Vision-based robust structural dynamic monitoring and modal identification using stationary and drone-based imaging systems.

Structural health monitoring (SHM) plays a critical role in assessing and managing the condition of the civil infrastructure and prevents sudden closures or catastrophic failure of the structures by timely detection of ongoing degradations. Vision-based SHM has emerged as a promising non-contact alternative to conventional SHM approaches for estimating structural response and modal parameters. However, vision-based system identification comes with several practical challenges due to low sub-pixel structural motion, hardware constraints, environmental noise, and platform-induced disturbances in mobile sensing platform systems. To address these challenges, this thesis proposes innovative non-contact, vision-based SHM methodologies for the extraction of dynamic characteristics and validates experimentally on laboratory and field conditions using both stationary and drone-mounted cameras. This thesis investigates phase-based motion magnification (PMM) and proposes a hybrid framework to extract dynamic parameters and to study the influence of low-frame-rate imaging on PMM in motion-magnified video. Subsequently, a decentralized vision-based modal identification technique is proposed by combining multi-camera fields of view to extract high-density spatiotemporal information, utilizing fiducial markers as inexpensive virtual sensors to recover three-dimensional structural vibrations and to estimate modal parameters for full-field vision analysis without deploying stereo camera systems. To address accessibility challenges associated with large civil infrastructure, a drone-based vision-driven framework is developed to extract absolute displacement responses and dynamic parameters via ego-motion compensation and homography-based mapping. The feasibility of drone-based modal identification is evaluated through laboratory and field experiments, highlighting practical limitations in the estimation of the dynamic measurements under low signal-to-noise ratio conditions. Finally, for dense displacement extraction and modal parameter estimation, a deep learning-based tracking framework is proposed for long-duration video sequences using a sliding-window and long-context weights strategy under limited GPU memory constraints. The proposed methodologies are validated through laboratory-scale studies and field deployments, demonstrating strong agreement between vision-based estimates and conventional contact sensors. Overall, this thesis establishes scalable, robust vision-based pipelines for non-contact structural vibration monitoring, advancing the practical implementation of camera-based SHM systems for real-world civil infrastructure.

Name thesis:degree_name

Ph D

Discipline thesis:degree_discipline

Civil and Environmental Engineering

Grantor dc:publisher

The University of Western Ontario

Year dc:date.issued

2026

Author dc:creator

• Mittal, Shivank

Advisor dc:contributor.advisor

• Sadhu, Ayan

• Structural health monitoring
• computer vision
• fiducial marker
• unmanned aerial vehicle
• vision-based dynamic measurement
• non-contact sensing
• modal identification
• deep learning
• phase-based motion magnification
• blind source separation
• correlation tracker
• Cotracker3

Statement dc:rights

• Attribution-NonCommercial-NoDerivatives 4.0 International

Language dc:language.iso

en

Handle dc:identifier.uri

https://hdl.handle.net/20.500.14721/39557

OAI identifier oai:identifier

oai:uwo.scholaris.ca:20.500.14721/39557