基于八叉树与KD树索引的点云配准方法

摘要

针对点云配准算法中KD树多维查询效率较低的问题,提出一种基于八叉树和KD树多层索引结构的点云配准方法.首先为模型点云数据建立八叉树全局索引,然后在八叉树叶子结点构建局部数据的KD树索引.对传统的ICP点云配准算法进行改进,通过叶子结点的全局索引值快速定位局部点云数据块,利用局部KD树索引加快最近点的搜索,计算最近点时利用欧氏距离阈值、点对距离差值和法向量阈值剔除部分噪声点.实验表明,改进算法提高了点云配准的效率和精度.%A multilayer index structure based on octree and KD tree is reported for low query efficiency problems in multi-dimensional queries of KD tree.First, the octree global index for model point cloud is established.Then, the local data KD-tree indexes are built in the octree leave nodes.To improve the traditional Iterative Closest Point algorithm, the local point cloud data is quickly located based on the global index of leave nodes.Using the local KD tree indexes, the searching speed of closest point is sped up.Part of the noise points are removed by the euclidean distance threshold, the difference of interval of point pair and the normals threshold.Experimental result indicates that the proposed method can improve the efficiency and accuracy of registration.