January 2017
·
9 Reads
·
1 Citation
Journal of Computer-Aided Design & Computer Graphics
A multi-granularity segmentation algorithm based on time-frequency analysis is proposed for human motion data segmentation. Firstly, in the time domain analysis stage, the motion sequence was pre-segmented by sparse reconstruction, and the original motion sequence was segmented into several independent behavior segments and hybrid motion primitives by multi-scale time correlation segmentation algorithm to realize coarse-grained segmentation. Secondly, in the frequency domain analysis phase, the feature extraction of each independent behavior segment was performed at the primary frequency, and then each independent behavior segment was segmented into multiple repetitive period segments by combining the zero velocity crossing point detection and the adaptive K-means algorithm to realize fine-grained segmentation. The experimental results show that the use of pre-segmentation and multi-scale segmentation strategy makes our algorithm obtain higher segmentation accuracy, and also facilitates more accurate recognition and segmentation of the hybrid motion primitive, and the feature extraction at the primary frequency makes the algorithm more robust to noise. © 2017, Beijing China Science Journal Publishing Co. Ltd. All right reserved.