Decentralized multi-target-tracking using a LIDAR sensor network

A LIDAR (LIght Detection And Ranging) sensor network to track walking persons inside a surveillance area is investigated. A small number of sensor nodes with spatially stationary and partially overlapping narrow LIDAR beams is chosen. As a consequence of this network topology, the area of surveillance is not fully covered with LIDAR beams and thus the accurate tracking of persons walking inside the area of surveillance is challenging, especially in a multi-target situation. To tackle this problem, multiple target tracking based on a sophisticated decentralized track-to-track fusion architecture is developed and evaluated in this paper: dynamic multi-hypothesis tracking (MHT) by independent local trackers is carried out in all sensor nodes; then local track favorites are sent to a fusion center where global track candidates are derived and fed back to the local trackers in order to improve the local tracking. With this architecture a track association success rate of (98.8 ± 0.3)% and a mean square position error of Δp = 6.7 cm were derived from 1000 pairs of random and intersecting trajectories of two walking persons (mean velocity 1.5 m/s). Furthermore, the tracking performance as a function of the target velocity was quantified. The performance of the proposed algorithm was also experimentally evaluated using general-purpose laser scanners.