Accurate full-field vibration measurement through computer vision is crucial for dynamic analysis and structural health monitoring, providing intuitive insights into dynamic behaviour. Recently, phase-based motion estimation (PME) has advanced rapidly due to its high spatial sensing capability and sub-pixel accuracy. However, inappropriate filter parameters can hinder accurate vibration estimation, and the complexity of structures makes it challenging for a single optimal filter to effectively match all measurement points. This study proposed a novel data-driven method for full-field PME that utilizes knowledge transfer-assisted optimal log-Gabor filters. The optimisation process adaptively balances self-optimisation and knowledge transfer among similar tasks based on iterative experience. Moreover, a morphological filling technique identifies edge pixels with local amplitudes higher than their surroundings, creating a closed-loop set, which is then iteratively filled using a dilation operation. An active pixel selection strategy is developed to define a threshold between active and inactive pixels based on pixel value variations. Overall, each pixel or region is matched with an optimal filter to yield more accurate vibration signals, which are presented through both instantaneous and frequency-specific operational deflection shapes. The proposed method is validated with a proof-of-principle test on a steel plate and applied to lab-scale flexible cable and real-size SSRM tests for full-field vibration estimation and dynamic analysis.

中文翻译：

---

使用许多知识转移辅助的最佳 log-Gabor 滤波器进行基于相位的全场振动测量和可视化


通过计算机视觉进行准确的全场振动测量对于动态分析和结构健康监测至关重要，可提供对动态行为的直观见解。近年来，基于相位的运动估计 （PME） 由于其高空间感知能力和亚像素精度而迅速发展。然而，不适当的滤波器参数会阻碍准确的振动估计，并且结构的复杂性使得单个最佳滤波器难以有效匹配所有测量点。本研究提出了一种新的全场 PME 数据驱动方法，该方法利用知识迁移辅助的最优对数 Gabor 滤波器。优化过程根据迭代经验自适应地平衡相似任务之间的自我优化和知识转移。此外，形态填充技术可识别局部振幅高于其周围环境的边缘像素，从而创建一个闭环集，然后使用膨胀操作迭代填充该像素。开发了一种活动像素选择策略，以根据像素值变化定义活动像素和非活动像素之间的阈值。总体而言，每个像素或区域都与最佳滤波器匹配，以产生更准确的振动信号，这些信号通过瞬时和特定频率的工作偏转形状呈现。所提出的方法通过在钢板上进行原理验证测试进行了验证，并应用于实验室规模的柔性电缆和真实尺寸的 SSRM 测试，以进行全场振动估计和动态分析。