Segmentation of Human Motion Sequence Based on Motion Primitives

Abstract

Motion capture data could describe human motion precisely. It’s anticipated some uncapturable human motions may be generated through easily capturable motions. A method was introduced to segment motion captured data, which could cut a long motion sequence into some unique motion primitives. This method involved 14 joints of body hierarchy. Singular Value Decomposition (SVM) was used to determine how motion data dimension changed, which could identify the segmentation frame. The validity of method was verified with an 8,401 frames motion sequence from Carnegie Mellon University Motion Capture Database, and it turned out to be valid in accordance with human intuitive judgment.