Abstract: To address the challenges of road surface extraction from large-scale vehicle LiDAR point clouds, including difficulties caused by obstacles occluding the road surface and variations in road topography leading to inaccuracies and noise in the extraction results, this paper introduces a vehicle LiDAR point cloud road surface extraction method based on radial gradients. The method involves initial data preprocessing, including pass-through filtering and voxel downsampling. Subsequently, the point cloud data is transformed into polar coordinates,which is projected onto a fan-shaped grid based on the hardware parameters of the vehicle's LiDAR system. Using the fan-shaped grid as a foundation, a radial gradient road surface point extraction is performed within a moving window, and a plane fitting technique is employed, optimized using the least-squares method to refine the extraction results. The KITTI dataset is chosen for road surface point extraction experiments. Results indicate that compared to other road surface extraction methods this approach exhibits robustness and high accuracy, with an average accuracy of 91.85%, an average completeness of 80.63%, and an average precision of 75.25% in road surface point extraction.

Key words: vehicle LiDAR point cloud, point cloud filtering, fan grid, radial gradient, road surface points