Comprehensive mitigation method for atmospheric turbulence errors in Vision-Based displacement monitoring of long-span bridges

Vision-based displacement measurement techniques for flexible structural bodies like long-span bridges have gained wide popularity in recent years. However, due to the long monitoring distance and large field of view, the visual monitoring system of long-span bridges is easily affected by atmospheric turbulence (AT). Temperature-induced air density fluctuations along the monitoring path create refractive index variations, resulting in degradation in image quality and measurement errors in displacement obtained from distorted images. In this paper, starting from the aspect of enhancing the feature information of degraded images, a comprehensive mitigation method of AT errors for visual displacement monitoring of long-span bridges is proposed. Firstly, the imaging tilt and blur caused by AT are decoupled utilizing the optics of imaging through turbulence. Subsequently, global tilt correction is achieved through optical flow alignment, followed by localized distortion mitigation using steerable pyramid decomposition and temporal phase filtering to address residual distortion. Meanwhile, a turbulence blur restoration transformer (TBRformer) model based on encoder-decoder architecture is designed to restore sharp images. By introducing multi-scale linear attention and an optimized feedforward neural network, the efficiency and the ability to restore the target feature details of the image are enhanced. Finally, the restored image sequence is transferred to the structural target displacement extraction algorithm, and a more accurate fitting vibration displacement curve is obtained through sub-pixel interpolation. The displacement measurement results of laboratory tests and field tests of a long-span bridge show that the proposed method can reduce the measurement error and improve the robustness of visual target tracking.