Playing hide and seek with a mobile companion robot

This paper addresses the problem of finding people in domestic environments utilizing a mobile robot. Companion robots, which should provide different services, must be able to robustly estimate the user´s position. While detecting people in an upright pose is mainly solved, most of the users´ various poses in living environments are hard to detect. We present a tracking framework that incorporates state-of-the-art detection modules, but also a novel approach for visually detecting the presence of people resting at previously known seating places in arbitrary poses. The method is based on a contextual color model of the respective place in the environment and a color model of the user´s appearance. The system has been tested by evaluating the robot´s capability to find the user in a 3-room apartment in a hide and seek scenario.