电子学报 ›› 2020, Vol. 48 ›› Issue (10): 2025-2032.DOI: 10.3969/j.issn.0372-2112.2020.10.021

• 学术论文 • 上一篇    下一篇

基于深度空间正则化的相关滤波跟踪算法

蒲磊1, 冯新喜2, 侯志强3, 查宇飞4, 余旺盛2   

  1. 1. 空军工程大学研究生院, 陕西西安 710077;
    2. 空军工程大学信息与导航学院, 陕西西安 710077;
    3. 西安邮电大学计算机学院, 陕西西安 710121;
    4. 西北工业大学计算机学院, 陕西西安 710072
  • 收稿日期:2019-04-18 修回日期:2020-03-21 出版日期:2020-10-25
    • 通讯作者:
    • 蒲磊
    • 作者简介:
    • 冯新喜 男,1964年10月出生,陕西富平人.1991年获西北工业大学博士学位,现为空军工程大学信息与导航学院教授、博士研究生导师,主要研究领域为信息融合、信号处理、目标跟踪等.E-mail:fengxinxi2005@aliyun.com
    • 基金资助:
    • 国家自然科学基金 (No.61571458,No.61703423)

Correlation Filter Tracking Based on Deep Spatial Regularization

PU Lei1, FENG Xin-xi2, HOU Zhi-qiang3, ZHA Yu-fei4, YU Wang-sheng2   

  1. 1. Graduate College, Air Force Engineering University, Xi'an, Shaanxi 710077, China;
    2. Institute of Information and Navigation, Air Force Engineering University, Xi'an, Shaanxi 710077, China;
    3. School of Computer Science and Technology, Xi'an University of Posts and Telecommunications, Xi'an, Shaanxi 710121, China;
    4. School of Computer Science, Northwestern Polytechnical University, Xi'an, Shaanxi 710072, China
  • Received:2019-04-18 Revised:2020-03-21 Online:2020-10-25 Published:2020-10-25
    • Corresponding author:
    • PU Lei
    • Supported by:
    • National Natural Science Foundation of China (No.61571458, No.61703423)

摘要: 近年来,结合深度特征的相关滤波算法由于较高的跟踪精度在视觉跟踪领域受到了广泛的关注.对训练样本的周期性假设一方面提高了计算效率,但是也引入了边界效应,限制了算法性能的进一步提升.通过对深度特征表达能力的深入挖掘,本文提出了一种新的跟踪框架.由于深层特征具有良好的语义信息,选取VGG网络第五层卷积特征提取目标的空间可靠区域,将该区域信息用于对样本进行裁剪并引入目标函数,建立空间约束模型,接着采用ADMM算法进行迭代求解.为了进一步提高算法的长时跟踪能力,提出一种简单有效的遮挡判断方法.实验结果表明,所提出的算法在跟踪精度和成功率上优于大多数先进的算法.

关键词: 视觉跟踪, 空间正则化, 深度特征, 相关滤波, 模型更新

Abstract: In recent years, the correlation filter based algorithm combined with deep features has received extensive attention. The period assumption of the training samples improves the computational efficiency, but also introduces the boundary effect, which limits the further improvement of the tracking performance. By exploring the deep feature representation ability, a new tracking framework is proposed. Since the deep features have good semantic information, the fifth layer convolution feature of VGG network is used to extract the spatially reliable region of the target, and the region information is introduced into the objective function to establish a spatial constraint model. Then iteratively solved by ADMM algorithm. In order to further improve the long-time tracking ability, a simple and effective method of occlusion detection is proposed. Experimental results show that the proposed algorithm outperforms most advanced algorithms in tracking accuracy and success rate.

Key words: visual tracking, spatial regularization, deep features, correlation filter, model update

中图分类号: