Automatic Modeling And Safety Compliance Analysis Of 3D Road Alignment Information Based On Point Clouds

Three-dimensional road alignment elements(3DRAEs)are the indispensable basic parameters to achieve "All-elements and full-cycle digitalization of transportation infrastructure ".The Light Detection and Ranging(Li DAR)represented by mobile laser scanning and terrestrial laser scanning,provides support for obtaining spatial information data such as point clouds with centimeter accuracy and centimeter density.“What you see is what you get” reproduces the spatial information of road infrastructure environment.However,with the characteristics of huge data volumes,lack of disorder and obvious differences carried by various instruments and scenarios,the classification results of point clouds are limited when classical deep learning networks are directly applied in road environment.Secondly,when segmenting the road surface point clouds into strips,the variety of road alignments might lead to the loss of road surface point clouds,and there are no corresponding relationships between discrete point clouds and road alignment parameters.Furthermore,the spatial data measured by traditional surveys is relatively sparse.Then the corresponding road safety analysis tends to be qualitative and lacks detailed analysis of any road section.Therefore,the main purpose of the thesis is to propose the end-to-end 3DRAEs(road boundary,centerline,cross slope and longitudinal slope)estimation model driven by massive point clouds and then perform detailed road safety analysis.The main works and contributions of this thesis are as follows:(1)Considering the point clouds of different road environments collected by different Li DAR devices,Octree module and Normal Vector module are added into Point Net++ network architecture.Octree-Point-Normal-Vector-Net++(OPNV)deep learning network is proposed for automatic and high-precision classification of different types of road infrastructure.Road point clouds can be extracted with the proposed network and the main infrastructure categories are distinguished automatically and accurately.(2)The concave envelope algorithm was constructed to identify the road point clouds boundary under the change of road alignment.B-spline algorithm was used to fit the road centerline and recover the horizontal alignment elements(straight line,curve,radius of curvature,et.al).The road surface point clouds were automatically divided into continuous strips with meter-level intervals by the quadrilateral principle.The total least squares model was built to calculate the cross and longitudinal slopes with adjustment solution.With the above operation,the 3DRAEs estimation model of the whole road was established automatically.(3)The road safety analysis is carried out from three dimensions of horizontal alignment,longitudinal alignment and cross alignment.The multi-index road alignment safety analysis is conducted.The early warning value and recommended speed value are given for safety operation of vehicles.Four data sets(data set 1-highway;data set 2-campus roads;data set 3-urban roads;data set 4-mountain roads)collected by mobile laser scanning and terrestrial laser scanning were applied to check the validation of the proposed method.The results demonstrate that the proposed model in this thesis is capable of achieving full-section and continuity analyses of roads,as well as providing early warnings and recommended speed values.The contributions of this work can be applied to road reconstruction projects and safety compliance assessments of in-service roads.