Dance motion analysis by correlation matrix between pose sequences

摘要

Since a dance motion is represented as a temporal sequence of poses, comparison of motion data is reduced to the comparison of individual poses. In the present paper, the pose comparison problem is basically solved by 2D correlation assuming that human motion is bound by gravity. Therefore, dance motion analysis is discussed based on correlation matrices that are calculated between two motion sequences captured by a motion capture system. A correlation matrix, in which each component represents a pose similarity between two poses, is defined between two motion sequences. A symmetric correlation matrix is also defined from symmetric similarities between two motion sequences. These correlation matrices are used for extraction of motion sequences that are similar or symmetric to a given dance motion, respectively. They are also used for motion difference analysis between two dancers performing a particular dance. In motion difference analysis, the correlation matrix is also considered between a motion sequence and a motion eigenspace. Some subtle motion errors can be extracted with the correlation matrix when a motion eigenspace is constructed over standard motion data.