One-step backtracking for occlusion detection in real-time visual tracking
Occlusion is a challenging problem in real-time visual object tracking. Most state-of-the-art methods learn the inaccurate appearance of the target when it becomes occluded by other objects in the scene. To address this issue, a novel one-step backtracking (OB) tracker for occlusion detection is proposed, which backtracks to one previous frame and detects occlusion by comparing the tracking result with OB result in each frame. An adaptive learning model update scheme is further proposed by computing the peak-to-sidelobe ratio of the response maps to improve the tracking performance. Experiments on several benchmark sequences show that the proposed tracker outperforms state-of-the-art approaches and achieves real-time visual tracking.
