Multi-Feature Integration and Occlusion Detect Mechanism Based Object Tracking Algorithm

Tracking failure happens in complex scenes such as background clutter and occlusion. Aiming at improving performance of the correlation filter based tracking algorithm, an object tracking algorithm that integrates multi-feature and occlusion detect mechanism is proposed. In this paper, multi-feature is utilized in the construction of object appearance model and background context model for better representing object and discriminating object from disturbance. In addition, an occlusion detection mechanism is introduced to effectively estimate whether the object is occluded or seriously disturbed, avoiding the useless model updating in the occlusion scene. Experimental results on the OTB-100 dataset show that the success rate score of the proposed algorithm is 0.631 and the precision rate score is 0.865 that better than other compared algorithms in many complex scenes.