为解决隧道环境下视觉位移监测算法性能边界不明确、失效机理未厘清及预警判据缺失的问题，以相位相关法（phase-only correlation，POC）和快速归一化互相关法（fast NCC，FNCC）2类代表性算法为研究对象，结合纯圆形、编码圆形、棋盘格、散斑4类标志物，构建以峰值相关性分数（peak correlation score, PCS）和像素位移误差为核心的“算法原理-标志物特征-环境干扰”协同量化分析框架。研究设计并完成10 024组涵盖噪声、光照、遮挡、旋转、缩放、多因素耦合、时序动态干扰与图像分辨率衰减的多维度受控试验，建立稳定工作、性能敏感、临界失效、完全失效的4级性能评价标准，定量标定不同算法-标志物组合的性能边界与失效临界点，并通过缩尺物理模型试验验证优选组合的工程适用性。结果表明： 1）POC与FNCC算法本征性能差异显著，POC计算效率为FNCC的1.5~2.0倍，但精度偏低且误差呈长尾分布，FNCC测量精度更优，复杂纹理标志物可有效提升2类算法匹配可靠性； 2）从机理层面揭示了POC算法对尺度变化的固有脆弱性，以及FNCC算法在局部畸变下的高PCS-低精度隐性失效特征，明确了标志物纹理特征与算法原理的耦合适配规律； 3）整像素匹配算法与亚像素精化方法需遵循形态适配原则，失配将导致测量精度急剧劣化； 4）采用水下隧道缩尺模型试验验证了POC+编码圆形标志物组合在隧道局部变形-复位全过程监测中的工程适用性，该组合实现了±1.5 mm微幅波动至17 mm级大变形的宽动态范围精准捕捉。
To address the issues of unclear performance boundaries, unclarified failure mechanisms, and lack of early warning criteria for visual displacement monitoring algorithms in tunnel environments, two representative algorithms—phase-only correlation (POC) and fast normalized cross-correlation (FNCC)—are taken as research objects to construct a collaborative quantitative analysis framework of “algorithm principle, marker feature, and environmental interference” based on four types of markers, namely pure circle, coded circle, checkerboard, and speckle, with peak correlation score (PCS) and pixel displacement error as core indicators. A total of 10 024 multi-dimensional controlled experiments are designed and completed, covering noise, illumination, occlusion, rotation, scaling, multi-factor coupling, temporal dynamic interference, and image resolution attenuation. A four-level performance evaluation standard, including stable operation, performance sensitivity, critical failure, and complete failure, is established, and the performance boundaries and failure critical points of different algorithm-marker combinations are quantitatively calibrated. The engineering applicability of the optimal combination is verified through a scale physical model test. The results show that: (1) POC and FNCC exhibit significant differences in intrinsic performance: POC has 1.5 to 2.0 times the computational efficiency of FNCC but lower accuracy with a long-tail error distribution, while FNCC achieves higher measurement accuracy, and complex texture markers effectively improve the matching reliability of both algorithms. (2) The inherent vulnerability of POC to scale changes and the high PCS but low accuracy hidden failure characteristic of FNCC under local distortion are revealed at the mechanism level, clarifying the coupling adaptation law between marker texture features and algorithm principles. (3) The integer-pixel matching algorithm and sub-pixel refinement method must follow the principle of morphological adaptation, as mismatch will lead to sharp deterioration of measurement accuracy. (4) The scale model test of an underwater tunnel verifies the engineering applicability of the POC plus coded circular marker combination in the whole process monitoring of local deformation and reset of the tunnel, demonstrating its ability to accurately capture a wide dynamic range from ±1.5 mm micro-amplitude fluctuations to 17 mm large deformations.
