Human Motion Behavior Segmentation based on Local Outlier Factor

Motion segmentation, which is a crucial technology in reuse of motion capture data, means automatically dividing a long motion sequence into several motion clips which have different semantics. In this paper, a new motion segmentation method based on Local Outlier Factor (LOF) is proposed. Our method is based on the assumption that motions with same type can form a cluster and the transition of two behaviors is the outlier of the two clusters. We use LOF to find a “bridge” between different semantic motion clips. The presented method mainly consists of four stages as follows: firstly, bone angle is employed to represent the motion characteristic in preprocessing stage; secondly, sliding window is used as the statistical unit and statistic histogram is built for all bone angles in sliding window; thirdly, LOF is calculated for every sliding window; finally the segmentation point can be obtained by locating the peak of LOF curve. The proposed method is evaluated on the Carnegie Mellon Motion Capture database. The experimental results show that the method can achieve the automatic segmentation for human motion capture data, and has better performance compared with the classical PPCA based segmentation method. Additionally, the proposed method shows high efficiency even dealing with long motion sequences.