多视立体网格的城市建筑物三维建模  

3D modeling of urban buildings based on multi view stereo mesh

作　　者：卢露媛 颜青松 曲英杰 李鹏[1] 邓非[1,2] LU Luyuan;YAN Qingsong;QU Yingjie;LI Peng;DENG Fei(School of Geodesy and Gcomatics*Wuhan University,Wuhan 430079,China;Key Laboratory of Urban Land Resources Monitoring and Simulation,MNR,Shenzhen,Guangdong 518000,China)

机构地区：[1]武汉大学测绘学院,武汉430079 [2]自然资源部城市国土资源监测与仿真重点实验室,广东深圳518000

出　　处：《测绘科学》2021年第8期112-119,共8页Science of Surveying and Mapping

基　　金：自然资源部城市国土资源监测与仿真重点实验室开放基金资助课题项目(KF-2018-03-025)。

摘　　要：针对城市建筑通常是由平面屋顶和垂直墙体连接而成的分段平面对象,该文提出了一个大规模城市建筑物的有效建模框架,在该框架中有两个主要步骤:分类和建模。首先采用基于征集的随机森林分类方法将大范围场景进行分割,然后采用假设和选择的建模策略对多种建筑物进行三维建模。该算法以多视点立体视觉系统生成的曲面网格为输入,输出具有不同细节层次的简化三维模型。实验表明,该方法能对多种外观的建筑物进行不同细节层次的建模,模型外观简洁规整,并能在一定程度上弥补原始数据中的噪声和遮挡问题。In view of the fact that urban buildings are usually piecewise planar objects connected by flat roofs and vertical walls, in this paper, we proposed an effective modeling framework for large-scale urban buildings. There were two main steps in our framework: classification and modeling. First, a random forest classification method based on feature set was used to segment a wide range of scenes, and then a variety of buildings were modeled in 3 D using modeling strategies of assumptions and selection. The algorithm took the surface mesh generated by the multi-view stereo vision system as input and output clean models with different level of detail. Experiments showed that this method could be used to model buildings with various appearance at different levels of detail, and the appearance of the model was clean and regular, and the noise and occlusion in the original data could be compensated to some extent.

关 键 词：网格分割 随机森林 建筑三维建模 多边形表面重建 

分 类 号：P231[天文地球—摄影测量与遥感]