关键词: 人体运动捕获, 运动分割, 内蕴均值, 主测地线分析（PGA）, 概率主测地线分析（PPGA）

Abstract: Due to the inherent nonlinearity of human motion capture data, linear methods could not always find intrinsic dimensionality of motion data, In view of this, automatic segmentation approaches using Principle Geodesic Analysis（PGA） and using Probabilistic Principle Geodesic Analysis（PPGA） are proposed in this paper. Both of these approaches regarded the motion as an ordered poses and segment the motion where there is a local change in the distribution of poses. The segmentation approach using PGA assigned a cut when the intrinsic dimensionality of a local model of the motion suddenly increased. The segmentation approach using PPGA placed a cut when the distribution of poses is observed to change. The experimental results show that they can segment automatically, and provide the good performance.

Key words: human motion capture, motion segmentation, intrinsic mean, Principle Geodesic Analysis（PGA）, Probabilistic Principle Geodesic Analysis（PPGA）