A Novel 3-D Point Cloud Registration Method With Quaternions Optimization and Data Snooping

Various algorithms derived from the errors-in-variables (EIV) model are always commonly used for estimating transformation parameters in 3-D point cloud registration, which is attributed to the flexible and robust nature of its mathematical framework. Despite the effectiveness of conventional methods in handling a range of transformation problems, limitations may arise when dealing with scenarios that involve large rotation angles and arbitrary scale ratios. Moreover, the presence of complex external environmental variables can introduce outliers in datasets, which significantly affect registration outcomes. To tackle these challenges, we propose to abstract the transformation model of point cloud registration as a generalized EIV model, utilizing quaternions to represent spatial rotation parameters. Subsequently, we introduce a sensitivity analysis-based data snooping algorithm to identify outliers. Unlike previous studies, our method employs a new algorithm utilizing quaternions to improve computational efficiency, removing the requirement for exact initial parameter values at the beginning of iterations. Additionally, the proposed algorithm utilizes sensitivity analysis to build test statistics, with the goal of improving outlier detection and enhancing resistance to their negative effects. Our method has demonstrated superior efficiency and precision compared to traditional methods in both simulations and real experiments.