摘要：3D hand tracking in recent years has become a hot spot in Human Computer Interaction(HCI), and it also exists many difficulties. Our research roots in the observation model which is used to measure similarity between hand image and hand model for hand tracking. Firstly, using edge feature extracted from hand image and hand model projection contour to get a probability function, which measures the similarity between hand model and hand image according their chamfer distance. Secondly, using skin color segmentation algorithm to get binary hand image, at the same time we can get hand model silhouette from hand model projection points. After getting the binary hand image and hand model silhouette we then build another probability function based silhouette feature. Finally we combine the two functions to build the observation model to evaluate the probability of hand state according current observation. Applying the observation model into the weight calculation stage of particle filter can effectively increase calculation accuracy. On the other hand, we introduce hand constraints to hand model, utilize these constraints in sampling stage not only can reduce sampling degree of freedom(DOF) but also can revise the incorrect samples.
关键词：Hand Tracking;Edge Detection;Distance Transform;Skin Color Segmentation;Observation Model;Particle Filter
