Coarse-to-Fine Robust Registration Method Based on Cauchy Kernel Function and Feature Constraints

To address the robustness challenges of point set registration under occlusion, noise interference, and large rotation angle scenarios, this paper proposes a hierarchical robust registration framework (HRRF), enhanced by Cauchy kernel functions and feature constraints in a coarse-to-fine manner. Traditional iterative closest point (ICP) algorithms exhibit limitations in noise sensitivity and matching stability. Our method achieves robust registration through a three-stage optimization framework. First, a point set initialization strategy integrating curvature-density feature extraction and hybrid sampling is designed to enhance feature representation capability. Second, in the coarse stage, a Median Absolute Deviation (MAD)-based dynamic threshold screening mechanism is introduced, combined with a Cauchy kernel function weighting to suppress outlier matches. Finally, in the fine registration stage, rigid transformation parameters are optimized via feature constraint-enforced parameter adjustment. Experimental results demonstrate that the proposed algorithm significantly reduces the average registration error compared to conventional methods in noisy and occluded 2D point set datasets. Notably, it maintains stable convergence even under large rotation angle conditions, providing a more robust registration solution for complex 3D reconstruction tasks.