顾及扫描线分布特征的地铁隧道移动扫描点云超分辨率方法  

Subway Tunnel Mobile Laser Scanning Point Cloud Upsampling Method Based on the Distribution Characteristics of Scan Lines

作　　者：张家文 梁嘉辉 张秋昭[2] 段伟 张开坤 ZHANG Jiawen;LIANG Jiahui;ZHANG Qiuzhao;DUAN Wei;ZHANG Kaikun(China Coal Jiangsu Survey Design and Research Institute Co.,Ltd.,Wuxi 214000,China;School of Environment Science and Spatial Informatics,China University of Mining and Technology,Xuzhou 221116,China;Nanjing Institute of Surveying,Mapping&Geotechnical Investigation Co.,Ltd.,Nanjing 210019,China)

机构地区：[1]中煤江苏勘测设计研究院有限公司,江苏无锡214000 [2]中国矿业大学环境与测绘学院,江苏徐州221116 [3]南京市测绘勘察研究院股份有限公司,江苏南京210019

出　　处：《金属矿山》2024年第12期231-239,共9页Metal Mine

基　　金：徐州市重点研发计划(社会发展)-社会事业项目(编号:KC23295);自然资源部国土卫星遥感应用重点实验室开放基金项目(编号:KLSMNR-G202222)。

摘　　要：基于深度学习的隧道点云超分辨率技术能够将稀疏点云进行上采样,获得更加真实、丰富的物体点云信息,为高精度的三维重建、病害识别提供数据基础。基于“插值生成点云+深度学习位置优化”的思想,提出了一种基于扫描线分布特征的地铁隧道移动扫描点云超分辨率模型和方法。应用扩张K近邻采样原理,改善了生成的插值点云空间分布。设计了融合扫描线空间分布特征的点云位置优化深度学习网络模型。在点距离特征提取过程中融合了扫描线空间分布特征,增强了网络模型对点云扫描线分布特征的感知与利用能力,从而改善地铁隧道移动扫描点云超分辨率处理效果。利用南京某地铁隧道实测点云数据制作点云超分辨率数据集,对所提模型进行了验证。结果表明:测试数据集的超分辨率结果CD值达到3.72,HD值达到50.57,相比经典的Grad-PU模型分别降低了19.31%和14.22%,对于低分辨率点云中扫描线间隙较大、点云密度不均匀的区域,能够取得更加均匀且准确的超分辨率处理结果。The tunnel point cloud upsampling algorithm based on deep learning can upsample sparse point clouds to obtain more realistic and richer object point cloud information.This can provide a data foundation for high-precision 3D reconstruction and disease identification.Based on the concept of"interpolation-generated point clouds and deep learning position optimization",this paper proposes a point cloud upsampling method for mobile scanning point clouds of subway tunnels based on the distribution characteristics of scan lines(DCSI).The application of K-nearest neighbor sampling principle improves the spatial distribution of the generated interpolated point cloud.A point cloud position optimization deep learning network model is designed that integrates the spatial distribution characteristics of scanning lines.In the process of extracting point distance features,the spatial distribution characteristics of scanning lines are integrated to enhance the network model's perception and utilization ability of point cloud scanning line distribution characteristics,and improve the effect of super-resolution processing of subway tunnel mobile scanning point clouds.A point cloud super-resolution dataset was created with real point cloud data from a subway tunnel in Nanjing to validate the model proposed in this paper.The experimental results of point cloud upsampling indicate that the CD value reached 3.72 and the HD value reached 50.57,which were 19.31%and 14.22%lower than Grad-PU model,respectively.For regions with large scan line gaps and uneven point cloud density in low resolution point clouds,more uniform and accurate super-resolution processing results can be obtained.

关 键 词：点云超分辨率 地铁隧道 移动激光扫描 Grad-PU 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]