Architectural Images as Safety Data: Automated Crack Detection for Proactive Building Maintenance

Abstract

This study applies deep learning-based computer vision to the domain of architectural image analysis, specifically for automated crack detection on building facades. As architectural imagery increasingly serves as critical data for structural health monitoring, we introduce a YOLOv13-based detection system capable of processing visual data from drones or ground-based surveys with high speed and over 90% accuracy. By framing cracks not only as structural defects but as visual indicators of material fatigue and environmental stress, this work bridges methodological approaches from engineering, digital design, and visual studies. The proposed system supports a shift from reactive to proactive maintenance, illustrating how computational image analysis can enhance building safety while contributing to interdisciplinary discussions on the interpretive and diagnostic value of architectural images in socio-technical contexts.