三维重建中点云配准算法研究

Research on point cloud registration algorithm in 3Dreconstruction

为了提高三维重建中点云配准的效率和精度,提出了一种基于由粗到精(CTF)方式的点云配准算法,当两帧点云位姿差别较大时,首先利用基于FPFH特征点的采样一致性初始配准算法(SAC-IA)进行两点云的粗配准,为精配准过程提供较好的初始位姿;然后再采用传统ICP算法的改进算法EM-ICP算法对点云位姿做refine,进行点云的精确配准;同时,在执行算法过程中通过GPU进行多线程并行加速。实验结果表明,所提配准算法与只使用单一的配准算法相比,配准精度有较大的提高;与SAC-IA+ICP算法相比,在保证配准精度的基础上,提高了配准效率。

In order to improve the efficiency and accuracy of point cloud registration in 3Dreconstruction,apoint cloud registration algorithm based on the from coarse to fine(CTF)method is proposed.When the difference between the point cloud poses of two frames is large,firstly based on FPFH sampling consistent initial registration algorithm for feature points(SAC-IA)performs coarse registration of two point clouds to provide a better initial pose for the precise registration process;then,the improved algorithm of traditional ICP algorithm,EM-ICP algorithm,is used to Refine the point cloud pose to perform accurate registration of the point cloud;at the same time,perform multi-threaded parallel acceleration through the GPU during the execution of the algorithm.The experimental results show that the registration algorithm proposed in this paper has greatly improved the registration accuracy compared with using only a single registration algorithm.Compared with the SAC-IA+ICP algorithm,the registration accuracy is greatly improved.The amplitude reduces the algorithm running time and improves the registration efficiency.