Video-Based Displacement, Velocity and Acceleration Measurement of Structural Elements Using Object-Tracking Algorithms

Abstract

Civil structures experience forces from external events, operational loads, and environmental conditions that induce vibrations and can lead to deformation that can compromise their integrity and safety. Precise computation of their displacement, velocity and acceleration aids in understanding their dynamic behaviour, detect failures early, optimise design and guarantee long-term durability. Traditional sensor-based approach, need physical contact, offer limited measurement points, are expensive, face accessibility issues, and lack scalability. These drawbacks prompt the need for non-contact, vision-based measurement techniques that offer flexibility, affordability, and comprehensive motion tracking. In this project, we propose a video-based method that utilises OpenCV based object tracking algorithms to measure kinematics such as displacement, velocity and acceleration of structural elements like steel cantilever beam, single degree of freedom pendulum and scaled portal frames. Several tracking algorithms such as Lucas-Kanade optical flow, CSRT (Channel and Spatial Reliability Tracker), MIL (Multiple Instance Learning), and KCF (Kernelized Correlation Filters) were leveraged to extract exact motion data over time by processing vibration videos of these structural elements. The videos are pre-processed to identify the structural element as the region of interest (ROI). Tracker-based approaches (CSRT, MIL, KCF) and Lucas-Kanade optical flow were utilised and displacement, velocity and acceleration were derived. The workflow involves frame extraction, grayscale conversion, and tracking algorithms, where Lucas-Kanade estimates vectors at the pixel level, while feature-based trackers record the trajectory of the beam. Plots of displacement vs time, velocity vs time and acceleration vs time depict vibration patterns and natural frequencies. The Fast Fourier Transforms (FFT) derived from both accelerometer data and video-based measurements are compared to validate the accuracy of the video-based analysis. By combining state-of-the art computer vision methods with structural analysis, this work represents a possible step toward modern, non-invasive methods for evaluating dynamic structural reactions. The comparative use of multiple tracking methods improves the approach’s reliability and applicability, paving the way for broader implementation in engineering processes. This approach can be leveraged to bigger structures. This method can be used in subsequent research to remotely examine the structural health of large-scale structures like buildings and bridges. Accuracy and efficiency can be further increased by incorporating deep learning for improved feature tracking and real-time processing.