基于PointNet++进行附属设施语义分割的隧道收敛变形分析  

Convergent deformation analysis of tunnels for semantic segmentation of ancillary facilities based on PointNet++

作　　者：卞政 石波[1] 吴凡 王静[1] 赵凯[1] 杨兴宜 BIAN Zheng;SHI Bo;WU Fan;WANG Jing;ZHAO Kai;YANG Xingyi(College of Geodesy and Geomatics,Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]山东科技大学测绘与空间信息学院,山东青岛266590

出　　处：《铁道科学与工程学报》2024年第11期4827-4839,共13页Journal of Railway Science and Engineering

基　　金：山东省自然科学基金资助项目(ZR2023MD048);山东省重点研发计划(重大科技创新工程)(2019JZZY010429);国家自然科学基金资助项目(41930535);山东科技大学科研创新团队支持计划(2019TDJH103)。

摘　　要：随着城市轨道交通日趋广泛,隧道结构变形引起的地铁安全事故凸显,亟需对运营期隧道进行变形检测。隧道衬砌作为隧道变形分析的研究对象,衬砌内表面存在的大量附属设施影响隧道收敛变形分析精度。为了提高变形分析精度,解决点云处理环节中存在的自动化程度低的问题,提出基于PointNet++点云语义分割的隧道收敛变形分析方法。首先利用深度学习方法进行点云语义分割,对隧道衬砌附属设施进行自动滤除。然后对隧道衬砌进行断面提取,利用随机抽样一致性算法(Random Sample Consensus,RANSAC)对隧道断面点云进行采样,分析隧道收敛变形程度,从Z+F PROFILER 9012A激光断面扫描仪获取山东省济南市地铁盾构隧道点云实测数据上并进行应用。研究结果表明:所提出的处理方法可以有效地将大规模隧道衬砌与连接紧密的附属设施分离出来,隧道附属设施总体分类精度达到96%,滤波结果较好地保留了隧道衬砌原始形态特征。在对隧道整体和局部收敛变形分析的重复性验证中,测试区间内隧道整体变形精度往返测长半轴平均偏差为1.04 mm,短半轴平均偏差为0.9 mm,测试区间内隧道局部收敛变形往返测标准差最小为0.773 mm,最大为0.938 mm,可以满足隧道收敛变形分析的精度需求。研究结果可以有效提升处理大规模隧道数据的自动化程度,具有良好的有效性与可靠性,对运营期地铁隧道收敛变形检测或监测有较好的实践应用意义。As urban rail transit becomes more and more extensive,underground safety accidents caused by deformation of tunnel structures are highlighted,and there is an urgent need for deformation detection of tunnels during the operation period.Tunnel lining is the main object of tunnel deformation analysis,and the existence of a large number of appendages on the inner surface of the lining affects the accuracy of tunnel convergence deformation analysis.In order to improve the accuracy of deformation analysis and solve the problem of low degree of automation in the point cloud processing link,a convergent deformation analysis method for tunnels based on semantic segmentation of PointNet++point cloud was proposed.Firstly,point cloud semantic segmentation was carried out using the deep learning method,and the tunnel lining appurtenances were automatically filtered out.Then,the tunnel lining was extracted from the section,and the point cloud of the tunnel section was sampled using the Random Sample Consensus Algorithm(RANSAC)to analyze the degree of convergent deformation of the tunnel.The point cloud of the underground shield tunnel in Jinan City,Shandong Province,obtained from the Z+F PROFILER 9012A laser section scanner was applied.Research results show that the proposed processing method can effectively separate the large-scale tunnel lining from the closely-connected appurtenances,and the overall classification accuracy of the tunnel appurtenances reaches 96%.The filtering results better retain the original morphological features of the tunnel lining.In the repeatability verification of the overall and local convergent deformation analysis,the average deviation of the long half-axis of the overall deformation of the tunnel in the test interval is 1.04 mm,and the average deviation of the short half-axis is 0.9 mm.The standard deviation of the round-trip measurement of the local convergent deformation of the tunnel in the test interval ranges from 0.773 mm at the minimum to 0.938 mm at the maximum.These results meet the ac

关 键 词：轨道交通隧道 激光点云 收敛变形分析 点云深度学习 随机抽样一致性 

分 类 号：U231.94[交通运输工程—道路与铁道工程]