Physically Based Polarization Reflection Separation for Laser Stripe Extraction on Highly Reflective Metal Surfaces

Polarization-based laser vision imaging offers high accuracy, noncontact sensing, and strong robustness to interference, making it promising for weld seam localization and intelligent welding. However, for planar filet-weld workpieces with highly reflective metallic surfaces, reliable laser stripe segmentation and precise extraction of centerlines and feature points remain challenging. This is because inter-reflection interference and camera saturation severely degrade measurement accuracy. In this article, we propose a novel physically based polarization reflection separation method using polarization features and spatial structural information for robust laser stripe extraction. The physical principles and polarization properties of direct- and inter-reflections from metallic surfaces are investigated through theoretical modeling and simulation. By combining the Stokes parameter S2 with a saturation region segmentation criterion, the inter-reflection regions of the laser stripe are automatically determined. An adaptive mask dilation approach is then introduced to suppress saturation-induced interference without complex parameter tuning. Finally, the centerlines and feature points are accurately extracted using the gray-gravity method with least-squares fitting. Experiments demonstrate superior segmentation and extraction performance under strong reflective interference, achieving an average feature-point extraction error of 2.36 pixels, a repeatability error within ±1.16 pixels, and an average computational time of 0.2531 s. Moreover, comprehensive evaluations across varying groove angles, laser rotation angles, exposure times, surface conditions, and dynamic robotic platforms further validate the robustness and effectiveness of the proposed method. This study highlights the potential of polarization imaging for addressing optical interference challenges in industrial welding scenarios, providing a solid foundation for intelligent and automated welding.