Internal wall layout estimation and automated 3D reconstruction of masonry buildings using building contours

Chenyu Zhang, Changhai Zhai, Weiping Wen, Penghao Ruan

Abstract: The digitization of urban building clusters represents a significant trend in current development, with three-dimensional (3D) building models being an essential component. However, the incomplete or missing information for many urban buildings has become a major obstacle to achieving this goal. Existing 3D reconstruction methods suitable for urban building clusters fail to account for internal wall information, particularly in scenarios where building data is incomplete or missing. This paper proposes a novel method that combines domain knowledge from the architectural design field with a physics-constrained generative adversarial network (GAN) to estimate internal wall layouts inside masonry buildings using building contours and limited available building parameters. Through the development of specialized image processing algorithms, the coordinates of the estimated wall layouts are extracted, enabling automatic 3D modeling of buildings with internal walls through Revit’s secondary development. Validated on two real-world masonry residential cases, the method achieved a mean pixel accuracy of 89.6% in reconstructing internal walls, demonstrating excellent predictive performance and applicability. Furthermore, the integration of the method and model proposed in this paper, along with the development of a User Interface (UI) system, has extended the automated workflow, allowing for the estimation of internal wall spatial layouts and the subsequent 3D building modeling, all starting from easily accessible urban-scale building information.

External IDs:dblp:journals/aei/ZhangZWR25