Automatic filtering of vehicles from mobile LiDAR datasets

Abstract This manuscript presents a new filter to automatically remove vehicles from point clouds of roads captured by a mobile LiDAR system. The algorithm works in two main steps. The first one consists of the normalization and segmentation of the point cloud in order to homogenize the density of points in all areas corrected as function of vehicle speed. The second step is based on detecting changes of slope on the transversal profiles of the roads digitalized by LiDAR, inside the contours defined by the traffic lines that limit the road space. Profiles are produced by the rotation of the LiDAR mirror. Once detected those points of the LiDAR echoes that correspond to the vehicle they are automatically removed. Finally, the point cloud is regenerated using commercial software that creates new points based on parametric surfaces, fitting the points of the neighborhood of each hole produced by the elimination of vehicle points. The algorithms developed are tested by checking the successful of the vehicle detection and the geometric comparison of longitudinal profiles and cross sections, before and after the filter application. Results show as all the vehicles were successfully detected and corresponding echoes cleaned. Road profile and cross sections reveal that the generated geometries are in agreement with the real geometries of the road.