Benchmarking Multi-Sensor Multi-People Tracking for Ground Robots: Dataset and Tools

We contribute a novel dataset and software toolsto benchmark people tracking modules designed for groundrobots. We consider a scenario where a robot moves, in asocially compliant manner, among people, tracking them accu-rately using multiple diverse sensors. The sequences recordedin the dataset are varied (e.g., in terms of people density, peoplebehaviors, presence of obstacles, and robot movements) to covera wide set of application scenarios. While we do not focuson benchmarking specific tracking algorithms, we include anexample with two people trackers, one using RGBD imagesand the other laser scans, providing insight into their relativeperformance.