Mobile robots following and guiding stroke patients during their rehabilitation program are in the focus of our research in rehabilitation robotics. To be able to act autonomously, it is crucial for the robot to extract long and precise movement trajectories of the patients. But already keeping track on one specific person in a crowded dynamic environment is inherently hard, since multi-sensor tracking as well as appearance-based re-identification are challenging tasks in realworld environments. Therefore, we aim for developing a coupled person tracking system that combines user tracking by spatial proximity with appearance-based user recognition. We analyzed all subcomponents of such a system and identified four essential parts, that significantly influence the overall performance. We show, that it is essential, to (1) accurately detect all persons in scene, (2) track people as long as no ambiguities occur, (3) visually re-identify the user otherwise, and (4) reduce the search space for re-identification to just relevant hypotheses using spatial proximity as criterion. In our experiments, we show, that by addressing all these aspects, our system significantly outperforms each approach, that excludes just one of these important parts.
展开▼
