Streamlined particle filtering of phase-based magnified videos for quantified operational deflection shapes. (1st September 2022)

• Record Type: Journal Article
• Authors: Valente, Nicholas A.; Sarrafi, Aral; Mao, Zhu; Niezrecki, Christopher
• Is Part Of: Mechanical systems and signal processing. Volume 177(2022)
• Publication Date: 2022-09-01
• Subjects: Phase-based video processing -- Computer vision -- Particle filter -- Data clustering -- Unsupervised learning -- Structural dynamics identification -- Total variation denoising; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2022.109233
• Languages: English
• ISSNs: 0888-3270

Abstract

Abstract: Non-contact optical measurements are commonly used in industrial and research domains to obtain displacement measurements. This can be attributed to their noninvasive advantages over traditional instrumentation approaches. Phase-based motion estimation (PME) and magnification (PMM) are targetless methods that have been utilized recently to extract qualitative data from structures via experimental modal analysis (EMA) and operational modal analysis (OMA). Transforming the motion-magnified sequence of images into quantified operating deflection shape (ODS) vectors is currently being conducted via edge detection. Although effective, these methods require human supervision and interference; such that, accurate characteristics of the structure are guaranteed. Within this study, a new hybrid computer vision approach is introduced to extract the quantified ODS vectors from motion-magnified images with minimal human supervision. The particle filter (PF) point tracking method is utilized to track the desired feature points in the motion-magnified sequence of images. Moreover, the k-means clustering method is employed as an unsupervised learning approach to perform the segmentation of the particles and assign them to specific feature points in the motion-magnified sequence of images. Total variation denoising is used to smooth the motion-magnified artifacts, which improves ODS vector extraction and provides a robust outlier removal. The results show that the cluster centers can be applied to estimate the ODS vectors, and the performance of the proposed methodology is evaluated experimentally on a lab-scale cantilever beam.