Indexed by:会议论文

Date of Publication:2017-01-01

Included Journals:EI、CPCI-S

Volume:773

Page Number:627-638

Key Words:Human motion sequence segmentation; Motion change characteristics; State variation; Hashing

Abstract:Motion sequence segmentation is a fundamental work in human motion analysis, which can promote the deep understanding as well as wide application of human motion sequences. Mainstream human motion sequence segmentation methods only focus on the data characteristics of the sequence and neglect the physical characteristics of the motions. This paper proposes a hashing based state variation segmentation (HBSVS) method to realize human motion sequence segmentation by analyzing the changing characteristics of the motions on time series. To improve the computational efficiency and merge motions from the same class, HBSVS adopts hashing method to construct the state descriptor of each motion on the sequence. Experiments on CMU motion capture database and UT-Interaction standard dataset show that HBSVS outperforms several state-of-the-art human motion sequence segmentation methods.