Rapid and reliable assessment of bridge integrity after earthquakes is a critical element of emergency response. Traditional inspections depend on manual visual evaluations, which can delay decisions about bridge closure or repair. This research addresses this limitation by developing a vision-based framework for structural health monitoring of bridges. The proposed approach integrates drone-acquired imagery with advanced computer vision algorithms—specifically, the YOLOv8 object detector and the Segment Anything Model (SAM)—to identify structural components and measure critical displacements, such as support-seat gaps and girder misalignments, in the aftermath of earthquakes. A novel marker-free tracking method (T-SAM) is introduced for dynamic monitoring, allowing displacement tracking from video without the installation of physical markers. A complementary static image analysis workflow (termed YSAMM) combines YOLOv8, SAM, and geometric calibration to quantify residual displacements and rotations between bridge elements from single post-event images. The methodology was validated through shake-table experiments and case studies. In dynamic tests on a full-scale wood-frame structure, the marker-free T-SAM tracking achieved millimeter-level accuracy, matching the performance of conventional sensor and marker-based measurements even under varying lighting conditions. In static image analyses, the system measured girder seating lengths and rotations with centimeter-level precision, enabling direct comparison with code-based safety thresholds. Key results indicate that the vision-based approach can reliably detect subtle damage indicators. For example, T-SAM maintained correlation coefficients of around 0.97–0.99 with physical sensor data, and the combined YOLO–SAM technique could accurately determine support seat loss within a few centimeters of the ground truth. Overall, this thesis demonstrates that an inspector-guided, yet automated, vision system can significantly enhance post-earthquake bridge inspections. By rapidly providing quantitative measurements of damage, the developed framework and tool enable more objective, data-driven decisions on bridge safety and reopening, enhancing resilience in disaster response while reducing reliance on time-consuming manual methods.
Leila Katebi (Thu,) studied this question.