地球信息科学学报 ›› 2021, Vol. 23 ›› Issue (8): 1497-1507.doi: 10.12082/dqxxkx.2021.200742

• 遥感科学与应用技术 • 上一篇    下一篇

基于改进RANSAC算法的复杂建筑物屋顶点云分割

刘亚坤1(), 李永强1,*(), 刘会云1, 孙渡2, 赵上斌1   

  1. 1.河南理工大学测绘与国土信息工程学院,焦作 454003
    2.江苏省太仓市自然资源和规划局港区分局,太仓 215400
  • 收稿日期:2020-12-07 修回日期:2021-03-24 出版日期:2021-08-25 发布日期:2021-10-25
  • 通讯作者: 李永强
  • 作者简介:刘亚坤(1992— ),男,河南周口人,硕士生,主要从事点云数据处理研究。E-mail: 757899220@qq.com
  • 基金资助:
    国家自然科学基金项目(41771491);国家自然科学基金项目(41701597);国家自然科学基金项目(U1810203);中国博士后科学基金项目(2018M642746)

An Improved RANSAC Algorithm for Point Cloud Segmentation of Complex Building Roofs

LIU Yakun1(), LI Yongqiang1,*(), LIU Huiyun1, SUN Du2, ZHAO Shangbin1   

  1. 1. School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454003,China
    2. Port area branch of Taicang natural resources and Planning Bureau, Jiangsu Province, Taicang 215400, China
  • Received:2020-12-07 Revised:2021-03-24 Online:2021-08-25 Published:2021-10-25
  • Contact: LI Yongqiang
  • Supported by:
    National Natural Science Foundation of China(41771491);National Natural Science Foundation of China(41701597);National Natural Science Foundation of China(U1810203);China Postdoctoral Science Foundation Project(2018M642746)

摘要:

屋顶模型重建影响到建筑物完整模型重建质量,屋顶面点云分割质量对屋顶模型重建具有重要意义。针对传统RANSAC算法在屋顶点云面片分割时易产生错分割、过分割等问题,本文顾及点云位置信息,提出一种对点云重新分配的改进RANSAC点云分割算法。算法暂时剔除非平面内点,选取平面内点集中3个点作为初始样本,平面拟合判定邻域是否有效,从有效邻域中选取标准差值最小的3个点为初始模型。利用RANSAC算法对屋顶点云进行分割。利用K近邻算法统计误分类点与面片的距离降低误分类,优化过分割面片并进行连通性分析,利用距离及法向量一致性检验的方法重分配非平面内点。为验证本文算法有效性,选取芬兰Helsinki地区的3栋相互独立的复杂建筑物屋顶以及上海某小区的6栋建筑物群屋顶作为实验数据。在2组数据中,本文提出的改进RANSAC算法分割屋顶面片的平均准确率分别为92.17%、87.82%,78%的建筑物屋顶不存在过分割。在第2组数据中,所有分割面片上的点与其对应的最佳拟合平面的距离的标准差的平均值为0.030 m。实验结果表明,本文算法分割建筑物屋顶面片的准确率较高,较好的抑制了过分割现象,且抗噪能力强。

关键词: 机载LiDAR, 屋顶点云, RANSAC算法, 种子点选取, 屋顶点云分割, 误分类判定, 面片优化, 点云重分配

Abstract:

Roof model reconstruction affects the quality of building complete model reconstruction, and the segmentation quality of roof point cloud is of great significance for roof model reconstruction. Aiming at the problems of wrong segmentation and over segmentation in the traditional RANSAC algorithm, this paper proposes an improved RANSAC algorithm to redistribute the point cloud, considering the location information of the point cloud. The algorithm eliminates the non planar points temporarily, and selects three points from the planar points set as the initial samples in the way of R radius neighborhood to fit them. The distance between the remaining points in the neighborhood and the fitting plane is calculated, and the neighborhood meeting the threshold requirements is classified as an effective neighborhood, three points with the minimum standard deviation are selected as the initial model, RANSAC algorithm is used to segment the roof point cloud. Aiming at the misclassification phenomenon in segmentation results, the distance between misclassification points and patches is calculated by k-nearest neighbor algorithm, and then the misclassification points are reclassified, at the same time, the angleθ and the distance d between patches are considered to merge the over segmented patches, the Euclidean distance based clustering segmentation algorithm is used to analyze the connectivity of the merged patches. By using the distance from a point to a plane and the consistency of the normal vectors between the point and the plane, the non planar points are redistributed. In order to verify the effectiveness of the algorithm, three independent roofs of complex buildings in Helsinki area of Finland and six roofs of buildings in a residential area of Shanghai are selected as experimental data. In the first group of experiments data, the average accuracy of the segmentation of roof patch is 92.17%, and the highest accuracy is 93.18%. In the second group of experiments data, the average accuracy of the segmentation of the roof patch is 87.82%, and the highest accuracy is 94.44%. The average standard deviation of the distance between the points on all the segmentation patches and the corresponding best fitting plane is 0.030 m. According to the above two groups of experiments data, 78% of the buildings have no over segmentation, and the average accuracy is 90%. The experimental results show that the algorithm has a high accuracy in extracting the roof plane slice, which can suppress the over segmentation and has a good anti noise ability.

Key words: airborne LiDAR, roof point cloud, RANSAC algorithm, seed selection, roof point cloud segmentation, judging the point of misclassification, patch optimization, point cloud redistribution