Towards a Robust People Tracking Framework for Service Robots in Crowded, Dynamic Environments

People tracking is an important prerequisite for socially compliant service robots that operate in human environments. In this paper, we take this challenge to new extremes by attempting to robustly track people in a 360-degree field around the robot in very crowded environments like a busy airport terminal. As a first contribution, we present a novel multimodal people tracking framework that is modular, flexible and fully integrated with ROS. We believe that our framework can be of great benefit to researchers as it covers the entire people tracking pipeline, including powerful visualization and evaluation tools. Secondly, we present a number of simple extensions that can make tracking in crowded scenarios more robust. Finally, we compare our tracking method against a complex multi-hypothesis tracking system. Our results on real and synthetic data suggest that it is not the choice of data association which has the largest impact, but the underlying models that, for instance, control track initiation, deletion, and handling of occlusions. By showing that the simpler data association may be sufficient, we provide reasoning why the available resources on a resource-constrained mobile robot might be better spent on other tasks such as higher-level perception and reasoning.