A Robust Correlation Filtering Tracker Based on Joint Reliability Evaluation of Target Position Changes

Abstract

Visual tracking is an important branch in computer vision. In complex scenarios, there exist various interference factors, e.g. background clutter, similar objects etc., making robust tracking based on correlation filter algorithm still a challenging task. In this paper, a correlation filter algorithm based on a novel adaptive multi-cue fusion strategy was proposed. First, a unified response map evaluation strategy was presented to assess the tracking reliability by combing the average peak correlation energy and the response peak historical information. Second, according to the cue reliability, an adaptive multi-cue fusion strategy was proposed to adaptively fuse two tracking cues, correlation filter and color histogram. The experimental results on OTB-2013 and UAV123 demonstrated that the proposed algorithm achieved competitive performance to the state-of the-art trackers.