基于无人机路径规划与深度学习的桥梁点云自动化分割研究  

Study on Automated Segmentation of Bridge Point Clouds Based on UAV Path Planning and Deep Learning

作　　者：万桂军 吴维维 王浩琛 冯东明[3,4] WAN Guijun;WU Weiwei;WANG Haochen;FENG Dongming(College of Civil Engineering and Architecture,Zhejiang University,Hangzhou 310058,China;China Overseas Construction Limited,Shenzhen 518055,China;Key Laboratory of Concrete and Prestressed Concrete Structures of the Ministry of Education,Southeast University,Nanjing 211189,China;School of Civil Engineering,Southeast University,Nanjing 211189,China)

机构地区：[1]浙江大学建筑工程学院,浙江杭州310058 [2]中海建筑有限公司,广东深圳518055 [3]东南大学混凝土及预应力混凝土结构教育部重点实验室,江苏南京211189 [4]东南大学土木工程学院,江苏南京211189

出　　处：《湖南大学学报(自然科学版)》2025年第3期60-72,共13页Journal of Hunan University:Natural Sciences

基　　金：国家重点研发计划资助项目(2023YFC3805900);东南大学新进教师科研启动经费资助(RF1028623149)。

摘　　要：为推动桥梁管养事业发展的数字化、智能化和精细化,保障其安全服役,提出基于无人机路径规划与深度学习的桥梁点云自动化分割方法.首先对桥梁结构进行倾斜摄影建模,根据模型提供的空间信息,对桥面板、桥侧、桥墩和桥底四部分分别进行无人机飞行路径精细化规划,并按照新路径执行航拍任务,进行三维重建.其次,通过实桥试验进行方法验证,根据目标分辨率确定无人机飞行高度、航向和旁向重叠率等飞行参数,编写KML文件导入无人机,经验证重建所得的桥梁三维点云模型精度达到毫米级.最后,制作点云语义分割数据集,将点云数据划分为背景、桥面板、桥墩和盖梁四类,采用轻量高效的RandLA-Net算法进行桥梁构件语义分割,结果MIoU值为98.77%,各类别构件IoU值在95.46%以上,在桥梁点云的自动化分割上取得了良好的效果.To promote the digitalization,intelligence,and refinement of bridge maintenance and management and ensure the safe operation of the bridge,an automated bridge-point-cloud segmentation method based on unmanned aerial vehicle(UAV)path planning and deep learning is proposed.First,the bridge structure is modeled through oblique photography,and based on the spatial information provided by the model,path planning is performed for the bridge deck,bridge side,bridge pier,and bridge bottom,respectively,to obtain a detailed path planning scheme for the entire bridge.UAV aerial photography and 3D reconstruction are carried out accordingly.Second,the method is validated through on-site experiments on actual bridges,and flying parameters such as flight altitude,heading,and lateral overlap ratio are determined based on the target resolution.A KML file is then generated and imported into the UAV to reconstruct the bridge’s 3D point cloud model with millimeter-level accuracy.Finally,a point cloud semantic segmentation dataset is created,and the point cloud data is divided into four categories:background,bridge deck,bridge pier,and cap beam.The lightweight and efficient RandLA-Net algorithm is used for semantic segmentation of the bridge components,achieving a mean intersection over union(MIoU)value of 98.77%and IoU values of over 95.46%for each category of components,verifying the validity of the selected algorithm on bridge point cloud segmentation.

关 键 词：桥梁工程 三维点云 桥梁管养 自动化分割 无人机路径规划 

分 类 号：U446.3[建筑科学—桥梁与隧道工程]