Person following and mobile robot via arm-posture driving using color and stereovision

Abstract

This paper describes an approach to arm-posture recognizing and interpreting by mobile robot while following a person. We combine color and stereo cues to detect and track a person in dynamic indoor environments in real-time. The arm-postures are very simple and can be described by a spatial relation between a person's face and hands. Color image processing and labeling techniques are used to detect candidates of face, hands and shirt. The localization and recognition take place by analyzing the labeled images and utilizing a range histogram. Using heuristics for the size and spatial relations we can locate the tracked person as well as the person's face and hands. Experiments conducted with a Pioneer 2 DX mobile robot equipped with an active color camera and a stereovision system have shown the presented approach to be repeatable and robust to noise.