计算机与现代化

• 图像处理 • 上一篇    下一篇

 TLD视频目标跟踪方法改进

  

  1. 南京航空航天大学计算机科学与技术学院,江苏南京210016
  • 收稿日期:2015-01-05 出版日期:2015-04-27 发布日期:2015-04-29
  • 作者简介: 金龙(1988-),男,吉林长春人,南京航空航天大学计算机科学与技术学院硕士研究生,研究方向:计算机视觉; 孙涵(1978-),男,副教授,博士,研究方向:数字图像处理,模 式识别与计算机视觉。
  • 基金资助:
     国家自然科学基金资助项目(61203246,61375021); 江苏省自然科学基金资助项目(SBK201322136)

 An Improved TLD Visual Target Tracking Method

  1. College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China
  • Received:2015-01-05 Online:2015-04-27 Published:2015-04-29

摘要:

TLD(Tracking-Learning-Detection)是Zdenek Kalal提出的一种新跟踪算法。该算法与传统跟踪算法的显著区别在于将传统跟踪算法与检测算法相结合来解决被跟踪目标在被跟踪过程中发
生的形变、部分遮挡等问题。对TLD算法进行改进,在算法检测模块引入基于Meanshift与Kalman的当前帧目标所在区域预估,有效缩小检测模块的检测范围,提高算法实时性及准确性;对原算法方差分
类器改进采用颜色特征分类器,提高算法对目标识别性能;对综合模块改进,提高算法目标跟踪成功率。通过实验对改进后的TLD及原TLD进行比较,实验结果表明,改进的TLD算法具有更高的跟踪准确性
及更好的跟踪实时性。

关键词: TLD, 目标跟踪, 目标区域预估, 颜色特征

Abstract:

Tracking-Learning-Detection (TLD) is a new tracking algorithm proposed by Zdenek Kalal. The most obvious difference between TLD and the traditional tracking
algorithm is that TLD combines traditional tracking algorithm with traditional detection algorithm to solve problems such as deformation and partial occlusion which may happen
when target is followed during the tracking procedure. The TLD is improved by this paper, based on Meanshift and Kalman, this paper introduces prediction for the area where
current frame target is located into testing module of the algorithm, which is supposed to narrow detection range of detector effectively and reduce the computation burden of
the algorithm; variance classifier in original algorithm is replaced by color feature classifier, which improve performance of target recognition. what’s more, the improvement
of integrated modular can increase the rate of success for target tracking. After comparison of the improved TLD and original TLD, the result shows that the improved TLD
algorithm has high accuracy and better precision for tracking.

Key words:  TLD, target tracking, target area prediction, color feature