Structural element recognition with computer vision for 3D BIM model generation

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

School of Science | Master's thesis

Department

Mcode

Language

en

Pages

56

Series

Abstract

Building Information Modeling (BIM) is the process of creating a digital representation of structural objects using computer software. In recent decades, BIM has greatly contributed to multiple phases of the construction life cycle, including design, operations and maintenance (O&M), and cost estimation. This advancement has also encouraged research on automation, one prominent topic being the reconstruction of 3D models, as manual modeling is labor-intensive and costly. In typical construction workflows, different parties exchange design drawings throughout the project, often requiring the same structures to be remodeled multiple times. Furthermore, many existing and aging buildings lack digital 3D models, motivating research into automated reconstruction methods that can recover BIM representations from available data in different forms. Among these, reconstructing 3D models from 2D drawings is a cost-effective off-site approach that remains relatively underexplored. While most studies focus on architectural floor plans, structural drawings contain geometric and spatial relationships that directly define the 3D form but require special attention due to their complexity. This study addresses this gap by proposing a computer vision based approach for recognizing structural general arrangement (GA) drawings and automatically generating a baseline for 3D BIM model creation in Tekla Structures. The extraction of constructional information is formulated as an image segmentation problem, where elements such as walls, openings, and grid lines are identified and converted into 3D geometries. The resulting pipeline demonstrates the feasibility of automated structural BIM reconstruction and establishes a foundation for future extensions. Potential directions include expanding the model to handle a wider range of structural elements, integrating profile and material information from annotations, and applying the framework to different construction types and disciplines to enhance its generalizability and practical impact.

Description

Supervisor

Kannala, Juho

Thesis advisor

Pitkänen, Henri
Shanmugavel, Vishakaraj

Other note

Citation