Captured Human Motion Segmentation Based on Dynamics and Principal Component Analysis

This paper proposes an approach for segmenting single actions from continuously captured motion sequences by examining the properties of active limbs. The target motions are related to sporting and dancing. In particular, two types of human sports motions are examined: 1) boxing and 2) hip hop dance. To segment continuous boxing motion sequences into single punches and combo punches, this paper employs the professional knowledge of boxing to compensate for the drawbacks of local minima or maxima detection based motion segmentation approaches. To segment continuous Hip-hop dance sequences into single actions, this papers explores a PCA (principal component analysis) based approach by finding the proper start frame and segment length to perform PCA transformation. Experiments proves that the proposed approach works well.