Analyzing Architectural Floor Plans Using Neural Networks

Balázs Szőke

Supervisor(s): László Szécsi

Budapest University of Technology and Economics

Abstract: In this paper, we introduce a system that reconstructs a representation containing the positions and shapes of the walls, and the locations of the doors, from an input floor plan image (2D). We also build a 3D model from this obtained representation. To achieve this, we utilize traditional image processing algorithms to identify wall and door features, and convolutional neural networks based on differentiable rendering to reconstruct wall geometry. The purpose of our paper is to prove that the concept of differentiable rendering can help achieve a flexible and lightweight solution to this task.
Keywords: Computer Vision, Image Processing
Full text:
Year: 2022