›› 2015, Vol. 28 ›› Issue (6): 13-.

• 论文 • 上一篇    下一篇

基于局部特征和Mean Shift的目标跟踪算法研究

席志红,李永佳,段炼   

  1. (1.哈尔滨工程大学 信息与通信工程学院,黑龙江 哈尔滨 150000;
    2.哈尔滨体育学院 财务处,黑龙江 哈尔滨 150000)
  • 出版日期:2015-06-15 发布日期:2015-06-20
  • 作者简介:席志红(1965—),女,教授,博士生导师。研究方向:图像处理与应用,视觉目标探测与识别,信号检测与估计等。李永佳(1989—),男,硕士研究生。研究方向:信息系统与网络通信技术。E-mail:787370269

Research on Object Tracking Method Based on Local Feature and Mean Shift

XI Zhihong,LI Yongjia,DUAN Lian   

  1. (1.College of Information and Communication Engineering,Harbin Engineering University,Harbin 150000,China;
    2.Financial Department,Harbin Institute of Physical Education,Harbin 150000,China)
  • Online:2015-06-15 Published:2015-06-20

摘要:

利用均值漂移进行目标跟踪的算法,在被跟踪目标出现旋转、尺度变化、噪声干扰等情况下,无法得到准确的跟踪结果。文中提出了基于当前流行目标跟踪算法和局部特征相结合的算法,基于局部特征—形状上下文(Shape Context)特征的Mean Shift目标跟踪算法。该算法首先提取目标的轮廓信息和特征,根据采样点之间位置和距离关系建立Shape Context直方图,最后所有点的Shape Context直方图构成了图像的Shape Context特征,最后根据Mean Shift算法进行跟踪。实验结果表明,该算法在跟踪目标出现尺度变化、旋转、噪声干扰和遮挡等情况下能够准确地跟踪物体,鲁棒性好。

关键词: 局部特征, 形状上下文, 均值漂移, 目标跟踪

Abstract:

Accurate tracking results are unobtainable by the traditional algorithm of Mean Shift due to noise,rotation and changed scales of the target etc.A shape context tracking algorithm for Mean Shift objects is proposed by combining the current popular target tracking algorithm and local features.The contour information and features of shape context are extracted from the target.Then a shape context histogram is established according to the locations and spacing of sample points and the histograms of all the points constitutes the shape context feature of the image  after which the Mean Shift algorithm is adopted for tracking.The experimental results show that the algorithm can accurately track the object with scale change,rotation,noise and is of good accuracy and robustness.

Key words: local feature;shape context;mean shift;target tracking

中图分类号: 

  • TP391.41