Deep-Learning for Lod1 Building Reconstruction from Airborne Lidar Data

Tee Ann Teo*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

A three-dimensional building model is an important geospatial information for a smart city. The objective of this study is to reconstruct OGC CityGML LOD1 prismatic building models from 3D lidar points automatically. A deep learning approach (i.e. Fully Convolutional Network, FCN) is developed to detect initial building regions for lidar data. After refinement, the building boundary needs regularization to reshape the irregular boundary into a regular building primitive. Finally, a 3D plane fitting is applied to shape the rooftop using lidar points inside building primitive. The test data was an urban area with the size of 1800m by 1200m. The lidar point density was 4 pt/m2. The experimental result indicated that the proposed method automatically reconstruct the LOD1 block model from lidar data. The accuracy of building detection reached 72% using lidar object height and intensity. The reconstruction showed high similarity with reference LOD1 building model.

Original languageEnglish
Title of host publication2019 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2019 - Proceedings
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages86-89
Number of pages4
ISBN (Electronic)9781538691540
DOIs
StatePublished - Jul 2019
Event39th IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2019 - Yokohama, Japan
Duration: 28 Jul 20192 Aug 2019

Publication series

NameInternational Geoscience and Remote Sensing Symposium (IGARSS)

Conference

Conference39th IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2019
CountryJapan
CityYokohama
Period28/07/192/08/19

Keywords

  • 3D building model
  • CityGML LOD1
  • deep learning
  • lidar.

Fingerprint Dive into the research topics of 'Deep-Learning for Lod1 Building Reconstruction from Airborne Lidar Data'. Together they form a unique fingerprint.

Cite this