摘要
针对传统背景建模存在的问题,文中基于低秩矩阵恢复原理,直接从视频序列中分离出前景物体和背景模型。已有低秩矩阵恢复算法的迭代计算过程中涉及大量的奇异值分解,而这些奇异值分解一般非常耗时且不够简洁,文中在非精确增广拉格朗日乘子法中引入线性时间奇异值分解算法,以得到更加有效的背景建模算法。基于实际视频序列实验,结果表明该改进算法具有更好的建模效果和较少的运算时间。
In this paper, a novel method is present based on low-rank matrix recovery, which can directly obtain background model as well as foreground objects from the video sequence. As the main computation of existing algorithms of low-rank matrix recovery is the singular value decomposition, most of which are time-consuming and not concise enough,the linear time SVD algorithm is introduced to the inexact augmented Lagrange multiplier method, and we get a more efficient background modeling algorithm. We test our algorithm on real video, and the experimental results show that our approach obtains good results and less time-consuming, compared to the exact and inexact augmented Lagrange multiplier method.
出处
《南京邮电大学学报（自然科学版）》
北大核心
2013年第2期86-89,96,共5页
Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition
基金
江苏省自然科学基金(BK2011758)资助项目
