The Robust Likelihood Model of State Measurement and Its Applications in Articulated Object Tracking

The establishment of the likelihood model of state observation with a strong robustness is one of the core issues in the study of moving hand tracking. This paper is dedicated to building a robust likelihood model of state observation, and do some study by using the method of gaining feature points from frame images of human hand. Firstly, based on physiological models and camera projection principle, we propose a basic idea that use gesture polygon to describe the image contour of hand gesture. Secondly, Lindeberg method is improved by designing the two types of response function to get the different types of feature points on multiscale space basen on the local area of vertex in the polygon, and a novel structural response mode is presented. Then we fuse the features in different scales by using Hausdorff distance and Hausdorff matrix, and present the likelihood model of state observation. Finally, the model is used for 3D motion tracking of human hand. Our theoretical analysis and experimental results show that the approach put forward in this paper has the advantages of a low time complexity and strong robustness, compared with the Lindeberg method.