一种三维实景模型单体建筑物提取方法  

A Method for Extracting Single Buildings from Real Scene 3D Model

作　　者：赖伟斌 张云生[1] 杜守基 邹峥嵘[1] 王伟玺 李晓明 LAI Weibin;ZHANG Yunsheng;DU Shouji;ZOU Zhengrong;WANG Weixi;LI Xiaoming(l.Sehool of Geoscienees and Info-Physies, Central South University, Changsha 410083, China;Research Center of Digital City Engineering, Shenzhen 518040, China)

机构地区：[1]中南大学地球科学与信息物理学院,湖南长沙410083 [2]深圳市数字城市工程研究中心,广东深圳518040

出　　处：《测绘与空间地理信息》2018年第4期30-32,36,共4页Geomatics & Spatial Information Technology

基　　金：卫星测绘技术与应用国家测绘地理信息局重点实验室开放基金(KLSMTA-201505);国家自然科学基金(41201472)资助

摘　　要：倾斜航摄仪的发展提高了三维实景模型生产的自动化程度,降低了生产成本。但目前软件自动生成的"一张皮"数据难以挂接GIS属性数据,因此有必要从三维实景模型中提取单体建筑物。针对此目的,本文提出了一种结合几何、纹理及语义信息的提取方法。首先利用几何、颜色以及语义信息将实景模型的三角面片分为地面、植被、建筑物立面和建筑物屋顶4类,然后通过组合验证建筑物立面和屋顶形成单体建筑物。采用"街景工厂"生产的2组三维实景模型数据进行实验,提取结果完整度大于90%,正确率高于93%。实验表明:本文方法能自动从三维实景模型数据中提取单体建筑物,可为后续的属性数据挂接提供基础。Development of aerial oblique camera improves the degree of automation for producing real scene 3 D model,and reduces the production cost. However,the ＂piece of leather data＂automatically generated by software is difficult to link GIS property data. In order to extract single buildings from real scene 3 D model,an extraction method combined geometric,color and semantic information is proposed. Firstly,geometric,color and semantic information is used to classify triangles to four categories as ground,vegetation,facade and building roofs. Then a single building is verified facades and building roof. Two groups of real scene 3 D model generated by Street Factory are employed to validate the proposed method. By comparing with ground truth data,the completeness is up to 90%,and the correctness is more than 93%. Experimental results reveal that the proposed method can automatically extract single building from real scene 3 D model data. It can provide a basic for subsequent attaching property data.

关 键 词：三维实景模型 建筑物单体提取 分类 拓扑关系 

分 类 号：P237[天文地球—摄影测量与遥感]