Multi-level deformation behavior monitoring of flexural structures via vision-based continuous boundary tracking: proof-of-concept study

• Published in: Measurement : journal of the International Measurement Confederation 2022-05, Vol.194, p.111031, Article 111031
• Main Authors: Shan, Jiazeng, Liu, Yuwen, Cui, Xiaoxuan, Wu, Hao, Loong, Cheng Ning, Wei, Zhihua
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Bending curvature; Circular plates; Computer vision; Curvature; Deep neural network; Deformation; Feasibility; Impact loads; Measurement; Neural networks; Reinforced concrete; Rigid-body dynamics; Salient object detection; Structural health monitoring; Vibration; Vibration measurement; Vision-based measurement
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2022.111031

•This study proposes the concept of measuring multi-level deformation behavior of flexural structures via the extraction of continuous structural boundaries from a computer vision technique.•This study explores the feasibility of using a salient object detection method to measure the curvature profile of a structure from the image frames of a video captured from a digital camera.•This study presents a pre-trained deep neural network that can accurately segment a salient object in an image for curvature measurement.•A novel approach for determining the curvature of a structure based on the geometrical information at its boundary is proposed.•The proposed technique is validated via two experiments and shown to be accurate for measuring the displacement and curvature profile of a structure. The concept of measuring the multi-level deformation behavior of flexural structures via the extraction of continuous structural boundaries using a computer vision technique is proposed. The feasibility of using a salient-object-detection method to estimate the deformation and curvature profile of target structures from the image frames of a video recording structural vibrations is investigated. A framework is proposed for performing this salient-object-detection-based vibration measurement technique via the aid of a pre-trained deep neural network. A method for determining the curvature estimated from the boundary extracted from the deep neural network is then introduced. The accuracy of the proposed technique is validated via two experiments. The first experiment measures the curvature of a semi-circular plate under rigid body motions. The second experiment tracks the deformation of reinforced concrete beams under impact loads. Both experiments verify that the proposed method is feasible for accurately measuring the vibration profile of the target structure.