测绘通报 ›› 2022, Vol. 0 ›› Issue (12): 14-18,23.doi: 10.13474/j.cnki.11-2246.2022.0350

• 实景三维构建技术及其应用 • 上一篇    下一篇

三维建筑模型单体化的深度学习技术实现

王文娜1, 张弓2, 吴侃1, 王瑞1,3, 亓立壮1   

  1. 1. 中国矿业大学环境与测绘学院, 江苏 徐州 221116;
    2. 中煤(西安)地下空间科技发展有限公司, 陕西 西安 710000;
    3. 赣南科技学院, 江西 赣州 341000
  • 收稿日期:2022-01-07 修回日期:2022-09-21 出版日期:2022-12-25 发布日期:2023-01-05
  • 作者简介:王文娜(1996-),女,硕士,主要研究方向为智能化遥感解译研究。E-mail:wwnwxm@163.com
  • 基金资助:
    江西省教育厅科技项目(GJJ191594);江苏省自然科学基金(BK20190642)

Realize building monomerization of 3D model based on deep learning

WANG Wenna1, ZHANG Gong2, WU Kan1, WANG Rui1,3, QI Lizhuang1   

  1. 1. School of Environment Science and Informatics, China University of Mining and Technology, Xuzhou 221116, China;
    2. China Coal (Xi'an) Underground Space Technology Development Co., Ltd., Xi'an 710000, China;
    3. Gannan University of Science and Technology, Ganzhou 341000, China
  • Received:2022-01-07 Revised:2022-09-21 Online:2022-12-25 Published:2023-01-05

摘要: 倾斜摄影三维建模技术能够快速建立目标区域的实景三维模型,但是该机制下的模型为连续不规则三角网(TIN),无法对其中具体的地理对象进行分析。本文采用深度学习语义分割方法提取建筑物,并与传统面向对象的影像分析提取结果进行对比。研究表明语义分割方法能够有效提取倾斜模型对应数字正射影像中的建筑物。通过对提取结果进行后处理得到建筑物矢量面,为每个建筑物添加基本属性信息,利用矢量面实现建筑物单体化并发布三维模型和数据服务,能够更有效地利用倾斜摄影模型展示研究区域地理空间和建筑物信息。

关键词: 倾斜摄影, 三维建模, 图像分割, 深度学习, 模型单体化

Abstract: The oblique photography 3D modeling technology can quickly obtain the real 3D model of the target area, but the model under this mechanism is a continuous triangulated irregular network(TIN), which makes it impossible to analyze specific geographic objects. In this paper, two methods of object-oriented image analysis and deep learning semantic segmentation are used to extract buildings and comparative analysis. The result shows that semantic segmentation method can effectively extract the buildings in the digital orthophotos corresponding to the tilt model. By post-processing the extraction results and giving the vector data the basic attributes of the building, the vector datas are used to realize the building monomerization and publish 3D models and data services, which can more effectively use the oblique photography model to show the geographical space and the building information.

Key words: oblique photography, 3D modeling, image segmentation, deep learning, model monomerization

中图分类号: