In the construction of buildings, the assessment of steel platforms via pre-compression testing and the subsequent monitoring of deflection at critical points is imperative for ensuring the safety of the construction process. However, the current non-contact visual measurement systems utilized for deflection detection are susceptible to heat haze and unsteady vibrations, affecting the collected data. To address these challenges, we aim to develop an automatic denoising framework. This framework is grounded in a three-stage multiscale clustering algorithm, designed to eliminate anomalous data obtained from visual systems to get true and accurate deflection. It integrates DBSCAN and k-means algorithms and denoises the data from both macro-whole and micro-local perspectives. Additionally, the k-distance optimization algorithm was utilized to achieve optimal parameters. The experimental results demonstrate that this method exhibits an average error of 4.56%, compared to laser displacement sensor measurements, thereby validating its feasibility.

中文翻译：

---

基于数字图像相关和三阶段多尺度聚类算法的室外钢结构挠度自动测量

在建筑物的施工中，通过预压缩测试对钢平台进行评估以及随后对关键点的变形进行监测对于确保施工过程的安全至关重要。然而，当前用于偏转检测的非接触式视觉测量系统容易受到热雾和不稳定振动的影响，影响收集的数据。为了应对这些挑战，我们的目标是开发一个自动去噪框架。该框架基于三阶段多尺度聚类算法，旨在消除从视觉系统获得的异常数据，以获得真实准确的偏转。它集成了DBSCAN和k-means算法，从宏观整体和微观局部两个角度对数据进行去噪。此外，还利用k距离优化算法来获得最佳参数。实验结果表明，与激光位移传感器测量相比，该方法的平均误差为4.56%，验证了其可行性。