Improved target tracking based on spatio-temporal learning

Object tracking is always in the core status with the develop of robotics. In this paper, we proposed a novel tracking framework based on Spatio-Temporal Context tracking (STC) and Template-Matching algorithm(TM). STC tracking method is a fast and simplified tracking method that strongly dependent on the context region (the surrounding background of the target). Therefore, it cannot correct the tracking error by itself and even lost the target during tracking. Under this circumstance, we make a closed-loop judgment whether the movement of the target between two neighboring image sequence larger than a constant or not. If the movement larger than the constant, the template matching is employed as sample detector to correct the error or re-track the target on line. The template-matching algorithm is an efficient and fast detection algorithm that find the maximum probability point in the image that similar to the template. In order to decrease the calculating time, the search region is not the whole image but the context region of STC tracking method. What's more, after tracking and detecting we update the scale parameter to adapt the change of the target's appearance. Finally, experimental result demonstrate that our tracking method have improve the robustness of tracking.