基于点密度调整的激光雷达点云配准改进方法
An Improved LiDAR Registration Alogrithm Based on Point Density Adjustment
王翊成 1李明磊 1魏大洲 2吴伯春2
作者信息
- 1. 南京航空航天大学电子信息工程学院 南京 211106
- 2. 中国航空无线电电子研究所 上海 200233
- 折叠
摘要
线扫激光雷达设备由于扫描机制,采集的点云数据会呈现条纹结构.为了改善点云配准结果,提出了一种基于点密度调整的激光雷达点云配准改进方法.首先利用八叉树算法和区域生长法将输入数据转化成超体素框架图.然后基于该框架来约束重采样过程,将重采样点沿法向量方向移动,再根据一个使点均匀分布的最小化能量函数来调整插入点分布,实现数据重采样.实验表明,该算法能改进点云质量,提高经典ICP算法和3D-NDT算法的配准精度,减少配准耗时.
Abstract
Due to the scanning mechanism,the point cloud data collected by line-scan LiDAR presents obvious stripe struc-tures.In order to guarantee the registration results,an improved LiDAR point cloud registration method based on point density ad-justment is proposed.First,the input data are transformed into a supervoxel frame by an octree algorithm and a region growing meth-od.Then,the resampling process is constrained based on the framework,and new points are projected along the normal direction.Then,the distribution of the insertion points are adjusted to minimize an energy function to make the points evenly distributed.Ex-periments show that the algorithm can improve the quality of point cloud and the registration accuracy of classical ICP algorithm and 3D-NDT algorithm.Furthermore,the registration time is reduced.
关键词
点云重采样/点云配准/迭代最近点算法/正态分布转换算法/线扫激光雷达Key words
point cloud resampling/registration/ICP alogrithm/NDT alogrithm/line-scan LiDAR引用本文复制引用
基金项目
国家自然科学基金(41801342)
中央高校基本科研业务费专项(NZ2020008XZA20016)
出版年
2024
国家科技期刊平台
NSTL
万方数据