基于几何特征的人体运动捕获数据分割方法

Human Motion Capture Data Segmentation Based on Geometric Features

运动捕获在虚拟仿真中被广泛使用,为了实现捕获数据的自动分割,提出了一种基于几何特征的分割方法,将人体运动数据分割为具有特定语义的运动片断,以及连接语义片断的过渡片断。首先选择一组几何特征,然后通过PCA法构造出用于分割的综合特征函数和对应的评价函数;最后对特征函数进行归一化、低通滤波和幅度跳变检测,检测出所有的语义片断,从而实现对运动数据的分割。实验表明,该方法具有较好的分割效果,有助于提高运动捕获的效率。

Motion capture is frequently used in virtual simulation. In order to segment the motion data into small clips automatically, an approach was proposed based on geometric features. The data can be automatically segmented into stability clips which describe the behaviors and transition clips which connect two stability clips. The approach was divided in three main stages. In the first stage, a set of geometric features was selected. In the second stage, PCA method was used to construct a final feature function and an evaluation function. In the last stage, stability clips and transition clips from the feature function were generated by normalization, low-pass filter and detecting the scope transition. Experimental results show that this approach provides the good performance.