摘要： 在基于稀疏低秩分解的运动目标检测方法中,由于核范数并非为矩阵的秩函数最佳近似,未考虑到运动目标的空间连续性,在动态背景干扰的情况下,运动目标检测的效果不理想。针对上述问题,提出加权截断p范数分析模型。该模型将观测视频分为静态背景、运动目标与动态背景3个部分,静态背景采用改进的非凸范数,即加权截断p范数进行低秩约束,根据动态背景与运动目标具有空间连续性的特点,分别使用l2,1范数进行结构性稀疏约束。实验结果表明,与鲁棒主成分分析模型、截断核范数模型、加权核范数模型以及相邻离群点低秩模型相比,该模型可有效去除动态背景扰动,并能提取到更精确的运动目标。

关键词: 背景建模, 运动目标提取, 稀疏与低秩理论, 加权截断p范数, 结构性稀疏范数

Abstract: In the moving object detection methods base on low rank and sparse decomposition,the nuclear norm is not the best approximation of the rank function of the matrix,meanwhile,the spatial continuity of moving object is not to be considered.As a result,the effect of moving object detection isn’t ideal on the circumstance of dynamic interference.To deal with this problem,a model named weighted truncated p-norm model is proposed.This model divides the observed video into three parts:the static background、the moving object and the dynamic background.The static background is constrained by an improved non-convex norm——weighted truncated p-norm.And according to the characteristic of spatial continuity,the dynamic background and moving object are constrained by structured-sparse l2,1-norm,respectively.Experimental results show that the WTPR model can remove noise of dynamic background effectively,also the more accurate moving object can be obtained than Robust Principal Component Analysis(RPCA)、Truncated Nuclear Norm Regularization(TNNR)、Weighted Nuclear Norm Minimization(WNNM)、Detecting Contiguous Outliers in the Low-rank Representation(DECOLOR).

Key words: background modeling, moving target extraction, sparse and low rank theory, weighted truncated p norm, structured-sparse norm

中图分类号: 

• TP391.41