点云驱动的城市道路停车视距检验及安全分析  

Inspection and Safety Analysis of Stopping Sight Distance on Urban RoadsfromLiDARPointCloud

作　　者：李浩 张子宜 翟鸿漾 付迪 王金 Li Hao;Zhang Ziyi;Zhai Hongyang;Fu Di;Wang Jin(Beijing Key Laboratory of Traffic Engineering,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学交通工程北京市重点实验室,北京100124

出　　处：《市政技术》2025年第2期10-18,34,共10页Journal of Municipal Technology

基　　金：国家自然科学基金项目(41801380);北京市自然科学基金(8232005,L221026)。

摘　　要：利用高密度激光雷达点云数据,针对城市双车道道路场景,采用PointNet++实现道路场景的自动分类,创建了纳入路侧环境的椭球体视域加宽模型,并引入内部形状描述子算法高效提取视域内关键地物特征点,继而搭建停车视距自动化检验模型,开展不同行车速度和路面条件下的三维停车视距检验,最后提出全路段停车视距满足率和单个视点停车视距符合率2个指标,以评价道路停车视距安全。通过城市双车道实例检验,结果表明,道路场景自动分类后的平均交并比、准确率、精确率和召回率分别为81.9%、91.7%、86.2%、91.1%。特征点提取后,点云数据量减少约85%,停车视距检验效率提高约70%。在同等路面条件下,2条城市双车道的全路段停车视距满足率平均值由行车速度为30 km/h时的83.7%降低至行车速度为60 km/h时的51.0%;在同等速度下,2条城市双车道的全路段停车视距满足率平均值由干燥路面时的68.8%降低至冰滑路面时的30.9%,说明行车速度提高或路面条件不良时,视距值显著降低。该研究可为营运期道路停车视距检验与评价提供详实的决策依据。For the urban double-lane highway scene,high-density LiDAR point clouds and PointNet++are used for automatic classification.A widening ellipsoid-view model is created,which is included in the roadside environment.The internal shape descriptor algorithm is introduced to efficiently extract the key feature points in the view.And then,the automatic inspection model of stopping sight distance(SSD)is built to carry out the 3D SSD inspection un-der different driving speeds and road conditions.Finally,two indicators of the satisfaction rate of SSD in the entire road and a single sight location are proposed to evaluate the road safety.Through the case study of the urban double-lane highway,the results show that the mean intersection over union,accuracy,precision and recall of road scene automatic classification are 81.9%,91.7%,86.2%and 91.1%,respectively.After feature point extraction,the amount of point clouds is reduced by about 85%,and the SSD inspection efficiency is improved by about 70%.Under the same pavement conditions,the average SSD satisfaction rate for the entire section of the two urban double-lane road decreases from 83.7%at the speed of 30 km/h to 51.0%at the speed of 60 km/h;at the same speed,the average SSD satisfaction rate of the entire section decreases from 68.8%on dry road to 30.9%on slippery road,indicating that the sight distance value decreases significantly when the driving speed increases or the road conditions are poor.The study can provide detailed decision-making basis for the inspection and evaluation of road SSD during the oper-ation period.

关 键 词：道路工程 停车视距 机器学习 视距满足率 激光雷达点云 视距安全分析 

分 类 号：U412.3[交通运输工程—道路与铁道工程]