Study of recognizing human motion observed from an arbitrary viewpoint based on decomposition of a tensor containing multiple view motions.