摘要
传统的正态分布变换配准算法处理初始位姿变换相差较大的两帧点云时,存在无法收敛或者陷入局部最小值的问题。为了提高算法的收敛性能,提出了一种基于LM方法改进的三维正态分布变换配准算法。在原始点云的体素滤波中,引入k最邻近搜索寻找距离重心最近的点作为替代点,提高点云数据在下采样后的精度。在算法的迭代优化步骤中使用列文伯格-马夸尔特法,通过计算因子ρ的值动态调节每次迭代过程的步长直到达到最优,避免无法收敛或陷入局部最小值的问题。实验数据表明,相比于传统的正态分布变换配准算法,所提算法在初始位姿变换相差较大时,精度更高且鲁棒性较好。
The traditional normal distribution transformation registration algorithm has the problem of being unable to converge or falling into local minima when dealing with two frame point clouds with significant initial pose transfor-mation differences.In order to improve the convergence performance of the algorithm,an improved three-dimensional normal distribution transformation registration algorithm based on the LM method is proposed.In the voxel filtering of the original point cloud,k-nearest neighbor search was introduced to find the point closest to the center of gravity as a substitute point,improving the accuracy of point cloud data after downsampling.In the iterative optimization step of the algorithm,the Levenberg-Marquardt method was used to dynamically adjust the step length of each iteration process by calculating the value of the factorpuntil it reaches the optimum,so as to avoid the problem of failure to con-verge or falling into a local minimum.Experimental data show that compared with the traditional normal distribution transformation registration algorithm,this algorithm has higher accuracy and better robustness when the initial pose transformation differs greatly.
作者
胡璇熠
崔更申
匡兵
邱德宪
HU Xuan-yi;CUI Geng-shen;KUANG Bing;QIU De-xian(School of Computer and Information Security,Guilin University of Electronic Science and Technology,Guilin Guangxi 541000,China;College of Mechanical and Electrical Engineering,Guilin University of Electronic Science and Technology,Guilin Guangxi 541000,China)
出处
《计算机仿真》
北大核心
2023年第11期306-310,共5页
Computer Simulation
基金
广西创新驱动发展专项资金项目(AA19046004)。
关键词
激光雷达
点云配准
正态分布变换
列文伯格-马夸尔特法
定位
Lidar
Point cloud registration
Normal distribution transformation
Levenberg-Marquardt method
Po-sitioning
