摘要
针对复杂场景下目标跟踪过程中目标遮挡、光照变化、快速运动等问题,提出一种压缩感知的矩阵低秩稀疏分解目标跟踪算法.该算法对跟踪区域提取特征向量压缩感知,用压缩域特征构建目标外观模型产生观测矩阵.采用非精确增广拉格朗日乘子法对观测矩阵低秩稀疏分解,获得各个候选目标的稀疏误差向量并构建误差矩阵.通过求解误差矩阵最小1-范数问题得到目标估计,并对目标模板字典在线更新适应目标外观变化.实验结果表明,算法在目标发生部分遮挡、光照变化、快速运动等复杂情况下,能够实现目标的鲁棒跟踪.
In order to solve target tracking under complex scene with occlusion,illumination changes and fast motion, a sparse and low- rank matrix decomposition tracking method with compressed sensing is proposed in this paper. Compressed sensing method is adopted for the image feature vector and using compressed domain characteristics build targets appearance model to create observation matrix for sparse and low-rank matrix decomposition. Inexact augmented lagrange multiplier is adopted for sparse and low-rank matrix de- composition to get sparse error vector of each candidate targets and build a error matrix of candidate goal set. By solve the matrix Min- imum I-norm problem to estimate target,and the target template dictionary online update to adapt the target appearance changed. The qualitative and quantitative experiment evaluations demonstrate that the proposed algorithm can track objects in complex scene with oc- clusion, illumination changes and fast motion.
出处
《小型微型计算机系统》
CSCD
北大核心
2017年第4期881-885,共5页
Journal of Chinese Computer Systems
基金
沈阳理工大学博士启动专项基金项目(BS-2015-03)资助
关键词
目标跟踪
压缩感知
矩阵低秩稀疏分鳃
稀疏表示
增广拉格朗日乘子法
target tracking
compressed sensing
sparse and low-rank matrix decomposition
sparse representation
augmented lagrange multiplier
