People perception from RGB-D cameras for mobile robots

Understanding how humans move through the scene is a key issue of decision-making for an autonomous mobile robot in crown people zones. So accurately detecting and tracking people from a mobile platform can help improve interaction effective and efficient. In this paper, we proposed a people detection and tracking system using combination of a several new techniques for mobile robots, plan-view maps, depth weighted histograms, and GNN data association. We proposed a spatial region of interest based plan-view maps to detect human candidates. Firstly, point cloud sub-clusters were segmented for candidate detection. Two different plan-view maps, named occupancy map and height map, were employed to identify human candidates from point cloud sub-clusters. Meanwhile, a depth weighted histogram was extracted to feature a human candidate. Then, a particle filter algorithm was adopted to track human's motion. Finally, data association was set up to re-identify humans which were tracked. Extensive experiments demonstrated the effectiveness and robustness of our human detection and tracking system.