Extraction Of Road Elements In High-Definition Maps Based On Vehicle LiDAR Data

Autonomous vehicles technology is one of the most popular technologies today.The requirements of unmanned driving for maps are not only to assist road planning,but also to coordinate maps with motion planning,positioning,prediction,and perception modules.Ordinary digital maps can no longer meet the requirements,requiring higher accuracy and more accurate positioning High-Definition map.This paper proposes a complete set of High-Definition map road feature extraction and segmentation processes.From the vehicle Li DAR data,the road features such as road surface,road marking lines,and rods on both sides of the road are extracted from the original point cloud.And give it specific semantic information and store it as the corresponding layer.Firstly,the ground points were extracted by point cloud filtering,and the scanning lines were separated according to the timestamps to restore the local spatial structure of the point cloud to a certain extent.At the same time,the calculation is reduced from three-dimensional to one-dimensional,which speeds up the operation of the algorithm.Then select the initial point based on the trajectory data,and use the strategy of early stopping to avoid confusion of other similar boundaries and improve the robustness of the algorithm.Then the density-based spatial clustering method(DBSCAN)is used to separate the noise and constraint points,and the noise points are optimized and constrained.This article uses the IQmulus & Terra Mobilita dataset for experiments.The F1 score reaches 98.89%,which performs well in complex road environments.Then based on the extraction of the road surface,segmentation threshold were used to correct the reflection intensity change of the laser point due to the incident angle.Road marking lines are all painted with bright colors.The maximum inter-class variance method(OTSU)is used to extract road marking laser points whose reflection intensity is significantly higher than that of road surfaces.Then use DBSCAN to cluster and separate the laser points,remove the noise points,and separate the different road marking lines to give it semantic information.Finally,we use the alpha-shape algorithm to extract the outline of road marking lines to simplify the storage structure.Finally,poles with distinctive features such as street lights,telephone poles,and street signs are extracted from non-ground points.According to the height of the laser point from the ground,use the elevation filter to remove the interference ofnon-rod-like parts such as tall trees,street lights and street signs,and low bushes at low places.The laser points were clustered by DBSCAN,and the minimum circumscribed circle radius was obtained for each cluster point,and the rod part was separated according to the minimum circumscribed circle radius.An algorithm combining DBSCAN and seed point growth ideas was used to extract the non-rod-like parts of these rods.At the same time,the voxelization method was used to further optimize the algorithm's operating efficiency.The results show that the High-Definition map road feature extraction and segmentation process proposed in this paper can extract various road features in complex road environments,and has high accuracy and robustness.