Abstract: In response to the problems of slow convergence and low accuracy in LiDAR point cloud registration in complex environments,this paper proposes a multi-stage optimization uncertainty ICP algorithm (CUBE-ICP).By using spatial distribution variance enhancement method and uncertainty regularization strategy,the accuracy and robustness of point cloud registration in complex scenes have been significantly improved.CUBE-ICP has developed a probability driven three-stage optimization framework: firstly,quantifying point cloud uncertainty based on a LiDAR error model.Secondly,capture the distribution characteristics of point clouds within the three-dimensional spatial unit through covariance matrices.Finally,the fusion of spatial distribution variance enhancement and uncertainty regularization constraints achieves a closed-loop approach from probabilistic modeling to robust optimization.The experimental results showed that the registration error of CUBE-ICP was significantly lower than mainstream algorithms such as ICP,3D-NDT,N-ICP,GICP,and LOAM in point cloud dual frame registration and continuous frame registration tasks.The CUBE-ICP algorithm proposed in this article has high performance advantages in the registration task of LiDAR point clouds,effectively solving the limitations of traditional ICP algorithms in processing complex scenes,and demonstrating stronger environmental adaptability and geometric feature adaptability.adaptability and geometric feature adaptability.

Key words: point cloud registration, ICP algorithm, CUBE algorithm, LiDAR, optimization in multiple stages