Abstract
Moore’s Law has driven key advancements in transistor and circuit technology, leading to innovations like 3D printers, compact SBCs, CPUs, and GPUs, greatly impacting robotics and computer science. Alongside, drone and Unmanned Aerial System (UAS) technologies have evolved, benefitting from Deep Learning and Computer Vision, thus enabling their application in a range of tasks from surveillance to inspections in remote or inaccessible areas. Despite these advancements, UASs face challenges in maneuvering through confined spaces, where there is a high risk of damage, compromising inspection data quality. This limitation undermines the performance of even the most sophisticated algorithms in such environments. In this thesis, the author introduces an integrated hardware and software framework for extracting precise, millimeter-scale measurements during infrastructural inspections in narrow, remote, or inaccessible areas. Utilizing stereoscopic images from a monocular camera mounted on a cascading, bi-directional rack-and-pinion actuator, the system employs photogrammetry and Structure from Motion (SfM) techniques to generate a detailed point cloud of the inspection target. This point cloud is then analyzed using an advanced 3D hierarchical segmentation model, enabling accurate measurements and forming a comprehensive geometric profile of the target. This framework boasts several advantages, including the ability to inspect the backside of a target using only its frontal view, developing a precise segmentation model without real-world data, and adjusting the UAS’s position to the most optimal location, irrespective of its initial landing site. Moreover, the thesis introduces an air-land hybrid locomotion system, allowing the UAS to maneuver efficiently in confined spaces without requiring additional actuators, relying instead on the UAS’s primary actuator. The hybrid locomotion system allows for the UAS to transit between operational mode through passive engagement mechanisms. This approach minimizes additional components, ensuring the UAS’s compactness. The dissertation proposes a comprehensive solution to transform traditional, labor-intensive inspection methods, known for their high risk, inefficiency, and cost. Demonstrated through the inspection of a rail viaduct bearing, the framework’s practicality and efficacy are highlighted. Its adaptability for other enclosed inspection applications is underscored, particularly those requiring a small form factor, automation, and precise measurements. The thesis concludes with a presentation of experimental results and real-world trials, emphasizing the potential of this research for real-world applications.