AR-Enhanced Indoor Construction Progress Monitoring Using Bim and Synthetic Data
Mathis Baubriaud*, Stephane Derrode, Rene Chalon and Kevin Kernn
ABSTRACT
Manual inspection of indoor construction sites for progress monitoring is time-consuming, error-prone, and inefficient. Automated solutions using Deep Learning (DL) and Augmented Reality (AR) offer significant potential, but are hampered by the scarcity of large labeled datasets, especially for complex indoor environments. This paper presents a novel and automated methodology for Indoor Construction Progress Monitoring (ICPM) that addresses this data bottleneck by leveraging Building Information Modeling (BIM) and synthetic data. Our approach uses a photorealistic graphics engine to generate a large, annotated synthetic dataset of Mechanical, Electrical, and Plumbing (MEP) components within BIM environments. A YOLOv8 instance segmentation model, enhanced with domain adaptation techniques, is trained on this synthetic data and integrated with an AR application on HoloLens 2 for real-time on-site progress monitoring.
Experiments demonstrate that the proposed synthetic data-powered model achieved a substantial improvement in mAP50 compared to models trained on limited real-world data. A preliminary on-site validation further highlights the practical potential of the AR-integrated system for efficient and reliable ICPM, demonstrating a feasible path towards accessible and user-friendly automated inspection tools that can be readily adopted by construction professionals on real-world sites.
