Efficient Unsupervised Behavioral Segmentation of Human Motion Capture Data

With the development of human motion capture, realistic human motion capture data has been widely implemented to many fields. However, segmenting motion capture data sequences manually into distinct behavior is time-consuming and laborious. In this paper, we introduce an efficient unsupervised method based on graph partition for automatically segmenting motion capture data. For the N-Frame motion capture data sequence, we construct an undirected, weighted graph G=G(V,E), where the node set V represent frames of motion sequence and the weight of the edge set E describes similarity between frames. In this way, behavioral segmentation problem on motion capture data may be transformed into graph cut problem. However, the traditional graph cut problem is NP hard. By analyzing the relationship between graph cut and spectral clustering, we apply spectral clustering to the NP hard problem of graph cut. In this paper, two methods of spectral clustering, tnearest neighbors and the Nystrom method, are employed to cluster motion capture data for getting behavioral segmentation. In addition, we define an energy function to refine the results of behavioral segmentation. Extensive experiments are conducted on the dataset of multi-behavior motion capture data from CMU database. The experimental results prove that our novel method is robust and effective.