Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (16): 1-12.DOI: 10.3778/j.issn.1002-8331.1805-0373

Previous Articles     Next Articles

Visual tracking via weighted global context-aware correlation filter

WAN Xin1, ZHANG Chunhui2,3, ZHANG Lin1, ZHOU Fan1   

  1. 1.College of Information Engineering, Shanghai Maritime University, Shanghai 201306, China
    2.School of Cyber Security, University of Chinese Academy of Sciences, Beijing 100049, China
    3.State Key Laboratory of Information Security, Institute of Information Engineering, Chinese Academy of Sciences, Beijing 100093, China
  • Online:2018-08-15 Published:2018-08-09


万  欣1,张春辉2,3,张  琳1,周  凡1   

  1. 1.上海海事大学 信息工程学院,上海 201306
    2.中国科学院大学 网络空间安全学院,北京 100049
    3.中国科学院信息工程研究所 信息安全国家重点实验室,北京 100093

Abstract: The related filtering algorithm based on the Context-Aware(CA) framework is a visual tracking algorithm, that is newly proposed. The disadvantages are that the context information is treated equally when dealing with fast motion, motion blur, occlusion and proportion change, and leads to reduce the robustness of visual tracking. For the above problems, this paper proposes a related filtering visual tracking algorithm based on the Weighted Global Context-Aware(WGCA) framework. Firstly, the original optimization problem is reconstructed. Secondly, according to the context-different areas and the size of the motion similarity of the tracking target, the different weights are given in the context region and the weight matrix is calculated. Finally, the one-dimensional and multi-dimensional solutions with the original and dual field are given. Based on the benchmark test set OTB-100, the results show that the framework significantly improves the robustness of filter and the tracking speed is comparable to the CA framework. However, the tracking accuracy and success rate are improved by 7% and 14% respectively.

Key words: visual tracking, correlation filter, weighted global context-aware

摘要: 基于上下文感知(Context-Aware,CA)框架的相关滤波算法是新近提出的一种视觉跟踪算法,其不足是在处理快速运动、运动模糊、遮挡、比例变化等情形时同等对待上下文信息,降低了视觉跟踪的鲁棒性。针对上述问题,提出了基于加权全局上下文感知(Weighted Global Context-Aware,WGCA)框架的相关滤波视觉跟踪算法。重构了原始的优化问题;根据上下文不同区域与追踪目标运动相似度的大小,赋予上下文区域不同的权值,计算出权值矩阵;给出了单通道和多通道情形的原始域、对偶域的闭式解。通过在基准测试集OTB-100上进行实验,结果表明该框架显著提高了相关滤波器的鲁棒性,其跟踪速度与CA框架相当,但跟踪精度和成功率较后者分别提高了7%和14%。

关键词: 视觉跟踪, 相关滤波器, 加权全局上下文感知