3D Human Motion Tracking Using Integrated Particle Filter and Mean Shift

• Main Authors: Shih-Yao Lin, 林士堯
• Other Authors: I-Cheng Chang
• Format: Others
• Language: en_US
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/bb5w6q

碩士 === 國立東華大學 === 資訊工程學系 === 96 === This study proposes a novel model-based algorithm, dynamic kernel based progressive particle filter, for 3D unconstraint human body tracking. The articulated full body motion tracking is the large number of degrees of freedom to be estimated. The aim of this study is to reduce the computational cost and improve the accuracy. The progressive particle filter consists of three principal approaches: the multiple predictions technique of particle filter, the iterative mode seeking algorithm of mean shift and hierarchical searching approach. The hierarchical searching approach decomposes the high dimensional space into three subspaces, global motion layer, local inside motion layer and local outside motion layer, for decreasing the searching range. Natural human motion is usually non-linear and non-Gaussian, the system applies the random sampling technique of particle filter to predict the unconstraint motion of human body. The mean shift trackers are embedded into each particle to improve the accuracy by iterative mode seeking. Moreover, the study proposes a dynamic kernel model, which can automatically adjust the kernel bandwidth of mean shift trackers according to the probability of each state by improving the searching effectively. This study combines the progressive particle filter and dynamic kernel model for reducing the computational cost and improving the accuracy simultaneously.