基于改进ICP算法的点云自动配准技术

Automatic Registration Technology of Point Cloud Based on Improved ICP Algorithm

在零件的型面检测过程中,通常有数据采集、曲面重构、曲面配准和误差求取几个步骤。其中,曲面之间的配准是检测中关键的一环。针对传统的经典ICP(Iterative Closest Point)算法在配准过程中受噪声干扰大、鲁棒性差的缺点,在应用点云主方向贴合的粗略配准基础上,以经典的ICP算法为基础,提出了点云数据的欧氏距离阈值去噪和点云的方向矢量夹角阈值两种方法改进ICP算法,并应用改进算法作为点云之间的精确配准算法。对于经过初始配准的点云数据使用欧氏距离阈值法剔除点云间点对的噪声,并经点云各点间的方向矢量夹角阈值进行对应点采样,提高了传统ICP算法的效率和精度。经飞机和汽车零件点云配准实验验证,本算法的配准误差在±1μm内。算法具有设计简洁,响应快速的特点,有实际工作意义。

There are four steps in part surface error detection： data collection, surface reconstruction, surfaces registration and error calculation. Surface registration is the key step. According the defect of ICP （ Iterative Closest Point） algorithm, this paper proposes an improved ICP algorithm based on sample points taken by point - to - point distance and angle between the direction vectors of two points. Firstly, Principal Component Analysis （PCA） is introduced to register two point clouds roughly. Then it will eliminate the noisy points from point cloud using Euclidean distance threshold, and sample points by point - to - point distance and angle between the direction vectors of two point clouds. The effect of the algorithm is verified in the applications. The efficiency and accuracy is improved. Experiments show that the efficiency and accuracy is improved. The accuracy of registration between measuring points and the standard model is in allowing error range （ ≤1 μm）. The algorithm is simple and fast, and has significant usage in practical work.