LiDAR-Based Road Extraction for UGV in High Definition Map

Abstract

Extracting road information from the high-density point cloud collected by the vehicle-borne laser scanning system is essential for making High Definition (HD) maps. This paper proposes a three-dimensional (3D) laser point cloudbased pavement extraction method applicable to roads of different structure types. First, the filtering method based on cloth simulation is used to extract all ground point sets from the original point cloud data. The partial surface points in front of the vehicle are selected from the ground point to estimate the normal vector of the road surface model. Then, based on the uniformity of the point cloud density distribution of the road surface, all the road points are finally filtered out from the plane point concentration by calculating the continuity of the onedimensional and two-dimensional point density. After that, according to the characteristics that the road points are positively distributed, the method based on the skewness balance is used to estimate the separable road sign. We show that this method can accurately extract road surface information from point cloud data with an extraction rate of 96.1%.