Assessment of Defects in Steel Structures Using UAVs and Image-Based 3D Reconstruction

The resolution of visual image and the accuracy of 3D reconstruction of complex structural geometries using a camera carried by drones are important indicators to adapt the methodology as a quantitative bridge inspection tool. This thesis aims to determine and optimize the accuracy of defect and dimensional measurements using 3D reconstruction and image processing tools for multi-girder steel bridges. Additionally, a passive visual sensor is developed that amplified the strain in the base structure, making non-visual plastic deformations able to be captured visually. A safe data collection route is defined considering the constraints of diaphragms and spacing between girders with the objective of creating a high-resolution 3D reconstruction of complex girder geometry. Qualitative accuracy was explored in the 3D model to determine the camera angles, elevations and working distance. The lowest percent error achieved in this study for 3D quantitative results using ad hoc measurement on overall accuracy in the girder geometry was 5.92% using 332 images. Using the proposed systematic methodology as a guideline in a more realistic setup, a percent error of 32.82% was achieved using 43 images, which significantly reduces the flight time without sacrificing from accuracy.