Advancements in 3D displacement measurement for civil Structures: A monocular vision approach with moving cameras
Qilin Li, Yanda Shao, Ling Li, Jun Li, Hong Hao
Abstract
• A computer vision method for 3D displacement measurement using a single moving camera. • Monocular depth estimation is fine-tuned with a synthetic civil structure dataset. • Laboratory tests show accurate vibration measurement of beam structures using a UAV. This study proposes a novel computer vision approach for measuring three-dimensional (3D) vibration displacements in civil structures using a single, moving camera. The proposed approach, named monocular vision moving camera (MVMC), effectively segregates the 3D displacement measurement into in-plane and out-of-plane components, addressing each within a cohesive framework. The in-plane displacements are measured based on an advanced point tracking method, whereas the out-of-plane displacements are obtained using a deep learning based depth prediction network. This network is refined with a synthetic structural depth dataset to augment the measurement accuracy. To eliminate camera motion, the structure is assumed to contain a stationary area, in which point registration techniques are used to differentiate between the movements of the structure and the camera. Laboratory tests on cantilever beams validate the MVMC’s capability to accurately measure full-field 3D vibration displacements. The results exhibit cross-correlation coefficients exceeding 0.96 and relative errors under 30% when benchmarked against laser displacement sensors. Comprehensive ablation studies further assess the critical components of MVMC, offering insightful recommendations for the future integration of such advanced vision-based techniques in 3D displacement measurement.