基于点云分割的建筑平面图智能生成技术  

Point cloud segmentation-based technology for intelligent generation of architectural plan

作　　者：祁彦之 余润泽 李培振[1,2] QI Yanzhi;YU Runze;LI Peizhen(College of Civil Engineering,Tongji University,Shanghai 200092,China;State Key Laboratory of Disaster Reduction in Civil Engineering,Tongji University,Shanghai 200092,China)

机构地区：[1]同济大学土木工程学院,上海200092 [2]同济大学土木工程防灾国家重点实验室,上海200092

出　　处：《华中科技大学学报（自然科学版）》2023年第5期1-6,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家重点研发计划资助项目(2020YFC1512500).

摘　　要：为了使房屋检测中建筑平面图测绘更加智能化,提出一种基于点云分割的建筑平面图智能生成技术.首先,使用多站点三维激光扫描仪获取点云数据,将所有站点数据进行点云配准拼接,获得完整的建筑模型,同时对点云进行降采样预处理;然后,采用随机抽样一致性(RANSAC)算法对建筑平面进行提取,通过调整点之间距离的阈值、拟合的平面数量和迭代的次数来分割出不同构件的平面;最后,选取合适的中间高度将分割后的点云模型切片,构建投影函数执行投影生成平面图.使用Python编程语言并结合Open3D点云数据处理库对所提算法进行验证,通过对教学楼某层扫描数据进行分析,实现了含有墙、门窗位置信息的建筑平面图的智能生成.研究结果表明:该技术可以准确地对三维激光扫描点云进行分析处理,且平均尺寸误差在4%以内,具有较高的分类精度.To make architectural plan drawings in house inspection more intelligent,a point cloud segmentation-based technology for intelligent generation of architectural plan was proposed.First,a multi-site 3D laser scanner was applied to obtain point cloud data,and they were aligned and stitched together to achieve the complete building model.Meanwhile,the point cloud was preprocessed by downsampling.Then,the random sampling consistency(RANSAC)algorithm was used to extract the architectural planes,and they were segmented by adjusting the threshold value of the distance between points,the number of fitted planes and iterations.Finally,a suitable intermediate height was selected to slice the segmented model,and the architectural plan was generated by execution of the projection function.Python and the Open3D point cloud data processing library were utilized to validate the proposed algorithm.By analyzing the scanned data of a layer of a school building,the intelligent generation of architectural plan containing the location information of walls,doors and windows was achieved.Research results show that this technique can accurately analyze and process 3D laser scanned point clouds with an average dimensional error of less than 4%,and has a high classification accuracy.

关 键 词：三维激光扫描 点云分割 建筑平面图 平面提取 随机抽样一致性 

分 类 号：TU198.5[建筑科学—建筑理论]