摘要
针对基于L2范数的主分量分析(L2-PCA)易受离群数据的影响,使得传统的基于L2-PCA的视觉跟踪对目标遮挡的鲁棒性较差的问题,提出一种基于L1范数最大化主分量分析(PCA-L1)的视觉跟踪算法.利用PCA-L1对目标表观建模,以粒子滤波为框架估计目标的状态;为了适应目标变化并克服"模型漂移"问题,提出一种PCA-L1的在线更新方法以实现子空间的更新.通过实验验证并与现有算法进行了比较的结果表明,文中算法具有较优的跟踪性能.
The principal component analysis based on L2-norm （L2-PCA） is sensitive to outliers, which result in the visual tracking algorithms based on L2-PCA having lower robustness to occlusions. To alleviate this problem, a novel visual tracking algorithm via Ll-norm maximization principal component analysis （PCA-L1） is proposed in this paper. The proposed algorithm models the object appearance using PCA-L1 , and infers the states of object with particle filter. In addition, to adapt to changes of object appearance and avoid model drifting, an online PCA-L1 update method is proposed. The experimental results on several challenging sequences show that the proposed algorithm has better performance than that of the state-of-the-art tracker.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2013年第9期1392-1398,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61175035)
关键词
视觉跟踪
L1范数主分量分析
模板更新
粒子滤波
visual tracking
L1-norm principal component analysis
model update
particle filter
