Correlation Filter-based Object Tracking Algorithms

Abstract

Object tracking is one of the most important tasks in computer vision. It is widely used in traffic monitoring, robotics, automatic vehicle tracking and the like. Discriminant tracking method based on correlation filtering theory has made a series of new progress due to its high efficiency and robustness. Basic algorithms, improved algorithms and algorithms combined deep learning on correlation filter-based object tracking are studied in this paper. Color-based, scale-based, part-based, and bound effect-based are included in these algorithms. Despite the broad application prospects of correlation filter in the field of object tracking, it is still a very challenging for research direction due to complex scenes and the object factors. 32 representative algorithms are compared on the OTB2013 and OTB100 datasets, experiment results show that the algorithm adopted by multiple features combination has better accuracy and higher success rate in the face of occlusion or position error.