Equirectangular 360° Image Dataset for Detecting Reusable Construction Components

Abstract

Insufficient as-built data hinders the transition of the architecture, engineering, and construction (AEC) sector to a circular system. Combining reality capture and machine learning (ML) could help better detect reusable components. However, a comprehensive image dataset of on-site inventory for circular economy strategies has yet to be developed. This study introduces and describes the generation of a purpose-built, 360° dataset. Initial validation using the YOLOv8 object detection model demonstrates a 63.4% mean average precision (mAP50), making it viable for computer vision. Further exploration of automating building stock inventory using 360-degree images and ML for urban mining is needed.