Structured Road And Marking Extraction Based On Vehicle Laser Point Cloud Data

The vehicle-borne laser scanning system can quickly acquire the three-dimensional coordinates and reflection intensity of the target surface,and has the characteristics of realtime,high-precision,high-density,non-contact and automation,which provides a new way for the rapid acquisition and real-time updating of road information.However,the original point cloud data acquired by the vehicle-borne laser scanning system is large,disorderly,unevenly distributed,and contains a variety of targets.At the same time,the road environment scene is complex and the shape is varied.The road boundary and the traffic markings on the road surface are not prominent enough in the original point cloud,and are easily occluded,resulting in the absence of point cloud in the local area of the road.These factors have caused great difficulties in the automatic extraction of road information.Therefore,how to extract the characteristic information such as roads and markings from the vehicle-borne laser point cloud in a large and complex environment accurately,efficiently and completely has important research value in the present.The main research works of this paper are as follows:(1)For the problem that the laser spot cloud is disorderly and has noise points and redundant data,the common preprocessing methods of vehicle-borne laser point cloud data are summarized and discussed,including point cloud space index establishment and original point cloud denoising.And point cloud streamlining.(2)Combined with the distribution characteristics of point cloud law vector of structured roads,a method of road point cloud extraction based on normal vector similarity is proposed.Firstly,the original point cloud is filtered by non-ground point using the cloth simulation algorithm.Then the principal vector and curvature values of each laser point are calculated based on principal component analysis and surface fitting method respectively.Finally,based on the similarity of point cloud normal vector as a constraint,the improved region growing segmentation algorithm is used to extract the road point cloud.(3)The method of road marking point cloud extraction based on intensity feature image is studied and improved.Firstly,the road point cloud projection is transformed into a point cloud intensity feature image,and the road marking position is obtained based on the edge detection and connectivity analysis method.Then,the road marking point cloud is initially obtained from the road surface point cloud through the inverse projection transformation,and finally,the Otsu method is recommend into the point cloud intensity filtering to eliminate the road noise points and achieve accurate extraction of the road markings.(4)Designed and developed a software that includes functions such as point cloud display,point cloud preprocessing,road information extraction,etc,and select multiple sets of vehicle-borne point cloud data in different environments to verify the road and marking method proposed in this paper.The experimental results were analyzed and evaluated.