Recognition and synthesis of human actions from video