An Observation Likelihood Model Integrating Edge Feature with Silhouette Feature

Abstract

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.&nbsp

Research on Gesture Speed Estimation Model in 3D Interactive Interface

“…Gesture model is very important in the field of human-computer interaction, it is widely used and it also has more types. According to the skeletal structure of the hand, we describe the hand gesture in 26 Degree of Freedom (DOF), including 6 global freedoms and 20 partial freedoms [7]. The joint angles connecting palm and finger have two DOFs with flexision/extension and abduction/adduction motions, other joint angles only have one DOF with flexision/extension motion.…”