基于多视图像点云的建筑物屋顶分割算法研究  

Research on Building Roof Segmentation Algorithm from Multiple View Stereo Point Clouds

作　　者：陈鑫祥[1] 蒲冰鑫 俞建[1] 王瑞胜 钟若飞[2] CHEN Xinxiang;PU Bingxin;YU Jian;WANG Ruisheng;ZHONG Ruofei(Land and Resources Technology Center of Guangdong Province,Guangzhou 510098,China;College of Resource Environment and Tourism,Capital Normal University,Beijing 100048,China;School of Geographical Sciences of GZHU,Guangzhou University,Guangzhou 510006,China)

机构地区：[1]广东省国土资源技术中心,广东广州510098 [2]首都师范大学资源环境与旅游学院,北京100048 [3]广州大学地理科学学院,广东广州510006

出　　处：《地理空间信息》2022年第5期33-37,共5页Geospatial Information

摘　　要：建筑物屋顶不仅能作为基于点云的城市三维重建的重要参考依据,而且能用于房屋地籍测量。以多视图像点云为数据源,提出了一种自动分割建筑物屋顶面片的算法。该算法由建筑物单体化、屋顶面片识别、面片优化3个部分组成。首先利用基于颜色不变量理论的植被过滤算法、地面过滤完成建筑物单体化;然后计算点特征去除建筑物立面,并利用区域增长算法实现屋顶面片识别;最后利用基于主成分分析的面片平整技术进行优化。大规模实验结果表明,该算法能生成理想的平整屋顶面片,正确率可达96.46%。Building roof can be used as an important reference for not only 3D urban reconstruction from point clouds, but also house cadastral survey. This paper presented a method to segment building roof planes from a large scale multi view stereo point clouds, which consists of three parts: building monomer, roof planes identification and plane optimization. The vegetation filtering algorithm based on color invariant theory and ground filtering algorithm were implemented to complete the monomer of buildings. Then, we calculated the point feature to remove the building fa?ade and used region growing algorithm to segment roof planes. Finally, a PCA-based plane processing technique was used to optimize the plane.

关 键 词：多视图像点云 地籍测量 颜色不变量理论 主成分分析 

分 类 号：P231[天文地球—摄影测量与遥感]