysis. It is also the basis of other motion analyses, such as motion object tracing, object classification and activity recognition, event detection and so on.This dissertation introduces some applications and developments of video object segmentation techniques in aspects of multimedia standard, video surveillance and sports analysis. It also summarizes research works of video segmentation from all over the world, and analyzes the video object segmentation methods, such as spatial intra-frame segmentations, global motion estimations and spatio-temporal segmentations. According to the characteristics of diving sport, we do research on how to retrieve the moving objects from a dynamic scene video.A non-backtracking searching algorithm for block matching in video is proposed to address the problem of motion estimation. Based on the idea of the KMP algorithm of character string for fast pattern matching, the algorithm converts the block matching in the video image into the pattern matching of a character string. The next function value of each pixel in the sub-block can be easily gotten by taking advantage of the self-matching trait of the block. And the r-th pixel from the next array of the block can be also found out so that to keep the comparing without backtracks to the start position of matching while the j-th pixel matching is lost.Because the diving athletes always have simple dresses and naked skins, we can get a few connected components by representing the video frames by graph. In view of this assertion, an irregular block matching video segmentation method based on the local graph is proposed. First, images are represented by the graph data-structure, and then the hypothesis test is adopted to merge the adjacent image block. After segmenting the first frame, according to the continuity of the frame motion, it takes advantage of the irregular block matching method to find the foreground area in the current frame and take the correlation coefficient method of variable correlation in the clustering analysis as the matching criteria for reference. Experimental results on diving and gymnastics videos show that the proposed method can segment the video motion objects efficiently.