测绘通报 ›› 2022, Vol. 0 ›› Issue (9): 52-57.doi: 10.13474/j.cnki.11-2246.2022.0263

• 学术研究 • 上一篇    下一篇

一种基于倾斜摄影测量点云密度自适应分割的建筑物边界提取方法

刘禹麒1, 陈广亮1, 蔡岳臻1, 李名豪2, 陈定安2, 胡小中1   

  1. 1. 广州蓝图地理信息技术有限公司, 广东 广州 510650;
    2. 中山大学地理科学与规划学院, 广东 广州 510275
  • 收稿日期:2022-02-23 修回日期:2022-07-20 发布日期:2022-09-30
  • 通讯作者: 陈定安。E-mail:chantingon@mail.sysu.edu.cn
  • 作者简介:刘禹麒(1980—),男,硕士,高级工程师,研究方向为摄影测量、三维建模及应用等。E-mail:lyq657@126.com
  • 基金资助:
    中山大学横向结题项目续研预研经费(37000-14090113);广州蓝图地理信息技术有限公司科研项目(2021RD7)

A density-adaptive building boundary extraction method based on 3D point clouds obtained from oblique photogrammetry

LIU Yuqi1, CHEN Guangliang1, CAI Yuezhen1, LI Minghao2, CHEN Dingan2, HU Xiaozhong1   

  1. 1. Guangzhou Lantu Geographic Information Co., Ltd., Guangzhou 510650, China;
    2. School of Geometry and Planning, Sun Yat-sen University, Guangzhou 510275, China
  • Received:2022-02-23 Revised:2022-07-20 Published:2022-09-30

摘要: 针对倾斜摄影场景中建筑物单体化问题,本文提出了基于倾斜摄影测量点云数据的建筑物识别和边界提取自动化算法。首先,对点云进行预处理,去除地面点和噪声点;然后,对点云进行二维栅格化处理,按间隔距离预分割;最后,结合改进的大津算法和区域增长算法,从预分割点云识别其中的建筑物,并提取建筑物边界点。从广东省江门市和湛江市选取两处试验区域对算法进行测试,结果表明:区域内建筑物点云均能准确被分割识别,建筑物边界提取准确度分别为87.8%与92.3%,说明本文提出的方法对于倾斜摄影测量建筑物识别和边界提取的适用性较强。

关键词: 倾斜摄影, 单体化, 特征提取, 聚类分割, 点云数据

Abstract: To tackle the problems of individual building segmentation, this paper proposes an automatic building boundary segmentation method for point clouds derived from the oblique photogrammetry. First, the ground and noise points are filtered out through the pre-processing stage. Then, the point clouds are voxelized for the segmentation. An improved Otsu method and the distance-based region-growing algorithm are integrated to apply to the voxelized point clouds for the boundary extraction. We verify the proposed method with two datasets captured in the Jiangmen and Zhanjiang of the Guangdong Province. The results show that the segmentation accuracy achieve 87.8% and 92.3% for Jiangmen and Zhanjiang datasets, respectively. This suggests that the proposed method is highly applicable to the building segmentation.

Key words: oblique photography, monomerization, feature extraction, cluster segmentation, point clouds

中图分类号: