Active stereo vision-based mobile robot navigation for person tracking

In this paper, we propose a mobile robot architecture for person tracking, consisting of an active stereo vision module (ASVM) and a navigation module (NM). The first tracks the person in stereo images and controls the pan/tilt unit to keep the target in the visual field. Its output, i.e. the 3D position of the person, is fed to the NM, which drives the robot towards the target while avoiding obstacles. As a peculiarity of the system, there is no feedback from the NM or the robot motion controller (RMC) to the ASVM. While this imparts flexibility in combining the ASVM with a wide range of robot platforms, it puts considerable strain on the ASVM.Indeed, besides the changes in the target dynamics, it has to cope with the robot motion during obstacle avoidance. These disturbances are accommodated by generating target location hypotheses in an efficient manner. Robustness against outliers and occlusions is achieved by employing a multi-hypothesis tracking method - the particle filter - based on a color model of the target. Moreover, to deal with illumination changes, the system adaptively updates the color model of the target. The main contributions of this paper lie in (1) devising a stereo, color-based target tracking method using the stereo geometry constraint and (2) integrating it with a robotic agent in a loosely coupled manner.