作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2008, Vol. 34 ›› Issue (1): 201-203. doi: 10.3969/j.issn.1000-3428.2008.01.069

• 人工智能及识别技术 • 上一篇    下一篇

基于多尺度的轮廓匹配方法

刘 曙,罗予频,杨士元   

  1. (清华大学自动化系,北京 100084)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2008-01-05 发布日期:2008-01-05

Contour Matching Algorithm Based on Multiscale Space

LIU Shu, LUO Yu-pin, YANG Shi-yuan   

  1. (Department of Automation, Tsinghua University, Beijing 100084)
  • Received:1900-01-01 Revised:1900-01-01 Online:2008-01-05 Published:2008-01-05

摘要: 物体的形状轮廓是计算机视觉中一个重要的特征。该文提出了一种基于多尺度下局部特征的描述和动态规划的形状匹配和识别方法。考虑到轮廓在不同尺度下特征点的变化,给出了形状的多尺度描述方法。在轮廓分段匹配过程中,根据噪声和形变的程度不同对局部轮廓分别采用不同尺度滤波,避免了局部形变及噪声导致错误的匹配结果。该算法对于噪声、形变和适度的遮挡有较好鲁棒性,用于行人识别上取得了较好的效果。

关键词: 形状匹配, 轮廓的多尺度描述, 动态规划

Abstract: Shape feature is an important visual feature. This paper proposes an approach based on multiscale representation and dynamic programming for the shape matching and recognition. Considering the changes of shape feature points in different scale, multiscale representation of contours is proposed. And the process of contour segmentation can adjust automatically while the amounts of noise and deformation change. It is robust with respect to noise, shape deformation and occlusion. This paper successfullies apply the matching algorithm to the task of pedestrian recognition in the night vision system with good results, also demonstrates the superiority of the proposed method over three traditional shape matching methods.

Key words: shape matching, multiscale representation of contour, dynamic programming

中图分类号: