Reconstructing Photogrammetric 3D Model by Using Deep Learning

June Hao Hou, Chi Li Cheng*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

Abstract

In this paper, we propose a deep learning-based method to reconstruct 3D models generated with photogrammetry techniques. The mechanism to reconstruct the model is detecting corners and then segment the polyline contours based on the detected corners. Moreover, we take advantage of the generative design tool (Grasshopper) to build a training gym system, constantly producing unique datasets to train the neural network. In the part of feature engineering, we use series of external angles as the geometrical feature to train the model, which means that the model processes a small number array instead of using images or any kind of huge multi-dimensional data. The result of this research reveals that the neural network model trained by our training gym system can learn how to detect corners on building contours successfully. The details of the training gym, dataset format, dataset generator, the deep learning model, and the potential usages will be elaborated in the following content.

Original languageEnglish
Title of host publicationAdvances in Science, Technology and Innovation
PublisherSpringer Nature
Pages295-304
Number of pages10
DOIs
StatePublished - 2021

Publication series

NameAdvances in Science, Technology and Innovation
ISSN (Print)2522-8714
ISSN (Electronic)2522-8722

Keywords

  • Deep learning
  • Geometry simplification
  • Mesh reconstruction
  • Photogrammetry
  • Polyline segmentation

Fingerprint Dive into the research topics of 'Reconstructing Photogrammetric 3D Model by Using Deep Learning'. Together they form a unique fingerprint.

Cite this