Classification of Human Actions using Sensor Gloves and Hidden Markov Models

Abstract

An important part of programming by demonstration is the recognition of a human demonstrator's hand motions. Generally, the demonstrator's movement is segmented and a classification into actions relevant for manufacturing tasks is performed. In this paper, we describe the use of hidden markov models to obtain this action sequence also for everyday manipulation tasks, without prior segmentation of the input. As recognition features, we use information about the hand position, finger joint angles and hand-object contact points.