Robust global registration of point clouds by closed-form solution in the frequency domain

Point cloud registration is invariably an essential and challenging task in the fields of photogrammetry computer vision to align multiple point clouds a united reference frame. In this paper, we propose novel global method using robust phase correlation for low-overlapping clouds, which less sensitive noise outliers than feature-based methods. The proposed achieved by converting estimation rotation, scaling, translation spatial domain problem correlating low-frequency components frequency domain. Specifically, it consists three core steps: transformation from domain, decoupling translation, adapted shift estimation. first step, unstructured unordered 3D points are transformed via Fourier transformation, following voxelization binarization process. second decoupled sequential operations, including transform, resampling strategies, Fourier-Mellin transform. third parameters into tasks. solved method, matched decomposing normalized cross-power spectrum linearly fitting decomposed signals with closed-form solution ℓ1-norm-based estimator. Experiments were conducted different datasets urban natural scenarios. Results demonstrate efficiency majority rotation errors reaching 0.2 degree 0.5 m, respectively. Additionally, also validated experiments that versatile wide ranges overlaps various geometric characteristics.