A Robust Algorithm for Long-term Object Tracking Based on PTAV

Abstract

Long-term object tracking is one of the most challenging problems in computer vision due to various factors such as deformation, abrupt motion, heavy occlusion and out-of-view. In this paper, we propose a tracking method based on Parallel Tracking and Verifying (PTAV). Firstly, we replace the fDSST tracker with a better performance tracker ECO-HC. Then we add self-test mechanism for the tracker, which include backtracking check using forward-backward overlap rate and multi-peak detection mechanism. At last, we modify the parallel framework between the tracker and the verifier in the PTAV, so that the algorithm can get timey feedback about the abnormal information. We perform experiments on the benchmark OTB-2015. Results show that our method has better accuracy and robustness in case of occlusion, out-of-view and other interference factors.