A Novel Method for Human Motion Capture Data Segmentation

The segmentation of human motion capture data is a crucial step in data analysis, which serves as a good basis for data management and reuse. In this paper, we research about the equivalence relation between normalized cut model and weighted kernel k-means, and apply it to the behavior segmentation of human motion capture data. The frames of the motion sequences are regarded as high-dimensional independent points, which can be clustered by the combination of normalized cut model and weighted kernel k-means. The clustering results after the time sequence recovery are constructed a category string, and we use the suffix array to find out the valid substrings. After that, long characters, invalid substrings, and the segmentation points can also be found ultimately. This method can not only solve the NP-hard problem of the graph cut model, but also solve the problem of selecting the kernel matrix of the weighted kernel k-means. The experimental results show that the method based on normalized cut model combined with weighted kernel k-means (NCWKK) has a satisfactory segmentation performance.