Region feature space based target retrieval in video

A new robust target retrieval method in video is presented in this paper. The proposed approach uses spatio-temporal analysis to segment video in space-time domain. Then, a region feature space is defined according to the segment result, in which selected or given objects can be retrieved automatically in successive frames through local motion estimation. Various experiments show our algorithm is robust to partial occlusion, out-of-plane rotation and great relative movement among targets, scene and camera.