Semantic Segmentation of Building Point Clouds Using Deep Learning: A Method for Creating Training Data Using BIM to Point Cloud Label Transfer

Thomas Czerniawski, Fernanda Leite

Research output: Chapter in Book/Report/Conference proceedingConference contribution

5 Scopus citations

Abstract

Creating deep learning classifiers requires large labeled datasets; and creating large labeled datasets requires elaborate crowdsourcing systems and many hours of manual human effort applied to classification and data entry. Fortunately, much of this effort can be bypassed in the building industry because of as-built building information models (BIMs), a semantically rich form of facility information. From these BIMs, semantics can be transferred to point clouds. This paper presents a method for creating large labeled datasets for training deep neural networks to semantically segment point clouds of buildings. Geometry and attached semantics are extracted from a BIM. The geometry is registered with the point cloud and the BIM semantics are copied to the points in the point cloud. The presented method enables organizations with access to as-built BIMs to forgo the effort of creating large labeled datasets and instead use the embodied effort in their pre-existing BIMs.

Original languageEnglish (US)
Title of host publicationComputing in Civil Engineering 2019
Subtitle of host publicationVisualization, Information Modeling, and Simulation - Selected Papers from the ASCE International Conference on Computing in Civil Engineering 2019
EditorsYong K. Cho, Fernanda Leite, Amir Behzadan, Chao Wang
PublisherAmerican Society of Civil Engineers (ASCE)
Pages410-416
Number of pages7
ISBN (Electronic)9780784482421
DOIs
StatePublished - 2019
Externally publishedYes
EventASCE International Conference on Computing in Civil Engineering 2019: Visualization, Information Modeling, and Simulation, i3CE 2019 - Atlanta, United States
Duration: Jun 17 2019Jun 19 2019

Publication series

NameComputing in Civil Engineering 2019: Visualization, Information Modeling, and Simulation - Selected Papers from the ASCE International Conference on Computing in Civil Engineering 2019

Conference

ConferenceASCE International Conference on Computing in Civil Engineering 2019: Visualization, Information Modeling, and Simulation, i3CE 2019
CountryUnited States
CityAtlanta
Period6/17/196/19/19

ASJC Scopus subject areas

  • Computer Science(all)
  • Civil and Structural Engineering

Fingerprint Dive into the research topics of 'Semantic Segmentation of Building Point Clouds Using Deep Learning: A Method for Creating Training Data Using BIM to Point Cloud Label Transfer'. Together they form a unique fingerprint.

Cite this