Correlation filter tracking: Beyond an open-loop system

Abstract

Most existing Correlation Filter (CF) based trackers do not use any feedback from tracking output and can be considered as open-loop systems. They are prone to drifting when the object endures occlusion and large appearance changes. In this paper, we propose a generic self-correction mechanism for CF based trackers by introducing a closed-loop feedback technique. Our mechanism first detects the abnormality in tracking output using the Gaussian shape prior of a response map, and then estimates the tracking error by minimizing the discrepancy of tracking output and the expected response. An optimal offset is finally given to regulate the tracking process and prevent tracking drifting through a feedback loop. Extensive experiments have been conducted on four large-scale benchmarks, including OTB-2013, OTB-2015, TC-128, and UAV123@10fps. Experimental results show that our self-correction mechanism can be used to improve the overall performance of most CF based trackers by a large margin. Besides, our proposed Enhanced Dual Correlation Filters (EDCF) tracker outperforms the state-of-the-art methods and runs at a high speed nearly 80 fps.