查询结果:   马福峰,耿楠,张志毅.基于邻域几何特征约束的植株三维形态配准方法研究[J].计算机应用与软件,2016,33(9):184 - 189.
中文标题
基于邻域几何特征约束的植株三维形态配准方法研究
发表栏目
图像处理与应用
摘要点击数
786
英文标题
ON 3D PLANT MORPHOLOGY REGISTRATION METHOD BASED ON GEOMETRICAL FEATURE CONSTRAINT OF NEIGHBOURHOOD
作 者
马福峰 耿楠 张志毅 Ma Fufeng Geng Nan Zhang Zhiyi
作者单位
西北农林科技大学信息工程学院 陕西 杨凌 712100     
英文单位
College of Information Engineering,Northwest Agricultural and Forest University,Yangling 712100,Shaanxi,China     
关键词
三维形态配准 几何特征 特征相似度 迭代最近点算法
Keywords
3D morphology registration Geometrical feature Feature similarity ICP algorithm
基金项目
国家高技术研究发展计划项目(2013AA 102304);基本科技创新一般项目(QN2013056);科技创新一般项目(2014YB067)
作者资料
马福峰, 硕士,主研领域:计算机虚拟技术与图形学。耿楠,教授。张志毅,副教授。 。
文章摘要
为提高不同角度多次测量得到的植株点云配准速度和精度,提出一种基于植株点云邻域几何特征约束改进的三维形态配准方法。首先,针对点云量大并缺少拓扑信息,选取关键点集并估计其中每个点的支撑邻域来拟合出支撑曲面,进一步计算出邻域几何特征。其次,采用特征相似度的方法实现点云的初始配准。最后,在初始配准的基础上,加入两个新的夹角几何特征约束匹配点对改进ICP算法进行配准优化。利用bunny、兵马俑模型点云对算法的精度和通用性进行测试,并在实际应用中验证了配准效果和算法鲁棒性。结果表明,与传统的特征配准方法相比,该方法配准速度提高约10%以上,精确配准误差约为传统算法误差的1%。
Abstract
In order to improve the registration speed and precision of plant point cloud derived from multiple measurements in various angles, this paper proposes an improved 3D morphological registration method which is based on the geometrical feature constraint of neighbourhood of plant point clouds. First, aiming at the large amount of point clouds and lack of topological information, the method estimates the support neighbourhood of each point by selecting key point set to fit the support surface and then to further compute the geometrical feature of neighbourhood. Secondly, it employs the feature similarity method to implement the initial registration of point clouds. Finally, based on initial registration it adds two matching pairs of angle’s geometric feature constraints to improve the iterative closest point (ICP) algorithm and to optimise the registration. We tested the accuracy and universality of the algorithm with the point clouds of bunny and Terracotta Army models, and verified in practical application the effect of registration and the robustness of the algorithm. Results showed that compared with traditional feature-based registration method, the proposed method increased the registration speed about over 11.9%, and the error of precise registration was about 1% of that of the traditional feature-based algorithm.
下载PDF全文