基于l1/2-TV正则化RPCA的运动目标检测
Moving Target Detection Based on l 1/22-TV Regularization RPCA
赵俊豪 1蒋峥 1刘斌 1张玲2
作者信息
- 1. 武汉科技大学信息科学与工程学院,湖北 武汉 430081;武汉科技大学冶金自动化与检测技术教育部工程研究中心,湖北 武汉 430081
- 2. 武汉科技大学计算机科学与技术学院,湖北 武汉 430081
- 折叠
摘要
针对复杂环境下背景干扰导致运动目标检测精度下降的问题,提出了一种基于l1/2-TV正则化RPCA的运动目标检测方法.方法利用核范数来描述背景的低秩特性,采用l1/2范数描述更稀疏的运动目标,以抑制运动目标中的背景干扰.同时结合TV正则化约束运动目标的空间连续性,使运动目标更加完整.利用Frobenius范数检测背景干扰.采用交替方向最小化策略扩展的增广拉格朗日乘子法求解所提出的约束最小化问题.实验结果表明,所提方法能有效去除背景干扰,提高运动目标的检测精度、改善视觉效果.
Abstract
Focusing on the issue that the accuracy of moving target detection decreases due to background inter-ference in complex environment,a moving target detection method based on l1/2-TV regularized RPCA is proposed.The method represents video sequences as three components,i.e.,a low-rank background,a moving target,and noise.The algorithm uses the kernel norm to describe the low rank characteristics of the background.The l1/2 norm is used to describe the more sparse moving target to suppress the background interference in the moving target.At the same time,TV regularization is combined to constrain the spatial continuity of the moving target to make the moving target more complete.Frobenius norm is used to detect background interference.The augmented Lagrangian multiplier meth-od extended by the alternating direction minimization strategy is used to deal with the proposed constraint minimization problem.The experimental results show that the proposed method can effectively remove the background interference,improve the detection accuracy of moving targets and improve the visual effect.
关键词
运动目标检测/复杂环境/鲁棒主成分分析/全变分Key words
Moving object detection/Complex environment/Robust principal component analysis/Total variation引用本文复制引用
基金项目
国家自然科学基金青年基金(61902286)
出版年
2024
NETL
NSTL
万方数据