Finding Geometric and Topological Similarities in Building Elements for Large-Scale Pose Updates in Scan-vs-BIM

Fiona C. Collins, Alexander Braun, André Borrmann

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

1 Scopus citations

Abstract

Information-rich BIM models are rarely usable off-the-shelf for operations tasks. Change decisions made on the construction site can lead to significant differences between the as-designed and as-built state of buildings. The responsibility for keeping the digital representation in sync with its physical twin is not defined and will likely only fully be assigned when automatic methods facilitate the geometric update process. To this end, previous research succeeded in (1) identifying if an element was erected at the time and position it was initially designed, and (2) updating the parametric design geometry to fit its LiDAR-measured as-built state under a set of assumptions and threshold values. The research presented in this paper aims at updating the as-designed model in case of significant pose differences between the as-designed and as-built state. The method leverages graphs to encode the topological connectivity between geometric elements, once for the as-designed BIM model and once for the as-built point cloud. A similarity metric, namely the cosine distance, allows for a quantitative comparison of the topologically enriched point cloud clusters and their corresponding BIM element. The results show that a convincing type-wise similarity can be found in the feature space between the as-built point cloud clusters and the BIM elements. This similarity score becomes meaningful once the element’s topological arrangements are included. An instance-wise similarity score of above 90% is achieved for matching-pairs of free-standing columns and allows for a large-scale pose update in the as-designed BIM model.

Original languageEnglish
Title of host publicationAdvances in Information Technology in Civil and Building Engineering - Proceedings of ICCCBE 2022 - Volume 1
EditorsSebastian Skatulla, Hans Beushausen
PublisherSpringer Science and Business Media Deutschland GmbH
Pages517-530
Number of pages14
ISBN (Print)9783031353987
DOIs
StatePublished - 2024
Event19th International Conference on Computing in Civil and Building Engineering, ICCCBE 2022 - Cape Town, South Africa
Duration: 26 Oct 202228 Oct 2022

Publication series

NameLecture Notes in Civil Engineering
Volume357
ISSN (Print)2366-2557
ISSN (Electronic)2366-2565

Conference

Conference19th International Conference on Computing in Civil and Building Engineering, ICCCBE 2022
Country/TerritorySouth Africa
CityCape Town
Period26/10/2228/10/22

Keywords

  • GNN
  • Knowledge-driven as-built reconstruction
  • Scan-vs-BIM

Fingerprint

Dive into the research topics of 'Finding Geometric and Topological Similarities in Building Elements for Large-Scale Pose Updates in Scan-vs-BIM'. Together they form a unique fingerprint.

Cite this