电子学报 ›› 2016, Vol. 44 ›› Issue (9): 2235-2241.DOI: 10.3969/j.issn.0372-2112.2016.09.031

• 学术论文 • 上一篇    下一篇

面向运动目标检测的粒子滤波视觉注意力模型

刘龙, 樊波阳, 刘金星, 杨乐超   

  1. 西安理工大学自动化学院, 陕西西安 710048
  • 收稿日期:2014-06-11 修回日期:2015-12-30 出版日期:2016-09-25 发布日期:2016-09-25
  • 作者简介:刘龙 男,1976年生,博士,副教授,研究方向为智能信息技术与嵌入式系统.E-mail:Liulong@xaut.edu.cn;樊波阳 男,1990年生,硕士研究生,研究领域为智能信息技术与嵌入式系统.;刘金星 男,1989年生,硕士研究生,研究领域为目标检测与跟踪.
  • 基金资助:

    陕西省工业科技攻关项目(NO.2016GY-128);国家自然科学基金(No.61001140);西安市技术转移促进工程重大项目(No.CX12166)

Particle Filtering Based Visual Attention Model for Moving Target Detection

LIU Long, FAN Bo-yang, LIU Jin-xing, YANG Le-chao   

  1. Xi'an University of Technology, Xi'an, Shaanxi 710048, China
  • Received:2014-06-11 Revised:2015-12-30 Online:2016-09-25 Published:2016-09-25

摘要:

视觉注意力是机器视觉领域的研究热点,对目标检测、跟踪等技术发展具有积极意义,本文面向运动目标检测问题,构建了一种基于粒子滤波的视觉注意力模型.首先依据贝叶斯估计理论,推导了基于注意力的粒子权重计算方法;然后将运动注意力和目标颜色注意力分别作为自底向上(Bottom-Up)和自顶向下(Top-Down)注意力的输入,通过重要性采样、粒子权值计算、重采样等形成粒子注意力显著图,并确定目标位置;测试结果显示本文方法能够获取比其它方法更好的目标注意力显著图,并具有准确的目标检测效果.

关键词: 注意力模型, 运动目标检测, 粒子滤波, 融合

Abstract:

Visual attention is one of the research hotspots in the field of machine vision,which is positive significance for development of target detection and target tracking.This paper presents a particle filter based visual attention model that is applied to detect moving target.Firstly,according to Bayes estimation theory,the method of particle weight calculation is established by visual bidirectional (Top-Down/Bottom-Up) fusion attention.Then motion attention and target color attention are adopted as input of the attention model,and moving target saliency is calculated through the importance sampling,particle weight calculation,resampling and particle saliency map processing.Lastly,the target position is determined by distribution of particle.Different video complex scene test results show that this method is more effective and accurate than the traditional method for detection of moving target.

Key words: motion attention model, moving target detection, particle filtering, fusion

中图分类号: