Computer Vision Methods for Bridge Surface Damage Detection

After long-term service, bridges are prone to surface defects such as cracks, spalling, and corrosion. The timely identification and recording of these defects constitute an essential part of bridge inspection and maintenance. Conventional manual inspection relies on close-range visual observation and record keeping by inspectors; this approach carries safety risks when workers operate at height or beneath bridges, and its efficiency and consistency are readily affected by the experience of the personnel and by field conditions. In recent years, image-based computer vision (CV) methods have been increasingly applied to bridge surface damage detection. This paper adopts a literature review and a representative-study analysis to introduce bridge surface image data and commonly used CV methods, and to examine related research from three perspectives: crack detection and quantification, multi-type surface defect recognition, and CV-assisted bridge inspection. Existing studies show that CV can identify and localize common defects such as cracks, spalling, and corrosion, and can quantify the dimensions of certain defects when scale calibration is available. When combined with unmanned aerial vehicles (UAVs) and robots, CV reduces the need for close-range manual inspection and large-scale image screening. Nevertheless, detection performance remains affected by image quality, complex backgrounds, class imbalance, and shooting conditions, and the stability of models across different bridges and environments still has room for improvement. At present, CV is better suited as an auxiliary tool for bridge inspection, while significant defects and uncertain results still require confirmation by manual inspection.