基于车载激光点云的普通干线公路违法建筑自动化排查方法  

Automated Inspection Method of Ordinary Arterial Highway Illegal Construction Based on Vehicle-Mounted Laser Point Cloud

作　　者：聂世刚 诸雨 刘锦阳 徐阳 胡颖雷 NIE Shigang;ZHU Yu;LIU Jinyang;XU Yang;HU Yinglei(Henan Transportation Development Research Institute Co.,Ltd.,Zhengzhou 450000,China;School of Transportation,Southeast University,Nanjing 211100,China;Transportation Development Center of Henan Province,Zhengzhou 450000,China)

机构地区：[1]河南交通发展研究院有限公司,河南郑州450000 [2]东南大学交通学院,江苏南京211100 [3]河南省交通事业发展中心,河南郑州450000

出　　处：《交通运输研究》2025年第1期67-78,共12页Transport Research

基　　金：河南省交通运输厅科技项目(2021T1)。

摘　　要：为实现高效准确的普通干线公路违法建筑排查和治理,维护交通安全、提高运输效率,提出一种基于车载激光点云的普通干线公路违法建筑自动化排查方法。首先,构建一种基于二维和三维特征融合的混合道路边缘提取算法;其次,采用基于GB-ADBSCAN的点云聚类技术对违建点云整体进行识别,并通过调整特征数量、引入并行对称函数等改进PointNet++模型对单体违建对象进行分割,最终实现对遮挡环境下违建物的准确识别。最后,以河南省国道G207的激光点云数据为例,对所提出的排查方法进行验证。结果表明,所提出的混合道路边缘提取算法的提取精度达98.7%,提取效果显著优于基于扫描线、基于特征图像、基于空间特征的3种常用道路边缘线提取方法;通过实地勘测、实景图片对比可知,基于车载激光点云的违建物识别的整体准确率达87%;说明所提出方法能有效提高违建物点云的识别正确率和速度,实现路侧复杂环境下稳定、高效的普通干线公路违法建筑自动化排查。To achieve efficient and accurate inspection and management of ordinary arterial highway illegal construction,ensuring traffic safety and improving transportation efficiency,an automated inspection method of ordinary arterial highway illegal construction based on vehicle-mounted laser point clouds was proposed.Firstly,a hybrid road edge extraction algorithm was constructed based on the fusion of twodimensional and three-dimensional features.Secondly,the point cloud clustering technology based on GB-ADBSCAN was adopted to identify illegal construction point cloud as a whole,and the Point-Net++model was improved by adjusting the number of features and introducing parallel symmetry function to segment the single illegal construction,achieving the accurate inspection of illegal construction in sheltered environment.Finally,the proposed method was verified using laser point cloud data from national highway G207 in Henan Province.The results show that the proposed hybrid road edge extraction algorithm has an extraction accuracy of 98.7%,and the extraction effect was significantly better than that of three commonly used road edge line extraction methods based on scanlines,feature images and spatial features.As verified by field surveys and real-scene image comparison,the accuracy of unauthorized construction identifying based on vehicle-mounted laser point clouds reaches 87%,indicating that the proposed method can effectively improve the recognition accuracy and speed of the point cloud of illegal construction,and achieve stable and efficient automated inspection of highway illegal construction under the complex environment of roadside.

关 键 词：激光点云 违法建筑 特征融合 PointNet++ 自动化排查 

分 类 号：U412.2[交通运输工程—道路与铁道工程] U418.9