Work

A novel machine vision method for fast real-time low-frequency vibration measurement of engineering structures

Year: 2025

Type: article

Abstract: • Fast real-time measurement for structure vibration is realized. • A new method is able to enhance measuring generalization capability and accuracy. • Prototype of the measuring system is developed. • The new method finds a balance between calculation time and measuring accuracy. Deep learning-based object detectors exhibit strong generalization capabilities and have been widely applied in the field of vibration measurement. However, errors often occur in the generation of candidate boxes during the deep learning processing, leading to a decrease in the measurement accuracy. More importantly, high computational cost and slow inference speed hinder the efficiency of the vision methods in practical applications. To address these issues, this work proposes a new vision method, named as the Fast Real-Time Low-Frequency Vibration Measurement (FRLF-VM), to measure small structural vibration displacements. In this new method, the keypoint localization process is divided into two stages. In the first stage, the deep learning is used to localize the target region; while in the second stage, a new keypoint detection algorithm is developed by combining the data-driven methods to achieve precise localization of the target points. Furthermore, after completing the coarse localization in the first stage, the system automatically cancels the subsequent computations of the candidate boxes, which effectively reduces the computational costs. To further establish a temporal correlation between video frames, an efficient keypoint tracking algorithm is proposed. Lastly, using the binocular vision technology, small structural vibration can be precisely measured. Based on this new method, a prototype (including hardware and software) is designed, manufactured and evaluated. Experimental results demonstrate the superiority of the proposed method over existing popular techniques by achieving subpixel-level accuracy. As a result, the FRLF-VM method provides a new solution for real-time vibration measurement.

Source: Applied Ocean Research

DOI: 10.1016/j.apor.2025.104869

Authors: Yuansheng Cheng, Zhe Tian, Donghong Ning, Atilla Incecik, Zhixiong Li, Lu Liu

Institutions: Opole University of Technology, Ocean University of China, University of Strathclyde

Language: English

Topic: Structural Health Monitoring Techniques

Subfield: Civil and Structural Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $2,950