计算机科学与探索 ›› 2016, Vol. 10 ›› Issue (12): 1744-1751.DOI: 10.3778/j.issn.1673-9418.1603092

• 人工智能与模式识别 • 上一篇    下一篇

自适应低秩稀疏分解在运动目标检测中的应用

金  静1+,党建武1,王阳萍1,2,翟凤文1   

  1. 1. 兰州交通大学 电子与信息工程学院,兰州 730070
    2. 兰州宇信信息技术有限责任公司,兰州 730000
  • 出版日期:2016-12-01 发布日期:2016-12-07

Application of Adaptive Low-Rank and Sparse Decomposition in Moving Objections Detection

JIN Jing1+, DANG Jianwu1, WANG Yangping1,2, ZHAI Fengwen1   

  1. 1. School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China
    2. Lanzhou Yuxin Information and Technology Co., Ltd., Lanzhou 730000, China
  • Online:2016-12-01 Published:2016-12-07

摘要: 针对视频处理中运动目标的精确检测这一问题,提出了一种自适应的低秩稀疏分解算法。该算法首先用背景模型与待求解的帧向量构建增广矩阵,然后使用鲁棒的主成分分析(robust principal component analysis,RPCA)对降维后的增广矩阵进行低秩稀疏分解,分离出的低秩部分和稀疏噪声分别对应于视频帧的背景和运动前景,然后使用增量奇异值分解方法用当前得到的背景向量更新背景模型。实验结果表明,该算法能更好地处理光线变化、背景运动等复杂场景,并有效降低算法的延迟和内存的占用。

关键词: 运动目标检测, 低秩稀疏分解, 自适应的鲁棒主成分分析

Abstract:  Focusing in the precise detection of moving objects in video processing, this paper proposes an adaptive low-rank and sparse decomposition algorithm. It constructs an augmented matrix by background model and to-be-computed background vector, and executes low-rank and sparse decomposition to dimension-decreasing augmented matrix by RPCA (robust principal component analysis) and obtains low-rank part and sparse noise which corresponding to background and moving foreground. Then it updates background model by new background vectors using incremental SVD (singular value decomposition). The experimental results indicate that the proposed method can process complicated scenes including illumination change and background moving better, and reduces the latency of algorithm and memory occupation at meanwhile.

Key words: detection of moving objects, low-rank and sparse decomposition, adaptive robust principal component analysis