Research On Image Segmentation Of Moving People Fusing Depth Information

Image segmentation is the basis of image analysis, recognition and understanding. Image segmentation, partitioning an image into different regions with some specific properties, has always been an import and challenging problem for many years. The main difficulties lie in the great variability of images and the presence of noises. In this paper, the segmentation of moving people and its application have been researched.First, based on a single camera, this paper introduces frames subtraction to detect moving objects. Meanwhile, much work is done in skin segmentation and face extraction. In different color spaces, skin color model is established and used to separate the color image into skin and non-skin regions through experiments. Then, based on Marr's theory, binocular stereo vision's basic theory, method and related techniques have been researched in this paper. The key contents are composed of fundamental theory, camera calibration and stereo matching. We give prominence to stereo matching. Based on the algorithm of area correlation, the pyramid frame matching is introduced. The constraint conditions of stereo matching are also used to increase the speed of matching process. The dense disparity map of moving people is extracted by stereo matching. Fusing the extracted depth, we complete the segmentation of moving people and improve the result. At last, an application about the segmentation of moving people-a people counting system is introduced, which comes from Dalian Everspry Image Technology Corporation Limited. The results of experiments and system tests are analyzed in the end of this paper.