Visual system for tracking and interpreting selected human actions

Abstract

This paper describes an autonomous vision system for realization of tasks consist of following a person with a mobile robot as well as interpreting some static and dynamic commands signaled by hand. Detection of the person is realized on the basis of color image segmentation combined with stereovision analysis. The elaborated algorithms of face detection and localization improves quality of tracking as well as makes possible to recognize some nonverbal commands using geometrical relations of face and hands and in particular to recognize the pointing arm-posture.