欢迎访问《兵工学报》官方网站,今天是

兵工学报 ›› 2011, Vol. 32 ›› Issue (11): 1353-1358.

• 论文 • 上一篇    下一篇

基于分割模板加权Hausdorff距离矩阵的特征匹配算法

徐一鸣1,2, 刘晓利1, 刘怡昕1,3   

  1. (1.南京理工大学 瞬态物理国家重点实验室, 江苏 南京 210094;2.南通大学 电气工程学院江苏 南通 226000;3.南京炮兵学院江苏 南京 211132)
  • 收稿日期:2010-04-27 修回日期:2010-04-27 上线日期:2014-05-04
  • 通讯作者: 徐一鸣 E-mail:helloyiming@sina.com
  • 作者简介:徐一鸣(1981—), 男, 博士研究生

A Feature Matching Algorithm Based on Template Partition Weighted Hausdorff Distance Matrix

XU Yi-ming1,2, LIU Xiao-li1, LIU Yi-xin1,3   

  1. (1.National Key Laboratory of Transient Physics, Nanjing University of Science and Technology, Nanjing 210094, Jiangsu,3.Nanjing Artillery Academy, Nanjing 211132, Jiangsu,China)
  • Received:2010-04-27 Revised:2010-04-27 Online:2014-05-04
  • Contact: XU Yi-ming E-mail:helloyiming@sina.com

摘要: 针对电视制导系统图像匹配时遇到的大比例目标遮挡以及严重噪声干扰等情况,提出一种基于分割模板加权Hausdorff距离(HD)矩阵的特征匹配算法。将模板分割为若干个子模板;利用最小核相似区检测角点(SUSAN)算子提取图像的角点;分别计算子模板与搜索图像对应区域的基于角点响应函数的加权HD,构造出HD矩阵;经过角点密度矩阵修正得到相似性度量,采用Frobenius矩阵范数求得矩阵的最佳解,即对应于最佳匹配位置。在目标跟踪实验中加入大比例遮挡及严重斑点噪声(n=10),当模板与对应区域的特征点数量相差达到-43.75%~56.25%时,仍然可以实现配准。

关键词: 信息处理技术, 特征匹配, 最小核相似区检测角点, Hausdorff距离, 矩阵范数

Abstract: A feature matching algorithm based on template partition Hausdorff distance matrix was proposed to resolve the problems of large scale target occlusion and serious noises in TV guidance process. The template image could be divided into several sub-templates and the corners were detected by using SUSAN operator. Hausdorff distance matrixes between sub-templates and searching image were constructed by calculating weighted Hausdorff distance based on the corner response function. Corrected by corner density matrix, a similarity measure was obtained, and the matrix with a minimal Frobenius norm was corresponding to the best matching position. A target tracking experiment with partial target occlusion and serious spot noise (n=10) was carried out, and the target could be tracked accurately when the difference of feature points in template and correlation area ranged from -43.75% to 56.25%.

Key words: information processing technique, feature matching, smallest univalue segment assimilating nucleus corner, Hausdorff distance, matrix norm

中图分类号: