Safety Assessment of Lane Marking for Autonomous Vehicles Using Light Detection and Ranging (lidar) Data

Autonomous vehicle (AV) experts have identified lane markings as the most important road infrastructure element to consider for AV compatibility with existing roadways. To ensure a safe and comfortable experience for riders, AVs must have ample preview distance of lane markings so that appropriate maneuvers can be calculated in time. This paper provides a method for assessing lane marking occlusion along a roadway provided a point cloud representation collected by light detection and ranging (LiDAR) scanners. An octree representation of the point cloud and a set of target planes on the pavement surface allows for querying any potential obstructions between an AV’s sensor field of view and the road’s surface. Using this algorithm, three road segments with a vertical crest followed by a sharp horizontal curve were identified to pose challenges for AVs, owing to their low lane marking visibility. Potential AV trajectories were modelled using Bézier curves at a variety of speeds and reaction times. Radius and centripetal acceleration were extracted at each point along the trajectory and compared against minimum thresholds for recommended radius and comfort levels. The goal of this work is to create a novel methodology to test and improve lane marking visibility in preparation for the mass deployment of AVs, as well as to promote data-driven decisions for smart infrastructure upgrades.