Skip to Main Content
In image sequence, target's features has two components: the spatial features which include the local background and nearby targets, and the temporal features which include all appearances of the targets seen previously. In this paper, we develop a multi-target visual tracking method based on mixed spatio-temporal features learning model which is a probabilistic inference model considering the above components. The proposed model combine the incremental appearance descriptor update strategy which can update descriptor dynamically according to previous appearances during tracking, and mix probabilistic data association which take targets' spatial features into account. In addition, we also apply the incremental update strategy into HSV histogram and region covariance descriptor, and compare these two descriptors in multi-target visual tracking. The results validate the proposed method in tracking moving multi-target in video streams.