电子学报 ›› 2016, Vol. 44 ›› Issue (3): 627-632.DOI: 10.3969/j.issn.0372-2112.2016.03.020

• 学术论文 • 上一篇    下一篇

运动目标检测的l0群稀疏RPCA模型及其算法

周伟1, 孙玉宝1,2, 刘青山1, 吴敏3   

  1. 1. 江苏省南京信息工程大学大数据分析技术重点实验室, 江苏南京 210044;
    2. 江苏省大气环境与装备技术协同创新中心, 江苏南京 210044;
    3. 南京军区南京总医院医学工程科, 江苏南京 210002
  • 收稿日期:2015-01-20 修回日期:2015-05-22 出版日期:2016-03-25
    • 通讯作者:
    • 孙玉宝
    • 作者简介:
    • 周伟 男,1990年生于江苏苏州.南京信息工程大学硕士研究生在读,研究方向为图像处理与模式识别、稀疏表示与压缩感知术. E-mail:zyxxpyzhouwei@gmail.com;刘青山 男,1975年生于安徽合肥,教授,博士生导师,研究方向为图像与视频分析、大数据处理与分析. E-mail:qsliu@nuist.edu.cn;吴敏 女,1973年生于江苏南通.南京军区南京总医院,高级工程师,研究方向为压缩感知理论与应用、EEG信号处理. E-mail:njzywm@163.com
    • 基金资助:
    • 国家自然科学基金项目 (No.61300162,No.81201161); 江苏省自然科学基金 (No.BK2012045,No.BK20131003); 中国博士后基金 (No.20110491429); 江苏省博士后基金 (No.1101083C); 江苏省光谱成像与智能感知重点实验室基金 (No.30920130122003)

l0 Group Sparse RPCA Model and Algorithm for Moving Object Detection

ZHOU Wei1, SUN Yu-bao1,2, LIU Qing-shan1, WU Min3   

  1. 1. Jiangsu Key Laboratory of Big Data Analysis Technology, Nanjing University of Information Science & Technology, Nanjing, Jiangsu 210044, China;
    2. Jiangsu Collaborative Innovation Center on Atmospheric Environment and Equipment Technology, Nanjing, Jiangsu 210044, China;
    3. Department of Medical Engineering, Nanjing General Hospital of Nanjing Area Command, Nanjing, Jiangsu 210002, China
  • Received:2015-01-20 Revised:2015-05-22 Online:2016-03-25 Published:2016-03-25
    • Supported by:
    • National Natural Science Foundation of China (No.61300162, No.81201161); Natural Science Foundation of Jiangsu Province,  China (No.BK2012045, No.BK20131003); Post-doctoral Foundation of China (No.20110491429); Postdoctoral Foundation of Jiangsu Province (No.1101083C); Key Laboratory Fund for Spectral Imaging and Intellisense of Jiangsu Province (No.30920130122003)

摘要:

经典的鲁棒主成分分析(Robust Principal Component Analysis,RPCA)目标检测算法使用l1范数逐一判别每一像素点是否属于运动目标,未能考虑到运动目标在空间分布的连续性,不利于提升运动目标检测的鲁棒性.本文提出了一种基于l0群稀疏RPCA模型的运动目标检测方法.首先运用Ncuts算法进行区域过分割,生成多个同性区域,将其作为群稀疏约束的分组信息;第二步构造基于l0群稀疏RPCA模型,运用群稀疏准则判别过分割后的各同性区域是否为运动目标,采用交替方向乘子算法对模型进行快速求解,约束过分割形成的同性区域具有相同检测结果,进而将背景环境和运动前景分离,能够更加准确地度量运动目标的区域边界,且对复杂的背景扰动更加鲁棒,达到了运动目标鲁棒检测的目的.

关键词: RPCA模型, l0群稀疏, 过分割, 交替方向乘子法, 运动目标检测

Abstract:

Classical robust principal component analysis (RPCA) algorithm uses l1-norm to discriminate whether each pixel belongs to motion object,without taking advantage of the spatial continuity distribution of the movement object.Thus,the accuracy and robustness of detection algorithm is negatively influenced.In order to use the spatial continuity prior,this paper proposes an object detection method based on l0 group sparse regularized RPCA model.First,our algorithm over-segments the video sequence into multiple homogeneous groups using normalized cuts over-segmentation algorithm,and these groups are treated as grouping information for group sparse constraint.The next step is to construct l0 group sparse regularized RPCA model,which uses l0 group sparse criterion to discriminate whether each homogeneous group belongs to motion object.Alternative direction of multiplier method (ADMM) is adopted to solve our l0 group sparse RPCA model quickly.Homogeneous groups are restricted to have the same detection results.Thus,our model can be more robust to background movement and increase the detection accuracy of object boundary.

Key words: RPCA model, l0 group sparse, over-segmentation, alternative direction of multiplier method (ADMM), moving object detection

中图分类号: