Multiple Pedestrian Tracking Using LiDAR Network in Complex Indoor Scenarios

A LiDAR network is exploited for multiple human target tracking in a complex indoor scenario. LiDAR is scattered across the surveillance area, which includes several regions. The human targets maneuver rapidly and may come close to each other. Meanwhile, lots of static objects in the scenario generate a mass of static clutter measurements. In this work, a track-before-detect (TBD) processing framework is developed for human trajectories. Initially, individual and overall clutter maps are constructed to eliminate static clutter. Then, 3-D region growth, 3-D projection (3DP), and tracklet association are applied to address the problem of multiple maneuvering targets and extended targets. In the actual experiment, five pedestrians maneuver through an area comprising two corridors and a lobby. All human trajectories can be well detected by a network containing three LiDARs without false track or missed detection even though the scenario includes closely parallel tracks and crossed tracks. The average position error of human tracks could be as low as 24 cm.