Color vision based person following with a mobile robot

Abstract

Describes a vision based interface that enables the mobile robot to detect and follow a person through office environments. The proposed algorithm works by estimating the separate distributions of colors on a person's face and shirt. Using heuristics on vertical alignment of face and shirt color as well as bordering on each other, we are able to extract the person reliably and to guide the robot to desirable places. A fast and adaptive scheme of color processing has been tested in changing lighting conditions and using very long image sequences. The elaborated algorithm utilizes a linear Kalman filter to perform smooth tracking. The algorithms we have developed have been implemented and tested on a mobile robot equipped with an active camera. The developed system is able to track a person at approximately 10 frames per second.