基于机载LiDAR点云的建筑物多细节层次建模方法  

Building multiple level-of-detail reconstruction method based on airborne LiDAR point cloud

作　　者：崔扬 杨伯钢[1,2] 李晓亮 刘鹏[1,2] 董志海 CUI Yang;YANG Bogang;LI Xiaoliang;LIU Peng;DONG Zhihai(Beijing Institute of Surveying and Mapping,Beijing 100038,China;Key Laboratory of Urban Spatial Information Engineering,Beijing 100038,China)

机构地区：[1]北京市测绘设计研究院,北京100038 [2]城市空间信息工程北京市重点实验室,北京100038

出　　处：《北京测绘》2023年第10期1380-1385,共6页Beijing Surveying and Mapping

基　　金：国家自然科学基金杰出青年基金(41725005);北京市自然科学基金(4214069);城市空间信息工程北京市重点实验室开放基金(20220104)。

摘　　要：机载激光雷达点云(LiDAR)重建大范围建筑物模型一直是实景三维建模的热点问题,由于机载激光点云稀疏、建筑物立面数据缺失,给模型自动化重建带来了极大的挑战。为了解决该问题,本论文研究基于图优化理论重建建筑物多细节层次(LOD)模型。首先,对原始点云进行自动化滤波,分离地面点和非地面点,对地面点构建数字高程模型(DEM),对非地面点半自动提取单体化建筑物点云。然后,基于滚球法(Alpha shape)提取建筑物边界,利用通用图优化方法(G2O)对误差线进行全局一致性改正,获得规则化的二维边框。并基于建筑物屋顶三维高程及DEM高程值重建建筑物LOD1模型。其次,根据点云的高程差异生成高程栅格图,从高程栅格图提取建筑物轮廓线,对轮廓线进行简化、规则化、聚类,并将规则后的边界线拉伸获取建筑物立面结构,弥补建筑物立面数据缺失对建模的影响。最后,将屋顶平面相交、建筑物立面裁剪,对候选平面进行二元图割全局优化,选择最能表达建筑物结构的平面,以此重建建筑物LOD2模型。本论文选择北京市2017年机载激光点云进行实验,结果表明,本文提出的方法可以稳健地重建建筑物多细节层次模型,LOD2模型距离偏差为0.22 m,LOD1模型距离偏差为0.71 m,整体模型精确度高;针对1147个建筑物,LOD2和LOD1模型重建时间分别为1290s和1097s,具有较高的自动化程度和建模效率。The reconstruction of large-scale building models using airborne light detection and ranging(LiDAR)point clouds has always been a hot issue in real scene 3D modeling.Due to the sparse of airborne LiDAR point clouds and the lack of build facade data,which bring great challenges for automatic reconstruction.To solve the problem,this paper studies the reconstruction of building multiple Level-of-Detail(LOD)based on the graph optimization theory.Firstly,the original point clouds are automatically filtered,the ground points and non-ground points are separated,the digital elevation model(DEM)is constructed from the ground points,and the individual building is extracted semi-automatically from the non-ground points.Secondly,based on the Alpha shape method,the building boundary lines are extracted,and the error lines are corrected globally consistently by using the general graph optimization method(G2O)to obtain regularized two-dimensional(2D)boundary lines.And the LOD1 models of the buildings are reconstructed based on DEM elevation and the 3D elevation of the building roofs.Thirdly,the elevation raster map is generated based on the elevation differences of point clouds,the building contours are extracted from the elevation raster map,the contours are simplified,regularized,clustered,and then regular boundary lines are stretched to obtain the building façade structure,which compensate for the impact of missing building façade data on modeling.Finally,the roof planes are intersected,the building façades are clipped,the binary graph cut energy function is globally optimized for the candidate planes,and the planes that best express the building structure are selected,so as to reconstruct the building LOD2 models.This paper selects airborne LiDAR point clouds from Beijing in 2017 for experiments,and the results show that the proposed method can robustly reconstruct building multiple level-ofdetail models,which with the distance deviation of 0.22 meters for LOD2 model and 0.71 meters for the LOD1 model,the overall models

关 键 词：机载激光点云 误差线全局一致性改正 建筑物LOD1模型 高程栅格图 弥补数据缺失 二元图割 建筑物LOD2模型 

分 类 号：P258[天文地球—测绘科学与技术]