基于核协作表示的目标跟踪算法

Object Tracking Algorithm Based on Kernel Collaborative Representation

大多数现有的基于稀疏表示的视觉跟踪算法大都通过模板的线性组合来重构目标,但是没有考虑模板与模板之间以及模板与候选目标之间的非线性关系,造成算法对目标的判别能力下降,在复杂环境下容易跟踪失败.为了解决上述问题,提出一种结合核协作表示的目标跟踪算法,利用核函数将候选目标与模板映射到高维核空间,得到它们的非线性表示,并在高维核空间求解目标的稀疏系数,提高算法对目标的判别能力.为提高跟踪速度,选用l2最小化方法.实验结果表明,本文算法在跟踪精度与鲁棒性方面都有较大提高.

Most existing visual tracking algorithm based on sparse representation just used a linear combination of the templates for re- constructing the target. However,the nonlinear relationship between the templates and the candidates which may cause a decline about the discriminating ability, and further lead to the tracking task failure in the complex environment are neglected. To address this problem we discussed, an innovative tracking algorithm based on kernel collaborate is proposed in this paper. The templates and the candicates are mapped to a high-dimensional kernel space by using a kernel function, and can obtain the nonlinear representation of the templates and the candidates. Then,the sparse coefficient of the target which improve the ability of discriminant for a fight target can be achieved in a high-dimensional kernel space. Moreover, al2 norm minimization method is used for improving computation speed. The experimental result shows that the proposed algorithm is improved both in tracking accuracy and robustness in a sense.