电子学报 ›› 2017, Vol. 45 ›› Issue (9): 2272-2280.DOI: 10.3969/j.issn.0372-2112.2017.09.031

• 学术论文 • 上一篇    下一篇

采用低秩与加权稀疏分解的视频前景检测算法

常侃1,2,3, 张智勇1, 陈诚1, 覃团发1,2,3   

  1. 1. 广西大学计算机与电子信息学院, 广西南宁 530004;
    2. 广西大学广西多媒体通信与网络技术重点实验室 培育基地, 广西南宁 530004;
    3. 广西大学广西高校多媒体通信与信息处理重点实验室, 广西南宁 530004
  • 收稿日期:2016-08-16 修回日期:2017-01-19 出版日期:2017-09-25
    • 通讯作者:
    • 常侃
    • 作者简介:
    • 张智勇,男,1991年9月出生于广西扶绥.现为广西大学计算机与电子信息学院计算机技术专业硕士研究生,主要研究方向为稀疏表示、低秩矩阵及应用等.E-mail:zhangzhiyong1160@163.com;陈诚,男,1993年10月出生于安徽舒城.现为广西大学计算机与电子信息学院信息处理与通信网络系统学术硕士研究生,主要研究方向为图像处理、稀疏表示等.E-mail:vigil1993@163.com;覃团发,男,1966年7月出生于广西宾阳.1997年于南京大学获博士学位,现为广西大学计算机与电子信息学院副院长、教授、博士生导师.主要研究方向为多媒体通信、无线传感器网络等.E-mail:tfqin@gxu.edu.cn
    • 基金资助:
    • 国家自然科学基金 (No.61401108); 广西自然科学基金 (No.2016GXNSFAA380154)

Video Foreground Detection by Low-Rank and Reweighted Sparse Decomposition

CHANG Kan1,2,3, ZHANG Zhi-yong1, CHEN Cheng1, QIN Tuan-fa1,2,3   

  1. 1. School of Computer and Electronic Information, Guangxi University, Nanning, Guangxi 530004, China;
    2. Guangxi Key Laboratory of Multimedia Communications and Network Technology(Cultivating Base), Guangxi University, Nanning, Guangxi 530004, China;
    3. Guangxi Colleges and Universities Key Laboratory of Multimedia Communications and Information Processing, Guangxi University, Nanning, Guangxi 530004, China
  • Received:2016-08-16 Revised:2017-01-19 Online:2017-09-25 Published:2017-09-25
    • Supported by:
    • National Natural Science Foundation of China (No.61401108); Natural Science Foundation of Guangxi Zhuang Autonomous Region,  China (No.2016GXNSFAA380154)

摘要: 传统的鲁棒主成分分析模型能较好地解决视频前景检测问题.但是,若该模型的假设条件不能满足,算法性能会变差.针对此问题,本文提出了一种低秩与加权稀疏分解模型,通过对前景矩阵加权以增强其稀疏性.在建立加权矩阵的过程中,采用光流法获取每帧的运动矢量,以区分真实运动区域.其次,进一步提出一种增强模型,通过将加权矩阵作用于观测矩阵及背景矩阵,防止前景与背景的错误分离.实验结果表明,在无噪和有噪的情况下,提出的算法均能有效地分离监控视频中的前景和背景.

关键词: 前景检测, 运动目标检测, 鲁棒主成分分析, 低秩表示, 光流法

Abstract: The traditional robust principal component analysis (RPCA)model is able to solve the video foreground detection problem well.However,if the basic assumptions are violated,this model will have poor performance.This paper proposes a low rank and reweighted sparse decomposition model,where the foreground matrix is reweighted so as to enhance its sparsity.When the weighting matrix is established,the optical flow method is used to get the motion vectors in each frame in order that the real moving areas can be recognized.Afterwards,based on the newly proposed model,an enhanced decomposition model is also developed.Since the weighting matrix is applied to both the observation matrix and the background matrix,the enhanced model is able to prevent the foreground and the background from being wrongly separated.Experimental results show that the proposed algorithm can efficiently separate foreground and background components for video clips with or without noises.

Key words: foreground detection, moving object detection, robust principal component analysis (RPCA), low-rank representation, optical flow method

中图分类号: