Abstract
In this paper﹐we present a novel scheme for segmenting the primitive motions of the motion capture data﹒Our scheme first extracts the features from the motion capture data to represent the motion postures﹒Based on extracted features﹐our scheme then uses principle component analysis (PCA) to reduce the dimensionality of features﹒This low-dimensional feature will be used to describe the motion sequence﹒In our segmentation phase﹐we design a signal peak detection method to extract the segmentation positions of the primitive motions﹒We demonstrate our method with two applications﹒The first retrievals the similar motion clips from all primitive motions﹒The second compares primitive motions extracted by our method with the motion clips segmented by annotation﹒The empirical tasks show that the proposed automatic segmentation method is comparable to the results obtained by human annotation﹒