A Study Of Video Moving Object Segmentation

Video object segmentation is a fundamental problem in computer vision, and has been studied widely. Video object segmentation techniques have many applications, such as video coding, video retrieval, video surveillance, multimedia production, image understanding, pattern recognition, etc. The goal of object segmentation is to partition the video frames into a series of spatially correlated regions along the time axis. Such separated image representation can provide flexible operations for videos. The object-based way is used in MPEG-4. The type of object-based video coding has not only improved the video compression, but also provided the object-based accessing functionality. In MPEG-7, the content organization and retrieval are required in the object-based way.This paper made research of video segmentation algorithm. It introduces the background and current status of research on video object segmentation. Then, as the basic theories of video segmentation, analyzing, image segmentation, mathematical morphology and motion detection are analyzed. The paper as follows:1,The background is constructed by the fourth cumulation. The gauss noise in subtracting background is reduced by using the fourth variance. A efficient video segmentation method is implemented. At first, the background construction technique based on higher order statistics of difference frame image is used to construct a background image. Then the video object is segmented by subtracting background based on the fourth variance. Experiments show that the subtracting background based on the fourth variance can get improved performance than subtracting background directly.2,A background subtraction method based on edge detection video segmentation algorithm is presented. The method is accomplished by using three kinds of edge map synthetically to get the moving edge of video object. That three edge maps are detected from the frame difference, background subtraction and background. Then the resulting moving edge map is fed to extract object by the region filling and morphological. Experiments show that the subtracting background based on edge detection improves the segmentation performance.3,A video segmentation based on global motion (camera motion) estimation is presented. The method estimates the global motion four parameters by the singular value decomposition of matrix. Then the coarse mask of foreground object is obtained by using morphological motion filtering(MMF). The moving object is segmented from video by using the mask and the information of frame edge synthetically.4,A mosaic panorama for videos algorithm based on global motion estimation is implemented. Based on the four global motion parameters estimated above, the affine model parameters are estimated by gradient descent method. Then the panorama is obtained by using the affine parameters. The method can make panorama correctly. There is no seam and information lost in the panorama by using multiple frames. 5,A motion estimation algorithm is presented based on improving the Horn-Schunck optical flow estimation method. Firstly, the forward gradients and backward gradients of the current frame are estimated. Then the occlusion problem in computing optical flow is reduced through selecting the gradients adaptive. Experiments show that the method can reduce the occlusion problem in the boundary of object, and it is also robust to noise.