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 ›› 2019, Vol. 45 ›› Issue (2): 258-264. doi: 10.19678/j.issn.1000-3428.0049232

• 多媒体技术及应用 • 上一篇    下一篇

改进的SAMF目标跟踪算法

李大湘,吴玲风,李娜,刘颖   

  1. 西安邮电大学 通信与信息工程学院,西安 710121
  • 收稿日期:2017-11-08 出版日期:2019-02-15 发布日期:2019-02-15
  • 作者简介:李大湘(1974—),男,副教授,主研方向为刑侦图像检索、机器识别;吴玲风(通信作者),硕士研究生;李娜,博士;刘颖,教授、博士。
  • 基金资助:

    国家自然科学基金(61571361);陕西省国际合作交流项目(2017KW-013);西安邮电大学研究生创新基金(CXJJ2017004)。

Improved SAMF Object Tracking Algorithm

LI Daxiang,WU Lingfeng,LI Na,LIU Ying   

  1. School of Communication and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China
  • Received:2017-11-08 Online:2019-02-15 Published:2019-02-15

摘要:

在基于视觉的目标跟踪过程中,当目标被遮挡时,跟踪算法精度往往下降。针对该问题,在SAMF跟踪算法基础上,提出一种基于图像分块重检测的改进算法。通过寻找最佳目标位置的方法优化SAMF算法,提高目标跟踪的准确率。利用图像分块及样本逐一测试的方法设计重检测模块,当目标因遮挡而无法稳定跟踪时,启动重检测模块,根据重检测后的最大响应值找出目标中心点,并引入模型自动更新策略对目标位置进行更新,避免出现跟踪漂移的现象。采用9个目标跟踪标准测试集进行对比实验,结果表明,该算法较SAMF算法平均距离精度提高了38%,且优于KCF、CN、CSK等其他目标跟踪算法。

关键词: 目标跟踪, 相关滤波, 重检测, 遮挡, 模型更新策略

Abstract:

In visual tracking,the accuracy of the tracking algorithms often declines when the object is occluded.In order to solve this problem,an improved SAMF algorithm based on re-detection is proposed in this paper.Firstly,the SAMF algorithm is optimized using the method of locating the object position to improve the accuracy of tracking.Secondly,a re-detection module is designed and added into the procedure of tracking.When the object is occluded,it cannot be tracked stably.Therefore,the re-detection module is started,and the object is found by using the maximum response value.At last,an automatic updating strategy for the appearance model is introduced to avoid the tracking drift.The proposed algorithm is compared with other tracking methods on nine video sequences.Experimental results show that the average distance precision of the proposed algorithm is improved by 38% compared with SAMF algorithm,and is superior to other tracking algorithms such as KCF,CN and CSK.

Key words: object tracking, correlation filtering, re-detection, occlusion, model update strategy

中图分类号: