3D damage mapping and segmentation using neural radiance fields and advanced deep learning techniques
| dc.contributor.author | Kim, Geontae | |
| dc.contributor.examiningcommittee | Wu, Nan (Mechanical Engineering) | |
| dc.contributor.examiningcommittee | Fiorillo, Graziano (Civil Engineering) | |
| dc.contributor.supervisor | Cha, Youngjin | |
| dc.date.accessioned | 2024-07-26T20:19:50Z | |
| dc.date.available | 2024-07-26T20:19:50Z | |
| dc.date.issued | 2024-07-08 | |
| dc.date.submitted | 2024-07-12T22:31:23Z | en_US |
| dc.degree.discipline | Civil Engineering | |
| dc.degree.level | Master of Science (M.Sc.) | |
| dc.description.abstract | This thesis introduces a novel approach to Structural Health Monitoring (SHM) by integrating deep learning with advanced 3D reconstruction techniques, focusing on the efficient analysis of bridge structures for achieving digital twin in SHM. Traditional photogrammetry faces challenges in accurately reconstructing flat surfaces and rapidly assessing structural health. Addressing these issues, this research adopts the Nerfacto model from the Neural Radiance Fields (NeRF) within the Nerfstudio framework to enhance 3D reconstruction fidelity and utilizes strategically placed markers to improve camera pose accuracy. Additionally, the integration of the STRNet and Test Time Agumentation (TTA) significantly enhances crack detection capabilities. This approach allows for precise mapping of segmented cracks onto a 3D model of a bridge, offering a detailed and quantifiable assessment of structural damage. By combining these innovative technologies, the research provides a scalable, cost-effective solution for comprehensive structural assessments, paving the way for proactive maintenance strategies that ensure the longevity and safety of critical infrastructure. The integration of digital twin technology and detailed damage mapping in 3D also sets a new standard in the field, demonstrating substantial potential for future SHM applications. | |
| dc.description.note | October 2024 | |
| dc.identifier.uri | http://hdl.handle.net/1993/38338 | |
| dc.language.iso | eng | |
| dc.subject | SHM | |
| dc.subject | Damage detection | |
| dc.subject | Deep learning | |
| dc.subject | Crack segmentation | |
| dc.subject | 3D reconstruction | |
| dc.subject | Neural radiance fields | |
| dc.subject | NeRF | |
| dc.subject | Digital twin | |
| dc.title | 3D damage mapping and segmentation using neural radiance fields and advanced deep learning techniques | |
| local.subject.manitoba | no |