Over the past decade, computer vision-based measurement approaches have been proposed to achieve non-contact structural displacement measurement. In particular, phase-based optical flow (PBOF) provides accurate dynamic measurements but is limited to estimating small displacements. Conversely, template matching can measure large-scale displacements but suffers from pixel resolution constraints. This study proposes a novel complementary approach that synergizes these two methods, using optical flow to capture high-frequency components of structural movements and reducing drift error with measurements from template matching. First, the video is preprocessed to remove possible lens distortion and image noise. A foreground mask is then generated using the proposed intensity difference-based method to refine the selected region-of-interest (ROI). Subsequently, structural displacements are estimated using the proposed foreground-weighted normalized cross-correlation (FWNCC) algorithm. An improved phase-based optical flow (IPBOF), which is robust against background clutter and invariant to intrinsic phase wrapping issues, is then used to compute the velocity components. Finally, these two measurements are fused using a complementary filter to achieve sub-pixel level large-scale displacement estimation. The efficacy of the proposed approach was validated with a base-excited three-story building model in the laboratory. The results demonstrate the method’s sub-pixel level accuracy in drift-free large-scale displacement measurement and its robustness against background clutter. Limitations and future research directions are also discussed in the conclusion.

中文翻译：

---

使用互补策略进行基于视觉的稳健亚像素级位移测量


在过去的十年中，基于计算机视觉的测量方法被提出来实现非接触式结构位移测量。特别是，基于相位的光流（PBOF）提供精确的动态测量，但仅限于估计小位移。相反，模板匹配可以测量大规模位移，但受到像素分辨率的限制。这项研究提出了一种新颖的互补方法，可以协同这两种方法，使用光流捕获结构运动的高频分量，并通过模板匹配的测量来减少漂移误差。首先，对视频进行预处理，以消除可能的镜头失真和图像噪声。然后使用所提出的基于强度差的方法生成前景掩模，以细化所选的感兴趣区域（ROI）。随后，使用所提出的前景加权归一化互相关（FWNCC）算法来估计结构位移。然后使用改进的基于相位的光流（IPBOF）来计算速度分量，该光流对背景杂波具有鲁棒性并且对固有相位缠绕问题具有不变性。最后，使用互补滤波器融合这两个测量值，以实现亚像素级大规模位移估计。该方法的有效性在实验室中通过基极激励三层建筑模型得到了验证。结果证明了该方法在无漂移大规模位移测量中的亚像素级精度及其对背景杂波的鲁棒性。结论中还讨论了局限性和未来的研究方向。