Stereoscopic video object segmentation based on depth and edge information

An effective scheme for stereoscopic video object segmentation was presented based on depth and edge information. Disparity estimation algorithm using region information was firstly proposed to get reliable depth information. In the proposed method, a new adaptive-size window approach was introduced to stereo matching in order to overcome problems with fixed-size window. Then, an improved watershed algorithm was exploited to achieve initial segmentation of depth image and coarse object areas were obtained. For each depth segments, edge detection was implemented to obtain accurate contours of object. Finally, watershed algorithm was utilized to segment object again. Using the extracted objects, an efficient motion tracking algorithm based on region was performed to track the video objects of the rest frames. The experimental results show that the proposed method can get good disparity map and segment video objects successfully.