Exploring the Applicability of Computer Vision Models for Automated Code Compliance Checking

Abstract: Automated Compliance Checking (ACC) is a longstanding challenge traditionally addressed through rule-based systems operating on fully parametric Building Information Models (BIM). However, in practice, building information is still predominantly conveyed through two-dimensional (2D) drawings, limiting the applicability of existing approaches. While recent advances in Computer Vision (CV) have enabled reliable detection and segmentation of elements in design drawings, these techniques are commonly treated as standalone recognition tasks. We propose a lightweight, computational framework for ACC directly from 2D drawings using CV as an enabling technology. We evaluate the applicability of CV workflows to support ACC through a case study, focusing on regulatory requirements for bathrooms. The paper describes the full computational pipeline, from drawing preprocessing and element segmentation to machine-interpretable representations suitable for ACC and up to the compliance checking process. Results indicate that CV technology is a viable computational bridge between traditional drawing-based engineering practices and automated compliance methods.