An Occlusion-Resolving Hand Tracking Method

This paper proposes a novel algorithm for real-time hand detection and tracking, and this algorithm can successfully track a hand even when it is overlapped with other skin-color objects during tracking. An on-line adaptive learning approach associated with negative skin-color exclusion is used to fit the skin color distribution of each individual hand in various environments. When skin-color objects have been extracted, three states (separation, proximity and overlap) between tracked objects are defined. A separation template image of the tracking hand is created whenever it is in the proximity state, and a feature-point-based matching comparison by using the newly created separation template is conducted when it is in the overlap state. The experimental results show the proposed algorithm not only can obtain a highly accurate hand tracking rate in various situations, but also can run in real time with 30-45 frames per second.