Dataset and tools for benchmarking multi-sensor multi-people tracking for ground robots

We contribute a novel dataset and software tools to benchmark ROS 2 people tracking modules for ground robots. We consider a scenario where the robot moves, in a socially compliant manner, among people, tracking them accurately using multiple diverse sensors: two RGBD cameras and a planar LiDAR. The sequences recorded in the dataset are varied (in terms, e.g., of people density, people behaviors, presence of obstacles, and robot movements) to cover a wide set of application scenarios; their multi-dimensional description helps users to select appropriate subsets of sequences and to assess in which scenarios trackers manifest criticalities. The software tools provide an automated pipeline to compute tracking-relevant metrics on the dataset.