A focused review of recent advances in deep learning and computer vision for concrete crack detection and measurement

Abstract

Abstract Concrete cracking is one of the most prevalent indicators of structural deterioration, posing significant risks to the safety, serviceability, and longevity of civil infrastructure. Traditional manual visual inspections are labor-intensive, subjective, and prone to human error, making them inadequate for large-scale structural health monitoring (SHM). In recent years, computer vision and deep learning (DL) have emerged as transformative, automated alternatives. This paper presents a focused review of recent advancements in vision-based concrete crack detection and measurement, synthesizing findings from 27 representative and influential studies selected through an expert-driven, non-exhaustive screening process rather than a fully systematic protocol. The review traces the technological evolution from traditional image processing to advanced deep learning architectures, categorizing methodologies into image classification, real-time object detection, and pixel-level semantic segmentation. Furthermore, it highlights novel approaches for translating pixel-based detections into practical physical measurements (millimeters) using laser calibration and RGB-D cameras. The study also addresses critical challenges, such as domain gaps, data imbalance, and environmental noise, while outlining future research directions, including lightweight edge-computing models and multi-modal sensor fusion.