Research On Data Completion Technology Of Road Point Cloud Based On Edge Information

High-precision maps play a crucial role in autonomous driving technology.Highprecision maps with complete road information are essential for autonomous driving navigation.The road data obtained through vehicle-mounted Li DAR(Light Detection and Ranging)sensors is stored and processed in the form of point clouds.In the actual point cloud data acquisition process,due to different parameters of the acquisition equipment and object occlusion and other factors,the measured road point cloud data has the problem of missing data,which affects the integrity of road information.In this paper,the integrity of the road information is improved by studying the complementary road point cloud data.Based on the study of complementary algorithms for point cloud data in recent years,this project addresses the problem of missing scattered and disordered point cloud data in road scenes.Firstly,the scene data of roads are pre-processed for semantic segmentation,sampling and information fusion.Secondly,the missing areas are located and the boundary curves of the roads are extracted based on the geographic element information of the OSM map.Finally,the semantic information of the road scenes and the boundary curves of the roads are used as a priori knowledge to complete the missing areas.The specific research work is as follows.(1)Study of semantic segmentation of road scene point clouds.The Gran LA-Net semantic segmentation network model is proposed to address the lack of semantic information,large data scale and scattered structure of point cloud data.The model mainly includes a road scene segmentation module,a road feature sensing module,a random sampling module and an extended feature aggregation module.Based on the Nodes information of the OSM map,the road segmentation module divides the original large road scene data into road segments for input into the network model,in order to improve the network operation rate and reduce the loss of scene information.The road feature sensing module extracts road edge feature information and elevation feature information to enhance the network’s ability to sense edge and elevation features.The extended feature aggregation module segments the road scene based on point location,edge and elevation features to obtain accurate data semantic information.To verify the performance of the Gran LA-Net model,it is compared with RANSAC,Point Net and Rand LA-Net models for experiments.(2)Study of boundary extraction of a road’s missing regions.In the restoration process of scattered point cloud data,the lack of a priori knowledge of the complementation algorithm and its insensitivity to line and surface data have become bottlenecks in the complementation of road scenes.To address the above problems,this paper proposes an OSM a priori-based road boundary extraction algorithm.The algorithm first locates the missing area of the road scene based on the OSM map,and then extracts the road boundary curve of the missing area based on the edge point information and the centreline information of the OSM map.The algorithm can provide a priori knowledge of road boundary curves for restoration.To verify the performance of the algorithm,it is compared with the software annotated boundary curves for accuracy experiments.(3)Study of data complementation of road point clouds.In order to accurately repair road scenes,this paper proposes a road point cloud data completion algorithm based on edge information.The algorithm is based on the semantic information of the road scene and the curve a priori knowledge of the road boundary,and performs surface reconstruction and redundant data processing on the missing area to obtain the complete road data.To verify the performance of the algorithm,it is compared with the hole-filling method,polygon point filling method,PCN and PF-Net models for experiments.