基于改进ICP算法和级联ASPP算法的异形建筑表面重建  

Surface reconstruction of irregular buildings based on improved ICP algorithm and cascaded ASPP algorithm

作　　者：史振玮 SHI Zhenwei(Shanghai Cehui Information Technology Company Limited,Shanghai 200433,China)

机构地区：[1]上海测慧信息科技有限公司,上海200433

出　　处：《北京测绘》2025年第4期462-467,共6页Beijing Surveying and Mapping

摘　　要：由于异形建筑独特的几何形态和复杂的表面结构,在对建筑表面进行重建时,表面识别结果容易出现偏差,影响重建模型的精度。对此,设计基于改进迭代最近点(ICP)算法和级联检测分割(ASPP)算法的异形建筑表面重建方法。改进ICP算法实施点云配准,将从不同位置采集到的点云数据统一到同一个坐标系下,消除位置偏差,提高点云数据的配准精度。通过级联ASPP算法设计轻量级语义分割模型,聚合多尺度信息,确保算法的识别范围能全面覆盖视野范围。利用随机抽样一致性(RANSAC)算法自动化处理异形建筑的识别数据,结合最小二乘法原理进行墙面立体拟合,得到高精度的表面重建参数,实现异形建筑表面重建。测试结果表明,设计方法的重建结果接近真实世界的几何形态,在异形建筑表面各墙面上重建偏差值均低于1.5。Due to the unique geometric shapes and complex surface structures of irregular buildings,surface recognition during reconstruction often leads to deviations,which can affect the accuracy of the reconstructed model.To address this issue,this paper proposed a method for the surface reconstruction of irregular buildings based on an improved iterative closest point(ICP) algorithm and an atrous spatial pyramid pooling(ASPP) algorithm.The improved ICP algorithm performed point cloud registration,aligning point cloud data collected from different locations into a common coordinate system,eliminating positional deviations,and enhancing registration accuracy.The Cascaded ASPP algorithm was used to design a lightweight semantic segmentation model that aggregates multi-scale information,ensuring comprehensive coverage of the visible area.The random sample consensus(RANSAC) algorithm was utilized to automatically process the recognition data of irregular buildings.Combined with the least squares method,it was used for the three-dimensional(3D) fitting of wall surfaces to obtain high-precision surface reconstruction parameters,ultimately achieving the surface reconstruction of irregular buildings.Test results show that the reconstruction closely matches the true geometric form,with reconstruction deviations on various walls of the irregular building surface remaining below 1.5.

关 键 词：点云配准 迭代最近点(ICP)算法 级联检测分割(ASPP)算法 随机抽样一致性(RANSAC)算法 异形建筑表面 

分 类 号：P208[天文地球—地图制图学与地理信息工程]