光学学报, 2018, 38 (10): 1010005, 网络出版: 2019-05-09
基于l
p 空间力学模型的三维点云配准算法
下载: 533次
Three-Dimensional Point Cloud Registration Algorithm Based on l
p Spatial Mechanics Model
图像处理 激光光学 l
p空间 力学模型 点云配准 奇异值分解 最近点迭代 image processing laser optics l
p space mechanics model point cloud registration singular value decomposition iterative closest point
摘要
为了提高三维激光扫描点云的配准效率和精度,提出一种基于l
p 空间力学模型的点云配准算法。针对待配准的两组点云数据,首先计算两片点云的重心,通过重心化将两组点云移到以重心为原点的同一坐标系下,然后利用l
p 空间力学模型将复杂的两组点云数据分别简化为三个特征向量表示的模型,再根据两点云特征向量的对应关系利用奇异值分解方法求解刚体变换旋转矩阵,得到初始配准参数,最后使用改进的最近点迭代(ICP)算法实现两组点云的精确配准。本文算法可以处理无序散乱点云样本。相比于经典ICP算法,本文算法对Bunny点云数据的配准效率提高了72%,对Dragon点云数据的配准速度提高了4倍。实验表明,本文算法收敛速度较快,效果优良。
Abstract
To improve the registration efficiency and accuracy of three-dimensional laser scanning point cloud, we propose a point cloud registration algorithm based on l
p space mechanics model. In the algorithm, the center of gravity of the sets data is calculated first, and two point clouds are moved to the same coordinate system with the center as the origin through gravity-centralizing. The complex point sets to be registered are represented as three eigenvectors respectively with the space mechanics model. Then, the singular value decomposition method is used to solve the rigid body transformation rotation matrix according to the corresponding relationship between two point sets’ eigenvectors. Finally, with the initial registration result, the improved iterative closest point (ICP) algorithm leads to perfect registration. The proposed algorithm can deal with disordered and scattered cloud sample. Compared with the classic ICP algorithm, the proposed method increases efficiency by 72% for the Bunny point cloud and is 4 times faster for Dragon scanning data. Experimental results indicate that the proposed algorithm has a fast convergence rate and good effect.
赵敏, 舒勤, 陈蔚, 杨赟秀. 基于