计算机工程

• 图形图像处理 • 上一篇    下一篇

基于联合支持向量机的目标跟踪算法

孙俊涛,张顺利,张利   

  1. (清华大学 电子工程系,北京 100084)
  • 收稿日期:2016-03-01 出版日期:2017-03-15 发布日期:2017-03-15
  • 作者简介:孙俊涛(1990—),男,硕士,主研方向为机器视觉、目标跟踪;张顺利,博士;张利,教授、博士。
  • 基金项目:
    国家自然科学基金重点项目(61132007);国家自然科学基金-民航基金联合项目(U1533132)。

Object Tracking Algorithm Based on Joint Support Vector Machine

SUN Juntao,ZHANG Shunli,ZHANG Li   

  1. (Department of Electronic Engineering,Tsinghua University,Beijing 100084,China)
  • Received:2016-03-01 Online:2017-03-15 Published:2017-03-15

摘要: 为得到包含目标与背景的区分度以及目标自身特性的外观模型,给出一种联合支持向量机。结合一类支持向量机和二类支持向量机的特点,设计优化的目标函数,利用拉格朗日乘子法给出其对偶形式,实现求解步骤,并基于此提出目标跟踪算法,以加强目标外观模型表达的鲁棒性,提高对目标和背景的鉴别能力。在公开的测试视频集上的实验结果表明,该算法能够准确地跟踪目标,并且具有较好的稳定性。

关键词: 模式识别, 目标跟踪, 联合支持向量机, 外观模型, 目标函数

Abstract: In order to obtain the appearance model that contains both the distinction between object and background and the object characteritics,this paper presents a joint Support Vector Machine(SVM).Combining the characteristics of the one-class SVM and the binary SVM,it proposes an optimized object function and gives the solution steps by its dual form with Lagrange multiplier.Based on it,an object tracking algorithm is proposed.The robustness of the object appearance model is enhanced,and the ability to identify the object and background is improved.Experimental results on the public test videos show that the proposed algorithm can realize precise object tracking,and has good stability.

Key words: pattern recognition, object tracking, joint Support Vector Machine(SVM), appearance model, object function

中图分类号: