王洪雁^{1, 2,,},
张海坤¹
1.大连大学信息工程学院 大连 116622
2.浙江理工大学信息学院 杭州 310018
基金项目:国家自然科学基金(61301258, 61271379)，中国博士后科学基金(2016M590218)，重点实验室基金(61424010106)，河南省高等学校重点科研项目支持计划(14A520079)，河南省科技攻关计划(162102210168)

详细信息

作者简介:王洪雁：男，1979年生，副教授，博士，主要研究方向为MIMO雷达信号处理，毫米波通信，机器视觉
张海坤：男，1995年生，硕士生，主要研究方向为图像处理，计算机视觉

通讯作者:王洪雁　gglongs@163.com

中图分类号:TN911.73; TP391

计量

文章访问数:572
HTML全文浏览量:238
PDF下载量:68
被引次数:0

出版历程

收稿日期:2019-06-20
修回日期:2020-04-20
网络出版日期:2020-08-29
刊出日期:2020-11-16

Moving Object Detection Method Based on Low-Rank and Sparse Decomposition in Dynamic Background

Hongyan WANG^{1, 2,,},
Haikun ZHANG¹
1. College of Information Engineering, Dalian University, Dalian 116622, China
2. School of Information Science and technology, Zhejiang Sci-Tech University, Hangzhou 310018, China
Funds:The National Natural Science Foundation of China (61301258, 61271379), China Postdoctoral Science Foundation (2016M590218), The Key Laboratory Foundation (61424010106), The Henan Province Support Plans for Key Scientific Research Projects of Colleges and Universities (14A520079), The Henan Province Plans for Science and Technology Development (162102210168)


摘要
摘要:针对背景运动引起动目标检测精度显著下降的问题，该文提出一种基于低秩及稀疏分解的动目标检测方法。所提方法首先引入伽马范数($\gamma {\rm{ - norm}}$)近乎无偏地逼近秩函数以解决核范数过度惩罚较大奇异值从而导致所得最小化问题无法获得最优解进而降低检测性能的问题，而后利用${L_{{1 / 2}}}$范数抽取稀疏前景目标以增强对噪声的稳健性，同时基于虚警像素所具有稀疏且空间不连续特性提出空间连续性约束以抑制动态背景像素，进而构建目标检测模型。最后利用基于交替方向最小化(ADM)策略扩展的增广拉格朗日乘子(ALM)法对所得优化问题求解。实验结果表明，与现有主流算法对比，所提方法可显著改善动态背景情况下动目标检测精度。
关键词:前景检测/
动态背景/
低秩/
稀疏/
L_1/2正则化
Abstract:Focusing on the issue that the detection accuracy of moving object is significantly reduced by background motion, a low-rank and sparse decomposition based moving object detection method is developed. Firstly, in order to solve the problem that the nuclear norm over-penalizing large singular values lead to the optimal solution of the obtained minimization problem can not be obtained and then the detection performance is decreased, the gamma norm ($\gamma {\rm{ - norm}}$) is introduced to acquire almost unbiased approximation of rank function. In what follows, the ${L_{{1 / 2}}}$ norm is used to extract the sparse foreground object to enhance the robustness to noise, and the spatial continuity constraint is proposed to suppress dynamic background pixels such that the moving object detection model can be constructed on the basis of the sparse and spatially discontinuous nature of the false alarm pixels. After that, the Augmented Lagrange Multiplier (ALM) method, which is the extension of the Alternating Direction Minimizing (ADM) strategy, can be employed to deal with the acquired constrained minimization problem. Compared with some state-of-the-art algorithms, the experimental results show that the proposed method can significantly improve the accuracy of moving object detection in the case of dynamic background.
Key words:Foreground detection/
Dynamic background/
Low-rank/
Sparsity/
L_1/2 regularization



PDF全文下载地址:

https://jeit.ac.cn/article/exportPdf?id=8ce80d20-6f71-4ac2-b564-7c352f13f754