Recent works on human action recognition have focused on representing and classifying articulated body motion. These methods require a detailed knowledge of the action composition both in the spatial and temporal domains, which is a difficult task, most