To main content

ContextNET - Contextual classification of building components using deep neural networks

Abstract

In recent decades, the construction industry has made significant steps towards digitalisation. One of the most persistent challenges remains ensuring data accuracy, particularly in building component classification. Recent research has explored the application of machine learning algorithms for element classification within Building Information Modelling (BIM). These methods, while effective in handling individual geometric features, often overlook the critical influence of an element’s spatial context, leading to misclassifications in dense and interconnected models. To address this gap, ContextNet proposes an innovative deep-learning approach that integrates contextual information from surrounding components, enhancing classification accuracy by accounting for spatial relationships. By processing point clouds from both individual elements and their context, ContextNet enables more robust identification of building components within BIM models. Results demonstrate that ContextNet significantly improves classification accuracy and robustness, reducing common misclassifications in complex BIM datasets. The proposed framework enhances BIM data by enabling contextaware component recognition, supporting more accurate and dependable classification within the construction industry.
Read the publication

Category

Academic article

Language

English

Author(s)

  • Wojciech Teclaw
  • Artur Tomczak
  • Mateusz Kasznia
  • Marcin Luczkowski
  • Nathalie Labonnote
  • Eilif Hjelseth

Affiliation

  • SINTEF Community / Architecture, Materials and Structures
  • Poland
  • Norwegian University of Science and Technology

Year

2025

Published in

Advanced Engineering Informatics

ISSN

1474-0346

Volume

67

View this publication at Norwegian Research Information Repository