基于拟牛顿法改进的3D正态分布变换点云配准算法

针对3D正态分布变换算法在大型场景点云数据配准时效率低的问题,提出一种基于拟牛顿法改进的3D正态分布变换算法。3D正态分布变换算法主要通过牛顿迭代法进行两视点云最优转换参数求解,但是随着待配准点云数据量的增加,牛顿迭代法需要大量的时间计算Hessian矩阵,增加了算法整体的时间复杂度。本文算法通过拟牛顿法代替牛顿法求解Hessian,改善了3D正态分布变换算法针对大型场景点云数据配准需要大量时间去计算Hessian矩阵的问题。实验表明,本文算法针对大型点云数据不仅能够保持传统3D正态分布变换算法的配准精度,还能提高配准效率。

An Improved 3D Normal Distribution Transformation Point Cloud Registration Algorithm Based on Quasi-Newton Method

For the problem of low registration efficient with large scene points in the 3D normal distribution transform algorithm registration process, a 3D normal distribution transform algorithm based on Quasi-Newton method is proposed. The 3D normal distribution transform algorithm mainly uses the Newton iteration method to solve the two-view cloud optimal transformation parameters. However, with the increase of the amount of data to be registered, the Newton iteration method needs a lot of time to calculate the Hessian matrix, which increases the time of the whole algorithm complexity. In this paper, Hessian is solved by the method of Quasi-Newton method, which improves the 3D normal distribution transformation algorithm. Experiments show that the algorithm can not only keep the registration precision of traditional 3D normal distribution transform algorithm, but also improve the registration efficiency for large point cloud data.