电子学报 ›› 2021, Vol. 49 ›› Issue (12): 2330-2338.DOI: 10.12263/DZXB.20200816

• 学术论文 • 上一篇    下一篇

复杂动态背景下基于群稀疏的运动目标检测方法

王洪雁1,2, 张海坤2, 罗宇华3, 汪祖民2   

  1. 1.浙江理工大学信息学院,浙江杭州 310018
    2.大连大学信息工程学院,辽宁大连 116622
    3.上海精密计量测试研究所,上海 201109
  • 收稿日期:2020-07-31 修回日期:2021-06-15 出版日期:2021-12-25
    • 作者简介:
    • 王洪雁 男,1979年5月生于河南南阳.2011年毕业于西安电子科技大学获信号与信息处理专业博士学位,现为浙江理工大学信息学院特聘教授、硕士生导师.主要研究方向为:阵列信号处理、机器视觉、深度学习等.E-mail:gglongs@163.com
      张海坤 男,1995年4月生于安徽阜阳.现为大连大学信息学院控制科学与工程专业硕士研究生,主要研究方向为计算机视觉.E-mail:707274821@qq.com
      罗宇华 男,1979年生于江西吉安,上海精密计量测试研究所高级工程师, 研究方向为图像处理、元器件可靠性.E-mail:luoyuhua@163.com
      汪祖民 男,1975年生于河南省信阳,大连大学信息工程学院教授、硕士生导师,研究方向为物联网、智慧城市.E-mail:wangzumin@dlu.edu.cn
    • 基金资助:
    • 国家自然科学基金 (61301258); 浙江省自然科学基金重点项目 (LZ21F010002); 中国博士后科学基金资助项目 (2016M590218)

Moving Object Detection Method Based on Group Sparseness Under Complex Dynamic Background

WANG Hong-yan1,2, ZHANG Hai-kun2, LUO Yu-hua3, WANG Zu-min2   

  1. 1.School of Information Science and Technology, Zhejiang Sci-Tech University, Hangzhou, Zhejiang 310018, China
    2.College of Information Engineering, Dalian University, Dalian, Liaoning 116622, China
    3.Shanghai Precision Metrology & Test Research Institute, Shanghai 201109, China
  • Received:2020-07-31 Revised:2021-06-15 Online:2021-12-25 Published:2021-12-25
    • Supported by:
    • National Natural Science Foundation of China (61301258); Key Program of National Natural Science Foundation of Zhejiang Province, China (LZ21F010002); Supported by China Postdoctoral Science Foundation (2016M590218)

摘要:

为提高复杂动态背景下运动目标检测精度,基于低秩及稀疏分解理论,本文提出一种基于群稀疏的运动目标检测方法. 所提方法将观测视频分解为低秩静态背景,群稀疏前景及动态背景三部分. 所提方法首先使用伽马范数近乎无偏近似矩阵秩函数,以解决核范数过度惩罚较大奇异值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;其次,为利用前景目标边界先验信息以提升运动目标检测性能,每一帧使用过分割算法生成同性区域以定义群稀疏范数并用于约束前景矩阵;再次,为避免运动目标同时出现在稀疏前景和动态背景中,引入非相干项以提升二者可分性;最后,本文利用交替方向乘子方法(Alternating Direction Method of Multipliers,ADMM)求解所得非凸目标函数. 实验结果表明,与现有主流运动目标检测算法相比,复杂动态背景下本文所提方法可较好抑制动态背景从而显著提高复杂运动背景下运动目标检测精度.

关键词: 运动目标检测, 动态背景, 低秩, 群稀疏, 超像素

Abstract:

To improve the accuracy of moving object detection under complex dynamic background, based on the theory of low-rank and sparse decomposition, a group sparse based moving object detection method is developed. The proposed method decomposes the observed video into a low-rank static background, a group sparse foreground and a dynamic background. Regarding the problem that the nuclear norm over-penalizing large singular values leads to the optimal solution of the obtained minimization problem cannot be obtained and then the detection performance is decreased,the gamma norm is introduced to acquire almost unbiased approximation of rank function. In order to utilize the object boundary prior to enhance the moving target detection performance, each frame is over-segmented into homogeneous regions which are taken to define the group sparse norm to constrain the foreground matrix. Moreover, to prevent the moving object from appearing in the sparse foreground and dynamic background simultaneously, the incoherence term is introduced to enhance the separability of them. Finally, the obtained non-convex objective function can be solved using the alternating direction multiplier method(ADMM). The experimental results show that, compared with the state-of-the-art moving target detection algorithms, the developed method can suppress the dynamic background considerably and then improve the accuracy of moving object detection significantly under complex dynamic background.

Key words: moving object detection, dynamic background, low-rank, group sparse, superpixel

中图分类号: