Spatiotemporal projection of motion field sequence for generating feature vectors in gesture perception

An algorithm of generating a single feature vector representation from a sequence of motion fields has been developed for gesture perception. Four directional motion components in a motion field are projected onto x- and y-axes perpendicular to respective directions and four histograms are generated. Then the time evolution of such histograms are integrated by spatiotemporal projection and summarized in the form of a single vector that represents the motion history of the image in a scene. Gesture perception is carried out by a simple template matching using such motion-field sequence vector. In order to enhance the performance of gesture perception, three techniques have been introduced in template matching: normalization, smoothing and shifting. Preliminary gesture perception experiments using the vector representation developed in the present work have been carried out and the effectiveness of the algorithm has been demonstrated.