Evaluation of a UAS-based Bridge Inspection Framework with Automated Damage Candidate Suggestion and Human-in-the-loop Damage Assessment

Abstract. Unmanned aerial systems have shown considerable potential to improve bridge inspection procedures by enhancing safety, efficiency, and data quality. However, developing a comprehensive system that integrates safe flight planning, data acquisition, automated damage detection, and reporting while addressing practical challenges remains complex. In this paper, we present a UAS-based bridge inspection pipeline that has been validated through a real-world case study. The system generates safe and efficient flight routes, which were closely followed by the UAS, achieving an RMSE of 0.67 m and ensuring successful camera alignment and precise photogrammetric 3D models with a mean distance error of 3.2 cm compared to terrestrial laser scans. The damage detection system predicts potential damages, maps them onto the 3D model, computes various characteristics, and aggregates the predictions into a manageable set of damage candidates. A human-in-the-loop validation via a graphical user interface refines these results, producing a verified damage report in less than 4.25 hours, providing accurate, actionable data for effective bridge maintenance. Our experimental results indicate that the proposed approach is both practical and effective for comprehensive bridge inspections. This paper systematically evaluates the system, highlights key strengths and limitations, and provides critical insights for future improvements.