Context-Aware weighted for visual tracking

To solve the problem of target drift during the tracking process, this paper proposes a general framework based on the context-aware tracking framework. In the tracking process, the status of the correlation detection response map is observed in real-time through the APCE. When there is a single sharp peak on the correlation detection response map indicating that the tracking result is stable, the position of the correlation response maximum is taken as the target position, and the filter will perform zero-regression learning on the context of the target region. Conversely, when there are multiple peaks on the correlation detection response map indicating that the tracking result is unreliable, the position of the target is redetected in the multi-peak region, and the filter will perform weighted regression learning on these multi-peak regions. We apply the modified framework to the basic trackers DCF and SAMF for experimentation and compare the performance of it with the basic tracker and the corresponding tracker using the context-aware framework. The experimental results show that the tracker using our modified framework can accurately locate the target and avoid the target drift problem. Meanwhile, the precision rate and the success rate of the tracking are improved.