基于IVCCS的三维点云配准算法

针对传统迭代最近点(ICP)算法在数据丢失以及存在噪声点的情况下配准时间过长、精度较低等问题，提出了一种基于改进的体素云连通性分割(IVCCS)与加权最近邻距离比相结合的配准算法。利用双阈值体素去噪剔除初始种子体素中的噪声体素，解决原本体素云连通性分割算法(VCCS)中因单一约束条件导致种子体素错误剔除的问题，同时将体素云分层去噪来加快配准的运算速度；利用流约束聚类提取点云中的特征点，并依据最近邻距离比验证特征点是否为重合点，赋予不同的权重优化ICP最小目标函数，从而加快配准速度。实验结果表明，该算法相对于传统ICP算法迭代次数减少，在精度与速度方面均有显著提升，相比于基于快速点特征直方图(FPFH)的ICP算法配准精度提高了8.5%~24.7%，速度上提高了65.6%~92.3%，迭代次数减少了16.6%~38%。

3D point cloud registration algorithm with IVCCS

In order to solve the problems of long registration time and low accuracy in the case of data lossing and noise points existing in the traditional iterative closest point (ICP) algorithm, a new registration algorithm based on improved voxel cloud connectivity segmentation (IVCCS) combined with weighted nearest neighbor distance ratio was proposed. Double threshold voxel denoising was used to remove the noise voxel in the initial seed voxel, which was caused by a single constraint in the original voxel cloud connectivity segmentation algorithm (VCCS). Meanwhile, layered voxel cloud denoising was used to speed up the operation speed of registration. The feature points in the point cloud were extracted by flow constrained clustering, and whether the feature points were coincidence points was verified according to the nearest neighbor distance ratio. The minimum objective function of ICP was optimized by giving different weights, so as to accelerate the registration speed.Experimental results show that compared with the traditional ICP algorithm, the algorithm has reduced the number of iterations, and significantly improved the accuracy and speed. Compared with the ICP algorithm based on fast point feature histogram (FPFH), the algorithm has improved the registration accuracy by 8.5%-24.7%, the speed by 65.6%-92.3%, and the number of iterations decrease by 16.6%-38%.