基于点密度采样的建筑物点云立面分割方法-《集美大学学报（自然科学版）》

文章信息/Info

Title:: Building Facade Segmentation Method Based on the Point Density Sampling

作者:: 康星火1; 2; 陈水利1; 3; 吴云东1; 4; 蔡国榕1; 4; （1.厦门市无人机遥感应用工程技术研究中心，福建厦门 361021；2.集美大学理学院，福建厦门 361021；3.集美大学诚毅学院，福建厦门 361021；4.集美大学计算机工程学院，福建厦门 361021）

Author(s):: KANG Xinghuo1; 2; CHEN Shuili1; 3; WU Yundong1; 4; CAI Guorong1; 4; (1.Xiamen UAVRS Application Engineering Technology Research Center，Xiamen 361021，China;2.School of Science，Jimei University，Xiamen 361021，China;3.Chengyi University College，Jimei University，Xiamen 361021，China;4.Computer Engineering College，Jimei University，Xiamen 361021，China)

Keywords:: LiDAR point clouds; GSMOSAC（global sample and model optimize sampling and consensus)method; facade segmentation; point density

摘要:: 由于地面激光扫描仪扫描时常存在死角，导致点云缺失、密度不均匀等问题，使得建筑物立面难以完整分割，为点云后续三维重建带来了很大的困难。提出了一种基于点密度的指导采样方式，并对提取的模型进行再优化的分割算法，即GSMOSAC（global sample and model optimize sampling and consensus)算法。该算法改进了最小采样集的选取方式，并对采样模型进行优化处理，以提高所提取模型的可靠性。针对三种不同类型的激光雷达点云数据的实验结果表明，该算法的分割效果比传统的RANSAC算法和多结构(Multi-GS)算法都更好。

Abstract:: Since terrestrial laser scanner exists scanning corner which may lead to problems such as lacking of point cloud and uneven density，it is hard to complete segmentation of building facade and brings great difficulty for sequent 3D reconstruction.There exist a lot of algorithms related to building facade segmentation based LiDAR point cloud datas.RANSAC and Mutil-GS have obvious advantage in sampling strategy among these algorithms in the literature，but there exists shortcomings for model selection and subsequent optimization.Based on a guidance of sampling point density and optimizing the extracted model，this paper puts forward a Global Sample and Model Optimization Sampling and Consensus (GSMOSAC) algorithm.Comparing with the traditional RANSAC and Mutil-GS，the algorithm obtains a better segmentation quality in the light of the experiment results under three types of LiDAR point cloud datas.