基于深度光流的结构简谐位移响应监测方法

Structural Harmonic Displacement Response Monitoring Method Based on Deep Optical Flow

摘要: 传统接触式位移测量方法存在对场地要求高、现场施工周期长、经济性差等问题。鉴于此，本文通过图像预处理、相机标定、畸变校正，并结合稠密光流估计网络，提出一种基于深度学习的结构简谐位移监测方法。结果表明，所提出方法对规则正弦振动条件下的结构简谐位移响应时程具有较好的识别效果，视觉位移时程与激光位移计结果具有较高一致性。视觉位移时程相对于激光位移计基准结果的RMSE为0.208~0.246 mm，峰值误差为0.692~0.780 mm，相对误差为13.84%~15.60%。

Abstract: Traditional contact-based displacement measurement methods suffer from stringent site requirements, long on-site installation periods, and poor cost-effectiveness. To address these limitations, this study proposes a deep learning-based monitoring method for structural harmonic displacement response through image preprocessing, camera calibration, distortion correction, and dense optical flow estimation. The results show that the proposed method achieves good identification performance for structural harmonic displacement responses under regular sinusoidal excitation, and the vision-based displacement measurements are in good agreement with those obtained from the laser displacement meter. The RMSE between the vision-based displacement time histories and the laser-based reference results ranges from 0.208 to 0.246 mm, the peak error ranges from 0.692 to 0.780 mm, and the relative error ranges from 13.84% to 15.60%.