Robust Hand Tracker Using Joint Temporal Weighted Histogram Features

Herein, a novel automatic hand-tracking approach based on temporal histogram features is proposed. Our method utilizes the joint temporal weighted histogram (JTWH) to track the hand robustly. When tracking begins, the hand model is initialized using a hand detector. During the tracking process, the hand model is updated using the most recent frame data and the hand tracker uses the weighted temporal model to track the hand persistently and robustly. The weights are calculated using the temporal and spatial similarity between the hand model and the current tracked hand. Because hand movement can be fast and may produce deformation, a weighted histogram was selected for the single hand model. Experiments demonstrate the proposed algorithm’s ability to track the moving hand robustly in comparison with several traditional hand-tracking algorithms. The proposed approach is robust in the complex background, and it can track the hand quickly and effectively.