Hybrid semantic clustering of 3D point clouds in construction

  • In this work, the authors present an artificial intelligence (AI)-based semantic segmentation approach for three-dimensional (3D) point clouds which were generated from 2D images with a structure from motion (SfM) pipeline. The target application area are 3D scans of buildings in the construction domain. The goal is to differentiate basic objects, such as floor, ceiling, walls, or columns in a building during the construction phase. The authors utilize state-of-the-art neural networks (RandLA-Net) to label the 3D points, and the novelty of their approach lies in conducting the training of the network in the hybrid 2D/3D domain, rather than solely in the 3D domain. This allows us to benefit from the high efficiency and accuracy of pre-trained networks in the 2D domain. In our approach we register the 2D images to the 3D points cloud and thereby fuse the individual 2D segmentation results to create a consistently segmented 3D point cloud.

Download full text files

Export metadata

Additional Services

Share in Twitter Search Google Scholar
Author:Marcus ZeppGND
Parent Title (German):34th Forum Bauinformatik / 34. Forum Bauinformatik (Bochum, 06. - 08.09.2023)
Document Type:Part of a Book
Date of Publication (online):2023/09/07
Date of first Publication:2023/09/07
Publishing Institution:Ruhr-Universität Bochum, Universitätsbibliothek
3D Point Clouds in Construction; IFC Model Generation; Semantic Annotation; Semantic Clustering in Construction
First Page:423
Last Page:430
Institutes/Facilities:Lehrstuhl für Informatik im Bauwesen
Dewey Decimal Classification:Technik, Medizin, angewandte Wissenschaften / Ingenieurbau, Umwelttechnik
open_access (DINI-Set):open_access
faculties:Fakultät für Bau- und Umweltingenieurwissenschaften
Konferenz-/Sammelbände:34th Forum Bauinformatik / 34. Forum Bauinformatik (Bochum, 06. - 08.09.2023)
Licence (German):License LogoCreative Commons - CC BY 4.0 - Namensnennung 4.0 International