A road surface filtering method based on plane normal adaptive constraints

Abstract

Aiming at the problem that the road surface point cloud is difficult to filter due to the complex road conditions and road types, this study proposes a road surface filtering method with plane normal adaptive constraints. After the grid of point cloud data, extracting the road surface seed points by the approximate plane constraint method and establishing the eight-neighborhood grid. We use the improved random sample consensus plane fitting to extract the geometric features of the road surface and carry out the plane normal statistical analysis of the road surface grid. The normal distribution Quantile-Quantile plot and Chi-square test show that the normal vector distribution of the road surface grid conforms to the normal distribution. Establishing the plane normal adaptive parameters using the normal distribution probability-weighted mean and the normal distribution standard deviation, and the plane feature part is automatically constrained. After pretreatment by elevation constraint and filter point re-determination, the road surface area is grown by cluster analysis to obtain the ground point accurately cloud. Through experiments and analysis covering most types of road sections, such as expressway and urban roads, straight and curved roads, gentle roads, and undulating roads, the results show that the road surface point filtering precision, recall rate, and overall accuracy of this method are all above 97 %, which verifies the accuracy and reliability of the method.