Vibration detection of stay-cable from low-quality CCTV images using deep-learning-based dehazing and semantic segmentation algorithms

Abstract

This study aims to monitor excessive vibrations in the stay-cables of long-span bridges, which can cause structural defects or discomfort to the public. Traditional methods like structural health monitoring (SHM) systems and computer vision techniques are impractical and costly for monitoring all stay-cables. Therefore, this study proposes a cost-effective solution using surveillance cameras (CCTV) commonly used for monitoring traffic conditions. Deep learning and computer vision techniques were used for semantic segmentation, dehazing, and tracking to address technical challenges in selecting feature points and image quality. Synthetic image generation was employed to obtain sufficient training images with pixel-wise annotations. The proposed framework was first validated using laboratory-scale experiments and applied to actual CCTV images collected under various environmental conditions. Parametric studies confirmed the efficiency of image synthesis and additional loss functions. The proposed method provides a viable alternative to monitor cable vibrations in long-span bridges using existing CCTV systems.