Robot vision to recognize both face and object for human-robot ball playing

Abstract

This paper presents a robot vision system to recognize both face and object for human-robot interaction. Social demands and expectations for human interactive robots have been increasing in recent years. When such robot interacts with human through some object, for example a ball, it must be convenient and effective for the robot to recognize human face and the target object. But no really working robot has been reported so far. Thus the authors develop the robot vision as the first step to realize such working robot. We take ball playing as an example of human-robot interaction with object. The developed vision system has advantage of interactive trainability. And it is so robust that it can detect and track the face and ball even though the background image is complex and cluttered. Experimental results support feasibility and effectiveness of the system.