Robust Subpixel Centerline Extraction of Dense Laser Stripes Using PAE-Constrained Anisotropic Gaussian Modeling

Subpixel centerline extraction of dense laser stripes is critical for high-precision 3-D measurement systems, yet traditional methods suffer from significant localization errors under noise, blur, and closely spaced stripe conditions - challenges that severely limit measurement accuracy in industrial metrology. This article presents a novel extraction method based on the PAE-constrained anisotropic Gaussian framework (PAE-AGK) to achieve robust subpixel accuracy. Unlike conventional approaches that assume intensity continuity, the partial area effect (PAE) treats pixel intensity as area-weighted averages across edges, enabling precise estimation of subpixel edge position, orientation, curvature, and contrast under adverse conditions. These PAE-derived geometric features serve as physically meaningful priors to constrain anisotropic Gaussian kernel construction, eliminating iterative parameter optimization while ensuring geometric-photometric consistency. The method operates in three stages: Stage I - PAE-based local edge model and stripe geometry estimation; Stage II - anisotropic Gaussian kernel construction with parameters fully determined from Stage I; Stage III - coarse localization via scale-normalized second-order response and analytic subpixel refinement along the normal direction. Comprehensive experimental validation demonstrates superior performance over traditional methods, achieving root-mean-square error (RMSE) of 0.0131-0.0289 pixels under noise-free conditions and maintaining 0.0135-0.0992 pixels under noisy conditions - substantially outperforming classical algorithms including Steger, gray-gravity, and geometric-center methods. The method successfully extracts centerlines in dense scenarios with an interstripe distance as low as 12 pixels. Qualitative evaluation on diverse real-world materials validates practical applicability for precision measurement and 3-D reconstruction applications.