Abstract: In this paper, based on judging the angle between neighboring normal points of point cloud, an automatic registration algorithm is proposed to solve the problems of low accuracy and robustness of 3D point cloud auto-registration. The algorithm firstly computes the cosine of the angle between the normal vector of each point in the point cloud and the normal vector set of the neighboring point set, and then takes the cosine of each point in the neighborhood as the attribute feature vector of the point. The algorithm extracts feature points by feature classification. In accordance with the similarity of the geometric features, the matching pairs are preliminarily searched, then the Euclidean distance constraint is used to eliminate matching pairs. The least squares method is used to calculate the initial registration parameters, then the refined Iterative Closest Point（ICP） algorithm is used for fine matching. Experiments show that the algorithm has improved both in convergence speed and matching accuracy with respect to the classic ICP algorithm.

Key words: point cloud data, automatic registration, feature classification, matching pairs, normal vector angle, Iterative Closest Point（ICP）

摘要： 针对三维点云自动配准精度不高、鲁棒性不强等问题，提出一种基于判断点云邻域法向量夹角的自动配准算法。该算法首先计算点云中每个点的法向量与邻域点集的法向量夹角的余弦值，然后把邻域各点的余弦值作为该点的属性特征向量，进行特征分类提取特征点，根据几何特征的相似性初步搜索匹配点对，并采用欧式距离约束条件剔除匹配错误的点对；运用最小二乘法计算初始配准参数，再通过改进的迭代最近点（Iterative Closest Point，ICP）算法进行精匹配。实验证明，该算法相对于经典的ICP算法无论收敛速度还是匹配精度上都有提升。

关键词: 点云数据, 自动配准, 特征分类, 匹配点对, 法向量夹角, 迭代最近点（ICP）