Tracking a Varying Number of People with a Visually-Controlled Robotic Head
Y.Ban, X. Alameda-Pineda, F. Badeig, S. Ba, and R. Horaud
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS’17)
Multi-person tracking (MOT) using a robot platform is of crucial importance in human-robot interaction. In addition to the tracking problems, such as occlusions, changes in appearance, and a varying number of people, there are robot hardware constraints and limitations. In this paper, we propose a novel method which simultaneously tracks a varying number of persons and performs visual servoing. The complementary nature of tracking and of servoing enables the following features: (i) the ability of tracking multiple objects while compensating for large ego-movements and (ii) visual-control of the robot to minimize the effect that the person-of-interest disappears from the field of view. The proposed Bayesian variational formulation allows us to efficiently solve the probabilistic inference problem, by providing a closed-form solution, thus maintaining a reasonably low computational cost (the overall system works at 10 FPS). The experiments using the NAO-MPVS dataset report a significant performance increase of the proposed tracking/servoing method with respect to tracking-only methods.
Results on the NAO-MPVS Dataset (More results available on this page)
Funding from the European Union FP7 ERC Advanced Grant VHIA (#340113) is greatly acknowledged.