Computer vision-based approach for autonomous estimation of building envelope material stock

• Computer vision-based approaches for estimating building envelope stock. • Elemental-level automated and scalable stock estimation. • Integrates computer vision, GIS, and building measurement for stock estimation. • Supports circular economy by estimating materials for façade and roof recovery. Building envelope components, such as façades and roofs, typically have shorter life spans than structural elements and experience more frequent replacement rates. These components present critical opportunities for promoting circular economy (CE) strategies, particularly reuse, recycling, and material recovery, while minimising down-cycling and landfill. However, accurate and scalable building stock estimation at the elemental level remains a challenge due to the significant time and cost involved in conventional data collection methods. To address this gap, this study employs computer vision-based approaches to autonomously estimate building envelope material stock, focusing on façades and roofs within the Southport North Statistical Area Level 2 (SA2) region of Queensland, Australia. For façade classification, a ResNet50 model with an 80 % accuracy rate was deployed. Roof material and roof type, instance segmentation was undertaken using the YOLOv8 model, which achieved approximately 70 % accuracy. High-resolution street-level and aerial imagery supported the classification and segmentation tasks. Where image data were unavailable, weighted material intensity values were applied to supplement stock estimation. Unlike prior studies that relied on aggregated data, this study introduces a point-level spatial analysis framework to estimate stock at the individual building level. This granularity supports CE applications, such as material recovery planning and facility locating strategies. The integrated approach demonstrates a scalable pathway for autonomous, bottom-up envelope material stock estimation using computer vision, offering a significant advancement in closing the information gap for CE planning and applications.