Visually Gesture Recognition for an Interactive Robot Grasping Application

Abstract

Gesture based natural human-robot interaction paradigm has few physical requirements, and thus can be deployed in many restrictive and challenging environments. In this paper, we propose a robot vision based approach to recognizing intentional arm-pointing gestures of human for an object grasping application. To overcome the limitation of robot onboard vision quality and background cluttering in natural indoor environment, a multi-cue human detection method is proposed. Human body is detected and verified by merging appearance and color features with robust head-shoulder based shape matching for reducing the false detection rate. Then intentional dynamic arm-pointing gestures of a person are identified using Dynamic Time Warping (DTW) technique, whilst unconscious motions of arm and head are rejected. Implementation of a gesture-guided robot grasping task in an indoor environment is given to demonstrate this approach, in which a fast and reliable recognition of pointing gesture recognition is achieved.