A KLT-Based Approach for Occlusion Handling in Human Tracking

Occlusions significantly affect the result during human tracking. This paper proposes a novel occlusion detection and handling algorithm which is mainly based on the KLT (Kanade-Lucas-Tomasi) method. Instead of using KLT as a tracker, we apply it for occlusion detection to enhance tracking stability. In this paper, a combinational method of particle filter tracking and occlusion detection is proposed. Depending on the detection result, our method makes decisions that whether to update the appearance model and use the occlusion handling strategy. Our occlusion detector associates color information, KLT feature tracker and directions of feature points. Additional, the occlusion handling strategy is based on the information from detection. Moreover, the algorithm also can solve the drift problem. Experimental results on famous datasets prove that our method has better performance and robustness on occlusion detection and handling.