利用邻域特征点相关性的点云配准算法

Point Cloud Registration Algorithm Using Correlation of Neighborhood Feature Points

针对经典ICP进行点云配准容易陷入局部最优且处理时间长、精度低的问题,本文提出一种利用邻域提取特征点进行配准的优化算法。首先,通过邻域特征计算法向量和曲率特征从待匹配数据中选取特征点,通过特征点匹配得到配准平移和旋转参数,并利用配准参数对待匹配数据进行初始配准,然后,通过ICP算法对数据进行精细配准。试验证明,在利用特征点进行初始配准的基础上,解决了经典ICP算法容易陷入局部最优的问题,且具有良好的配准精度和运行效率。

Aiming at the problems that point cloud registration by classical ICP is prone to fall into local optimum,long processing time and low precision,this paper proposes an optimization algorithm that uses neighborhood extraction feature points for registration.The feature points are selected from the data to be matched by calculating the normal vector and the curvature feature of the neighborhood features,and the registration translation and rotation parameters are obtained through the feature point matching,and the registration parameters are used to perform the initial registration of the data to be matched,then,finely registered by the ICP algorithm.Experiments show that,based on the use of feature points for initial registration,the problem that the classical ICP algorithm tends to fall into local optimum is solved,and it has good registration accuracy and operating efficiency.