基于关节联动特征的运动捕获数据分割方法

Motion Capture Data Segmentation Based on Joint Linkage Features

运动捕获数据分割中,针对关节角度或关节点中心距离作为描述人体姿态的特征所存在的局限性,提出一种关节联动特征。该特征从人体运动学角度对人体关节之间的联动关系进行建模,将关节角度与关节点中心距离作为一组关节联动特征点转换到极坐标空间表示。通过计算任意时刻的人体姿态与初始时刻人体姿态的关节联动特征点的偏移量来描述人体姿态的变化。提取每一帧捕获数据的关节联动特征,并使用主成分分析方法对运动序列进行分割。对CMU人体运动捕获数据库的实验结果表明,使用该特征得到的运动捕获数据分割结果要优于使用关节角度或中心距离特征的分割结果。

In order to improve the performance of motion capture data segmentation, the joint linkage feature was proposed to overcome the limitation of the feature of joint angle and joint point center distance. The joint linkage feature modeled the linkage relationship between the joint angles from kinesiology, and conversion the joint linkage feature which consist of joint angle and joint point center distance to polar coordinate space. The transformation of human body posture was represented by feature point offset of initial moment and other times. Motion sequences were segmented by PCA method finally. The experimental results show that the proposed feature can divide motion sequence more efficiently than other geometric feature on CMU human motion database.