摘要
为了进一步提高多通道循环结构跟踪算法的鲁棒性,提出适用于线性和非线性条件的一种通用多通道关联滤波器(MCF).首先从时域入手推导得到MCF的通用形式;在此基础上,依靠线性核和高斯核函数分别得到线性和非线性条件下MCF最常用的两种形式.通过HOG这种多通道特征因子,在视频序列上进行实验.结果表明,在线性条件下,文中方法与频域法相比鲁棒性大致相同;在非线性条件下,该方法比现有方法具有更强的鲁棒性.
To further improve the robustness of the multi-channel circulant structure tracking, a novel general multi-channel correlation filter (MCF) is proposed, which can be applied in linear and nonlinear situations. First the general MCF is derived from the time domain; Then the mostly used two forms of MCF under linear and nonlinear situations can be got by means of linear and Gaussian kernel functions. Experiment is held on video sequences by HOG feature. Results show that proposed method has substantially the same robustness compared with frequency domain method under linear situation and is more robust than existing method under nonlinear situation.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2016年第3期456-463,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61203374)
中央高校基本科研业务费专项资金(2014G3322008)
关键词
循环结构跟踪
多通道关联滤波器
线性条件
非线性条件
时域
circulant structure tracking
multi-channel correlation filter
linear situation
nonlinear situation
time domain
