应用改进IRLS-ICP的植株点云配准  

Plant point cloud registration based on improved IRLS-ICP algorithm

作　　者：刘晓慧 耿楠[1] 张志毅[1] 胡少军[1] LIU Xiao-hui;GENG Nan;ZHANG Zhi-yi;HU Shao-jun(College of Information Engineering,Northwest Agriculture and Forestry University,Yangling 712100,China)

机构地区：[1]西北农林科技大学信息工程学院

出　　处：《计算机工程与设计》2019年第7期1964-1970,共7页Computer Engineering and Design

基　　金：国家863高技术研究发展计划基金项目(2013AA102304);基本科技创新一般基金项目(QN2013056)

摘　　要：为提高不同角度扫描得到的存在噪声和离群点的植株点云配准的鲁棒性和效率,提出一种基于改进的加权重迭代最近点(IRLS-ICP)植株点云配准方法。以获取准确的重心和协方差矩阵为目的,基于最小绝对值和,利用Huber损失函数建立目标函数,通过加权重迭代(IRLS)技术计算最优解;为加快最近点对的搜索效率,利用Delaunay三角剖分,建立以四面体为单元的点集搜索结构。实验结果表明,该配准方法对含有离群点和噪声的点云具有很强的鲁棒性,与传统的配准方法相比效率提高了约90.4%,精度提高了约0.6%。To improve the robustness and efficiency of the registration for point cloud of plants under noises and outliers during 3D scans,IRLS-ICP registration method was presented. A cost function based on the least absolute values was established using Huber function,and it was solved using IRLS approach and therefore accurate center and covariance matrix were computed. To accelerate the search efficiency of the nearest point pair,the Delaunay triangulation was used to establish a point set search structure with tetrahedron as the unit. Experimental results show that the registration method for plants is robust to noises and out- liers . Results also show that registration speed and accuracy can be increased by 90.4% and 0.6% respectively comparing with classic method.

关 键 词：点云配准 鲁棒 加权重迭代 最近点迭代 植株 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]