An improved correlation filter tracking method with occlusion and drift handling

摘要

Despite remarkable progress, visual object tracking is still a challenging task as objects usually suffer from significant appearance changes, fast motion, and serious occlusion. In this paper, we propose a correlation filter-based tracking method with reliability evaluation and re-detection mechanism (CF-RERM) to deal with drift and occlusion problems. We first propose a criterion that uses the fluctuation trend of the response values, the displacement difference of the object, and the peak-to-sidelobe ratio to comprehensively evaluate the reliability of the tracking process. Then, a re-detection mechanism with a two-stage screening strategy is proposed for implementing the re-detection task when the criterion is triggered. Experimental results show that our method has achieved considerable performance in terms of accuracy and success rate on widely used OTB-50, OTB-100 and Temple-Color-128 tracking benchmark dataset. In addition, CF-RERM is able to achieve real-time tracking speed.