Geometry-preserving adaptive multi-exposure fusion for robust structured-light rail profile measurement under extreme illumination

Structured-light-based rail profile measurement is highly susceptible to illumination variations in outdoor railway environments, where strong quasi-specular reflections, sensor saturation, and low signal-to-noise ratios frequently distort laser stripe geometry and degrade three-dimensional reconstruction accuracy. Although multi-exposure fusion (MEF) effectively extends the imaging dynamic range, existing methods are primarily designed for perceptual image enhancement and rarely consider the geometric fidelity required for precision optical metrology. To address this limitation, this paper proposes a geometry-preserving structured-light measurement framework that integrates illumination-aware adaptive exposure acquisition, stripe-width-guided multi-exposure fusion, robust subpixel centerline extraction, and dual-stage denoising. Specifically, an adaptive exposure scheduling strategy is first developed to dynamically generate an optimal exposure sequence according to ambient illumination, ensuring reliable acquisition of stripe information across both saturated and underexposed regions. A structure-preserving fusion method is then proposed by introducing laser stripe width consistency as a pixel-wise reliability prior, enabling radiometric enhancement while maintaining subpixel geometric integrity of the laser stripe. Furthermore, an improved Steger-based centerline extraction algorithm combined with polynomial-constrained refinement and region-growing-based outlier suppression is employed to improve robustness against saturation-induced distortion and ambient-light interference. Experimental results demonstrate that the proposed method maintains centerline extraction deviations within 3 pixels over illumination levels ranging from 213 to 25,327 lx and achieves absolute rail wear measurement errors below 0.09 mm. Under extreme outdoor illumination of 57,909 lx, the method remains highly stable, yielding repeatability standard deviations of 0.025 mm and 0.038 mm for lateral and vertical wear measurements, respectively. Comparison with a contact profilometer further shows a maximum absolute error below 0.085 mm with negligible degradation in repeatability under strong sunlight. These results demonstrate that incorporating geometric constraints into multi-exposure fusion substantially enhances the robustness and measurement accuracy of structured-light vision systems operating under uncontrolled illumination, providing an effective solution for high-precision outdoor optical metrology.