Correlation Filter Based Moving Object Tracking With Scale Adaptation and Online Re-Detection

Object tracking is a difficult work in complex situations including crowded environment, occlusion, out of view, and fast motion. Recently, many tracking strategies have been designed to handle the object tracking in complex conditions. However, most of the designed methods are inefficient to tackle the target aspect ratio variation and disappearance problems during the long-term tracking. Hence, it is most important to design a tracking algorithm that effectively reduce the drifting problem and recapture the target from the tracking failure. In this paper, we proposed a robust correlation filter-based moving object tracker with scale adaptation and online re-detection. First, we trained a translation filter using kernelized correlation filter with the multiple features for identifying the initial target location in each frame. Second, we used the high confidence score of the correlation output to reduce the model-drifting problem. Third, we introduced a new online re-detection strategy to relocate the target at the time of tracking failure. This re-detection component activated dynamically based on the present and historical confidence scores of the target. To tackle the aspect ratio and scale variation problems, we used detection proposal with the correlation filter method. The experimental evaluation on the several benchmark datasets proved that our results significantly better compared with the other methods.