Visual inspection remains the most fundamental and widely used method for assessing the condition of bridges. This process involves observation of structural surfaces at a close distance to identify visible signs of deterioration such as cracking, spalling, corrosion, and delamination. Traditionally, human inspectors perform visual inspections manually. This labour-intensive process is associated with many limitations, for example, subjectivity to an inspector’s interpretation, difficulty accessing structural components, management of large volumes of unstructured data and the lack of consistent historical records. Recent advancements in computer vision and artificial intelligence have enabled considerable progress toward automating visual inspections. However, the full automation of visual inspections in practical, real-world scenarios remains constrained by several challenges: (i) the continued need for human intervention, (ii) the limited availability of high-quality labelled datasets, (iii) the generalizability of existing models, and (vi) the lack of standardized inspection protocols. In this positioning paper, we present an overview of the current state of automated visual inspection for defects identification in bridges. It reviews key open-source datasets of defects and state-of-the-art deep learning models. We give our forward-looking perspective on fully automated defects identification systems that align with standardized visual inspection guidelines.
Khan et al. (2025) studied this question.