Towards automatic construction of photorealistic BIM window elements from single-picture input

Conference paper (2021)

Authors

Johan Luttun AECgeeks
T.F. Krijnen Urban Data Science - Architecture and the Built Environment
Research Group
Urban Data Science (Architecture and the Built Environment) (TU Delft)
Copyright: © 2021 Johan Luttun, T.F. Krijnen
Computer Vision VR Object detection BIM VR environment Photorealistic BIM
More Info
expand_more

Published Date

2021

Language

English

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Faculty

Architecture and the Built Environment

Department

Urbanism

Research Group

Urban Data Science

Abstract

Recent progress in computer vision and the pervasive use of Building Information Models (BIM) in the construction industry provide countless opportunities for decision making in architectural design. It allows capturing real-world geometrical data for assessing design options, such as windows products for replacement on the façade of an existing building. However, capturing information from the real world is a tedious task that involves data acquisition followed by costly processing steps. In this study, we focus on using a single 2D picture containing simple building elements in their context to generate a semantic 3D model of these elements. To that aim, we trained a detection neural network whose input constituted of labeled photographs containing windows. After training, the model is able to locate and segment elements of interest on unseen new data. With the use of Computer Vision methods, we derive accurate geometric outlines suitable to reconstruct photorealistic three-dimensional BIM models.