BIM-guided semantic fingerprinting for vision-based construction progress monitoring of cold-formed steel framing

Abstract

Construction progress monitoring remains largely dependent on manual inspection, limiting the frequency and reliability of updates on active sites. Existing automated approaches have demonstrated value but commonly rely on fixed camera infrastructure, depth sensing, or metric camera-to-BIM registration, requirements that are difficult to justify on budget-constrained residential projects, and most treat BIM as a passive post-detection reference rather than an active constraint during visual interpretation.

To address this gap, this study proposes a BIM-guided semantic fingerprinting framework for element-level progress monitoring of cold-formed steel framing using standard monocular video. A custom YOLOv8m detection model identifies wall panels, doors, and windows across video frames; visual odometry provides temporal reasoning, detecting façade transitions and suppressing duplicate detections as the camera moves along the building. Accepted detections are aggregated into semantic fingerprints and matched against BIM-derived façade representations through a reward-based scoring system, enabling façade identification and element assignment without metric registration; wall-panel coverage is assessed separately as spatial continuity along each face. A gap correction mechanism recovers missed detections using BIM spatial context, and a pyRevit extension supports end-to-end BIM extraction and in-model progress visualisation.

Evaluated on a single-storey residential cold-formed steel structure in Hamilton, New Zealand, the framework correctly identified all four building façades, matched element counts to ground truth, and reduced 2,050 raw detections across 983 frames to 20 unique element observations, a 99.02% reduction, without fixed infrastructure, depth sensing, or point-cloud reconstruction. The work is currently validated on a single proof-of-concept site, and further testing on independent projects is needed to establish broader generalisability.