欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 视觉与图像 • 上一篇    下一篇

基于相关滤波器的自适应目标跟踪

  

  1. 1. 中国石油大学(华东)计算机与通信工程学院，山东 青岛 266580；
    2. 中国科学院计算技术研究所智能信息处理重点实验室，北京 100190
  • 出版日期:2017-04-30 发布日期:2017-04-28
  • 基金资助:
    国家自然科学基金项目(61379106)；山东省中青年科学家奖励基金项目(BS2010DX037)；山东省自然科学基金项目(ZR2009GL014，
    ZR2013FM036，ZR2015FM011)；浙江大学CAD&CG 国家重点实验室开放课题(A1315)；中央高校基本科研基金项目(13CX06007A，
    14CX06010A，14CX06012A)

Adaptive Visual Tracking Based on the Correlation Filter

  1. 1. College of Computer and Communication Engineering, China University of Petroleum (Huadong), Qingdao Shandong 266580, China;
    2. Key Laboratory of Intelligent Information Processing, Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China
  • Online:2017-04-30 Published:2017-04-28

PDF (PC)

217

摘要： ：针对传统核相关滤波器(KCF)跟踪算法在遮挡情况下通常会跟踪失败问题，提出一
种目标边缘增强的自适应核相关滤波器算法。首先通过核相关滤波器找出目标位置，对目标位置
区域进行边缘增强，使目标特征更突出，提高分类器性能，然后通过计算目标位置的响应强度自
适应的改变模板的学习率参数，使检测模板适应外界环境的变化。实验选取15 段公开视频序列
进行测试，与现有几种相关滤波器进行比较，相对于结果最好的KCF 算法的平均中心位置误差
减少了9.6 像素，平均成功率提高了7.55%，平均距离精度提高了21.62%。实验结果表明在目标
被遮挡、旋转及快速运动等复杂情况下，该算法有较强的适应性，具有重要的研究和应用价值。

关键词: ：相关滤波器, 目标跟踪, 自适应跟踪

Abstract: Aiming at addressing the problem that traditional kernel correlation filter (KCF) algorithm
often fails in occlusion cases, we proposed an adaptive kernel correlation filter algorithm which can
enhance the target edge. First we employ kernel correlation filter to find the location of target. Then
we enhance the edge of location area in order to make the target characteristics more outstanding and
improve the classifier performance. At last, we calculate the response of the target location intensity
vector to change the template parameters adaptively. In experiments, we selected 15 sets of
challenging public video sequences to test the effectiveness of our method. compared with the KCF
algorithm which is best in existing correlation filters, the center position error of our method
reduced 9.6 pixels, the average success rate increased by 7.55% and the average distance precision
increased by 21.62%. The experiment’s results show that the proposed algorithm has strong
adaptability in obscured, rotating and rapid complex conditions. Furthermore, it has important theory
research and application value.

Key words: correlation filter, visual tracking, adaptive tracking