• CSCD核心中文核心科技核心
  • RCCSE(A+)公路运输高质量期刊T1
  • Ei CompendexScopusWJCI
  • EBSCOPж(AJ)JST
二维码

隧道建设(中英文) ›› 2025, Vol. 45 ›› Issue (8): 1459-1468.DOI: 10.3973/j.issn.2096-4498.2025.08.004

• 研究与探索 • 上一篇    下一篇

融合改进ORB特征点与LK-金字塔光流法的位移测量方法及工程应用

刘小林   

  1. (华南理工大学土木与交通学院, 广东 广州 510640)
  • 出版日期:2025-08-20 发布日期:2025-08-20
  • 作者简介:刘小林(1982—),男,湖南常宁人,华南理工大学土木水利专业在读博士,正高级工程师,主要从事结构监测方向研究工作。E-mail: 675671691@qq.com。

Displacement Measurement Method Integrating Improved Oriented FAST and Rotated BRIEF Feature Points With Lucas-Kanade Pyramid Optical Flow and Its Engineering Application

LIU Xiaolin   

  1.  (School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510640, Guangdong, China)
  • Online:2025-08-20 Published:2025-08-20

摘要: 为解决光流法在工程现场应用中易受光照变化干扰导致特征点检测失准、匹配偏差及跟踪成功率低的难题,以提升结构位移识别精度为目标,提出融合改进ORB特征点与LK-金字塔光流算法的结构位移测量方法。该方法包含2层模块: 首先,在特征检测层,构建FAST特征点自适应阈值模型,通过动态调整阈值解决固定阈值易引发的特征点误检与误配问题; 结合Hessian矩阵边缘响应值分析实现边缘伪特征点的精准剔除,并引入改进型非极大值抑制算法对特征点进行空间均匀化筛选,提升特征点集的稳定性与空间分布合理性。其次,在位移解算层,依托优化后的特征点集,设计多尺度分层金字塔光流追踪架构,采用由粗到精的迭代计算策略逐层优化运动矢量,有效克服大尺度位移工况下的追踪失效问题。研究结果表明: 1)该方法在复杂光照下的特征点匹配正确率达到76.0%,较SURF、SIFT和传统ORB等算法的正确率和计算效率均有提升; 2)位移结果误差小于4%,且在保持与DIC算法精度相当的情况下,效率更高; 3)盾构井施工过程的现场实测结果表明,该方法与人工测量的结果较为吻合,两者位移差异在±1.5 mm以内,结果可靠。

关键词: 光流法, 位移监测, ORB特征点, LK-金字塔, 盾构井

Abstract: The optical flow method applied at engineering sites is easily affected by illumination changes, leading to inaccurate feature point detection, matching errors, and a low tracking success rate. To improve the accuracy of structural displacement identification, a structural displacement measurement method integrating improved oriented FAST and rotated BRIEF (ORB) feature points with the Lucas-Kanade pyramid optical flow algorithm is proposed. The method comprises two modules: the feature detection layer and the displacement solution layer. At the feature detection layer, an adaptive threshold model for FAST feature points is constructed. By dynamically adjusting the threshold, the false detection and mismatching of feature points caused by fixed thresholds are reduced. In addition, by analyzing the edge response values of the Hessian matrix, edge pseudo-feature points are precisely removed. An improved non-maximum suppression algorithm is introduced to perform spatially uniform filtering of feature points, thereby enhancing the stability and spatial distribution rationality of the feature point set. At the displacement solution layer, based on the optimized feature point set, a multi-scale hierarchical pyramid optical flow tracking architecture is designed. An iterative computation strategy from coarse to fine is employed to optimize motion vectors layer by layer, effectively addressing tracking failures under large-scale displacement conditions. Experimental results show that: (1) under complex illumination conditions, the feature point matching accuracy of the proposed method reaches 76.0%, outperforming algorithms such as speeded-up robust features, scale-invariant feature transform, and traditional ORB in both accuracy and computational efficiency; and (2) the displacement error is less than 4%. While maintaining accuracy comparable to that of the digital image correlation algorithm, the proposed method is more efficient. (3) Field measurements during shield-launching shaft construction indicate that the results of this method align well with manual measurements, with displacement differences within ±1.5 mm, demonstrating its reliability.

Key words: optical flow method, displacement monitoring, oriented FAST and rotated BRIEF feature points, Lucas-Kanade pyramid, shield-launching shafts