一种用于点云配准的改进迭代最近点算法

An Improved Iterative Closest Point Algorithm for Point Cloud Registration

经典的迭代最近点算法(ICP)收敛速度慢,在源点云和目标点云初始姿态不佳时存在容易陷入局部最优解等问题。针对上述问题构建一种结合快速点云粗配准的改进ICP算法。改进的ICP算法首先利用重心重合法进行两个点云集预处理,缩小平移误差,提高点云重叠度;然后采用随机采样一致性算法(RANSAC)实现两个点云集的粗配准,使两个点云集具有相对较好的初始位置姿态;最后利用体素栅格和KD树对ICP算法进行改进,实现点云精配准。将改进算法和经典ICP、GICP算法进行对比实验,结果表明:相较于经典ICP和GICP算法,改进算法精度更高、速度更快。

In order to solve the problem of low convergence rate and local optimal solution of the classical iterative closest point(ICP)algorithm when point cloud initial posture is not good,an improved ICP algorithm with fast point cloud initial registration is constructed.Firstly,the two point cloud data sets is preprocessed by center of gravity coincidence,so as to reduce translation error and improve point cloud overlap.Then,the random sampling consistency algorithm(RANSAC)is used to realize the initial registration of two point cloud sets to acquire a better initial position and posture.Finally,the voxel grid and KD tree are used to improve the ICP algorithm to achieve accurate registration of point clouds.The experimental results show that the improved algorithm has better accuracy and speed than the classical ICP and GICP algorithms.