LiDAR-based Free Space Detection Method for Autonomous Vehicles

As a crucial environmental perception technology of autonomous vehicles, free space detection has gained more attention that detects road obstacles, such as curbs, vehicles and pedestrians generally. Besides, real-world road scenes also encompass potholes and water hazards, which also affect the adaptability of autonomous vehicles. Focused on the extended objects of road obstacles, this paper presents a LiDAR-based approach for free space detection to enhance the adaptability of autonomous vehicles. Firstly, ground and non-ground point clouds are segmented through grid height difference filtering and plane fitting. Then, boundary features of obstacles such as vehicles and pedestrians are extracted from non-ground point clouds based on height differences. On the other hand, boundary features of water hazards are extracted from ground point clouds based on horizontal distance, and those of curbs and potholes are extracted based on slope deviation. Finally, the extracted boundary feature points are merged, to obtain the free space by the ray-casting method. Experimental results based on the KITTI-ROAD dataset and real-world scenarios demonstrate the effectiveness of proposed method.