Target-free vision-based vibration measurement of bridge cables using high-resolution segmentation and complex phase analysis

Accurate cable vibration identification is fundamental to cable force estimation and structural health monitoring of long-span bridges. However, existing target-free vision-based methods remain vulnerable to complex backgrounds, resolution degradation, and unstable phase responses from slender cable regions. To address these issues, this study proposes a target-free cable vibration measurement framework, termed HCPA, which integrates high-resolution cable background segmentation (HCBS) with improved complex multi-scale phase analysis (CPA). HCBS employs a Swin Transformer encoder, a multi-scale atrous spatial pyramid pooling decoder, and an auxiliary decoder to extract cable regions from high-resolution bridge images with complex backgrounds. On this basis, CPA decomposes the segmented video into multi-scale and multi-orientation phase components, while amplitude-weighted spatial pooling and variance-weighted sub-band fusion are introduced to strengthen vibration-sensitive signals and suppress noise-dominated responses. A high-resolution cable dataset containing steel wires and bridge cables was constructed to train the segmentation model. Compared with the benchmark Sas-Net model, HCBS improves mIoU, mDice, mP, and mR by 1.66%, 0.87%, 1.01%, and 0.70%, respectively. In laboratory tests, HCPA achieves average absolute errors of 0.27%, 0.23%, and 0.31% for the first three modal frequencies, clearly outperforming the conventional phase-based video motion method. Resolution analysis shows that reliable frequency extraction strongly depends on preserving high-resolution cable information. In a field test on a suspension bridge, the maximum frequency identification error of HCPA is 1.95% relative to contact accelerometer measurements. Compared with representative vision-based methods, including BSS, KLT, PVM, LTA, and RAFT, HCPA provides a favorable balance between frequency identification accuracy and computational efficiency. These results demonstrate that HCPA provides a robust and practical solution for target-free cable vibration monitoring in real bridge environments.