Skip to Main Content
Single action segmentation is very important for the application of motion capture technique in many domains. In this paper, we propose an unsupervised approach to segment single actions from continuous captured motion sequences. The proposed method works by firstly finding the underlying structure of the input motion sequence. Consequently a point cloud of pairwise posture similarity matrix is constructed from the underlying structure found by PCA dimensionality reduction. By discovering the similar short clips of homogenous motion families, single actions could be segmented from the input continuous motion sequence. Experiments show that our approach works well.