Human motion segmentation based on structure constraint matrix factorization

The human motion recognition based on the segmented datasets is a hot multidisciplinary research topic in the field of computer vision. However, in reality, the collected data is without segmented. Therefore, the motion segmentation is crucial. Currently, common methods include principal component analysis (PCA), cluster algorithm and deep learning. However, the difficulty of human motion segmentation is summarized as follows: (1) The reasonable and easy-tocalculate similarity measurement criteria are designed for sequential motion data points. (2) In general, motion sequential data are a high-dimensional sequential data, which needs to be preprocessed by similarity measurement, data alignment as well as data dimensionality reduction.