Obstacle detection and tracking in an urban environment Using 3D LiDAR and a Mobileye 560

Abstract

In order to navigate in an urban environment, a vehicle must be able to reliably detect and track dynamic obstacles such as vehicles, pedestrians, bicycles, and motorcycles. This paper presents a sensor fusion algorithm which combines tracking information from a Mobileye 560 and a Velodyne HDL-64E. The Velodyne tracking module first extracts obstacles by removing the ground plane points and then segmenting the remaining points using Euclidean Cluster Extraction. The Velodyne tracking module then uses the Kuhn-Munkres algorithm to associate Velodyne obstacles of the same type between time steps. The sensor fusion module associates and tracks obstacles from both the Velodyne and Mobileye tracking modules. It is able to reliably associate the same Velodyne and Mobileye obstacle between frames, although the Velodyne tracking module only provides robust tracking in simple scenes such as bridges.