Vision-based Adaptive Hand Tracking by Fusion of Depth and Skin Color

摘要

Hand tracking is the first step of natural human computer interaction. The existing hand detection and tracking algorithms are easily influenced by environmental factors such as light. To improve robustness the existing algorithms have many restrictions. In this paper, an adaptive hand tracking by the fusion of depth and skin features is put forward. In the hand segmentation stage, hand area is obtained based on an adaptive depth threshold. In the hand tacking stage, via skin-color feature of YCbCr space, the new tracking algorithm firstly establishes skin normalized histogram to establish the model of hand state. Then this paper presents a new tracking algorithm under particle filtering tracking framework by depth information to obtain the hand tracking area accurately. Experimental results show that, the proposed algorithm can achieve an accurate and robust tracking performance under complex background.