Development of structural assessment approaches for bridges using vision based unmanned aerial systems

Abstract

This thesis aims to advance remote sensing techniques for bridge structures using Unmanned Aerial Systems (UAS) to (1) enhance the bridge inspection process particularly for crack detection and (2) to enhance the bridge assessment process by utilising UAS to measure bridge displacements.

To enhance the bridge inspection process, the study is trying to address the challenges associated with detecting and recording bridge defects during an inspection, and in particular, the detection of cracks. Crack detection is crucial because cracks are the most common and early indicator of material degradation or structural failure. Currently, the challenges in the prevailing approach include aspects such as the cost of access equipment, the tedious nature of manually recording cracks and inspector subjectivity to mention a few. To address these challenges research into the use of UAS and image processing techniques for bridge applications has risen significantly. Unfortunately, in the field, it is still challenging to detect cracks accurately and automatically due to noise in the images which a computer algorithm can perceive as cracks. Further, to properly understand the context of the cracking, individual images obtained using UAS need to be stitched into a panorama to recover the entire bridge element under consideration. However, image stitching where camera translates is not trivial and currently requires specialist software and equipment. In this study, taking a leaf from the visual cues from natural eyesight, a technique for crack detection is developed that improves the detection of cracks in noisy images obtained from portable platforms such as UAS. Further, a method for stitching images where the camera centre is allowed to translate is developed that improves the quality of the panorama without the need for specialist software or equipment.

UAS have also been proposed for use in bridge assessments for remote bridge displacement measurement which has the advantage of not fixing sensors physically to a bridge. Further, by hovering in flight, a UAS can ensure that the sensor is positioned in the most advantageous position. The main challenge of using UAS for bridge displacement measurement is that both the UAS and the bridge are in motion. This requires then that stable feature points can be seen by the UAS to determine its own motion, as only then can it reliably observe bridge motion. Achieving this requires overcoming two key challenges: (i) Maintaining features between frames. Unfortunately, in the field, finding and maintaining feature points during image tracking is difficult. Further, stable features maybe lost due to environmental changes and/or temporary occlusion by an object entering the scene such as a bird or a leaf. (ii) The absence of stationary objects close to the bridge mid-span. To measure bridge displacement to submillimetre accuracy typically requires the measuring camera to be zoomed in closely on the mid-span. This means a small field of view (FOV) around the mid-span, where (by the nature of the structure) there are rarely stationary objects. So, in essence a wide FOV is favourable for stabilisation, a zoomed in view, that is, narrow FOV is favourable for measuring displacement, representing two competing interests. Therefore, in this study (i) is addressed by proposing a technique to improve the robustness of UAS stabilisation techniques that requires only three feature points at a time. (ii) is addressed by developing an approach that utilises two different cameras that do not need to have an overlapping FOV. One wide angle camera for UAS stabilisation and one zoom camera to track displacement at a region of interest on a bridge. With this arrangement the need to have a stabilising feature in the field of view of the measuring camera is overcome and opens further benefits such as the ability to relate the measured displacement to the position of the load causing it.
These advances for both crack detection and displacement measurement developed in this study, contribute to bringing closer the use of UAS to enhance the traditional bridge management process.

Date of Award	Dec 2023
Original language	English
Awarding Institution	Queen's University Belfast
Sponsors	Commonwealth Scholarship Commission in the UK
Supervisor	Su Taylor (Supervisor) & David Hester (Supervisor)