欢迎访问《图学学报》 分享到:

图学学报

• 计算机视觉 • 上一篇    下一篇

基于球心点互斥的球目标识别方法

  

  1. 1. 河海大学地球科学与工程学院,江苏 南京 211100;2. 宁波市规划设计研究院,浙江 宁波 315042
  • 出版日期:2018-02-28 发布日期:2018-02-06
  • 基金资助:
    国家自然科学基金项目(41301406,41201439);江苏省自然科学基金项目(BK20130829)

Spherical Target Recognition Method Based on Mutual Exclusion of Spherical Centers

  1. 1. School of Earth Sciences and Engineering, Hohai University, Nanjing Jiangsu 211100, China;
    2. Ningbo Urban Planning and Design Institute, Ningbo Zhejiang 315042, China
  • Online:2018-02-28 Published:2018-02-06

摘要: 提出了一种基于球心点互斥的球目标识别方法,用于从大场景三维点云中自动识
别未知个数和未知半径的球目标。首先,根据专门设计的球面点响应函数滤除大量非球面点,
并根据法向与曲率将剩余的球面点映射到球心位置;然后,构建用以描述局部密度渐变规律的
球心点互斥树,通过剪枝操作将其分裂成若干子树,其分别对应不同球目标的球心点聚类;最
后,根据球心点局部密度和球面点覆盖率估计值确认真实存在的球目标。实验结果表明:基于
球心点互斥的球目标识别方法能够有效解决大场景三维点云中球目标的识别问题,即使是存在
严重遮挡的情况下,暴露表面不足整个球面6%的球目标也都能够被识别出来。

关键词: 球目标识别, 球面点响应函数, 球心点互斥, 聚类, 球面覆盖率

Abstract: A new spherical target recognition method based on mutual exclusion of sphere centers is
proposed to solve the automatically identification problems of unknown number and unknown radius
targets in large-scale 3D point clouds. First, an effective spherical point response function is specially
designed to remove most of aspheric points, and every remaining spherical point is mapped to a
sphere center by taking advantage of its normal and curvatures. Then, a novel tree-like structure for
describing distribution and local density change rules of these centers is constructed, through a series
of pruning operation complying with the mutually exclusion relationships between different sphere
centers, the tree is split into several sub-trees, and a sub-tree correspond to a possible sphere target.
Finally, the real sphere is confirmed by the local density of the root node of sub-tree and the coverage
rate of points on the sphere surface. The experimental results demonstrate that the proposed sphere
recognition method based on the mutual exclusion of sphere centers can effectively identify and
precisely loc ate various spherical targets in a large and cluttered scene. Even in the case of serious
occlusion, such as the exposed surface is less than 6%, the sphere can also be robustly identified.

Key words: spherical target recognition, spherical point response function, spherical center mutual exclusion, clustering, spherical coverage