电子学报 ›› 2017, Vol. 45 ›› Issue (4): 799-804.DOI: 10.3969/j.issn.0372-2112.2017.04.005

• 学术论文 • 上一篇    下一篇

基于自适应分层结构的压缩分布场跟踪算法

王亚文, 陈鸿昶, 李邵梅, 高超   

  1. 国家数字交换系统工程技术研究中心, 河南郑州 450002
  • 收稿日期:2015-09-23 修回日期:2016-05-19 出版日期:2017-04-25 发布日期:2017-04-25
  • 作者简介:王亚文 男,1990年8月出生,河南郑州人.国家数字交换系统工程技术研究中心硕士研究生,主要研究方向为计算机视觉.E-mail:15738321455@163.com;陈鸿昶 男,1964年4月出生,河南新密人.国家数字交换系统工程技术研究中心教授、博士生导师,主要研究方向为电信网安全.
  • 基金资助:

    国家自然科学基金(No.61379151,No.61521003);国家科技支撑计划(No.2014BAH30B01);河南省杰出青年基金(No.144100510001)

Object Tracking by Compressive Distribution Fields with Adaptive Hierarchical Structure

WANG Ya-wen, CHEN Hong-chang, LI Shao-mei, GAO Chao   

  1. National Digital Switching System Engineering R & D Center, Zhengzhou, Henan 450002, China
  • Received:2015-09-23 Revised:2016-05-19 Online:2017-04-25 Published:2017-04-25

摘要:

为了提高分布场跟踪算法的运算效率,增强其在复杂背景下的鲁棒性,提出基于自适应分层结构的压缩分布场跟踪算法.该方法充分考虑目标区域像素值分布情况,引入k-means算法对首帧标记的目标区域进行聚类分析,根据聚类结果自适应的产生分布场结构.针对分布场模型维数较高的缺点,融合压缩感知方法对分布场进行压缩,降低模型维数,提高算法效率.此外,改变原始分布场跟踪算法采用的局部搜索跟踪策略,利用随机抽样的方式来提高算法跟踪精度.实验结果表明,提出的算法与当前流行的跟踪算法相比,具有更好的表现.

关键词: 分布场, 压缩感知, 目标跟踪, 聚类分析

Abstract:

In order to improve the efficiency of tracking algorithm based on distribution fields and the robustness of the algorithm under complex background,tracking algorithm by compressive distribution fields with adaptive hierarchical structure is presented.Distribution of pixel values in target region is considered in this method,k-means algorithm is introduced to analyse the distribution of pixel values in the first frame,adaptive hierarchical structure of distribution fields is built according to the clustering results.For the problem that the dimension of distribution field model is high,compressive sensing is combined to compress distribution fields,which can reduce the model dimension and improve the efficiency of tracking algorithm.Furthermore,local search strategy in original distribution fields tracking algorithm is changed,random sampling is used to improve the tracking accuracy.Experimental results demonstrate that the proposed algorithm outperforms the state-of-the-art tracking algorithms.

Key words: distribution fields, compressive sensing, object tracking, cluster analysis

中图分类号: