High Definition LiDAR mapping of Perth CBD

High Definition (HD) 3D maps are gaining popularity in automation. For the emerging industry of autonomous driving, point cloud-based HD mapping is particularly interesting due to its robustness to weather conditions in data collection. It can help driverless and robotic vehicles to precisely localize and navigate in large complex environments. We present a scalable technique to effectively construct an HD 3D map of an urban environment. We demonstrate the application of our approach by constructing a 3D map of the Central Business District (CBD) of Perth, Australia. Our mapping covers a ground floor area of approximately 4 km
²
. We use Ouster LiDAR sensor for data collection. On one hand, ours is a unique approach to construct large-scale HD maps in urban settings. On the other, we provide the first-ever instance of HD point cloud map of Perth CBD, which is in-line with the recent developments of autonomous driving in Western Australia. To construct the map, we extract closed spatial loops from the frames, which are aligned with a 3D normal-distribution transform, and meticulously fused to construct a single closed loop HD sub-map. The sub-maps are combined to construct the full large-scale map of the region. We adopt closed loop formation to suppress error propagation in our mapping. We provide binary annotation to split the map into roads and other objects. Finally, we also demonstrate an additional utility of our mapping technique in self-localization of vehicles using Eigen features in the map.