稀疏表示的Lucas-Kanade目标跟踪

提出一种新的目标跟踪算法,将稀疏表示应用于LK(Lucas-Kanade)图像配准框架.通过最小化校准误差的L1范数来求解目标的状态参数,从而实现对目标的准确跟踪.对目标同时建立两个外观模型:动态字典和静态模板,其中动态模型由动态字典的稀疏表示来描述目标外观.为了解决由于动态字典不断更新造成的跟踪漂移问题,一个两阶段迭代机制被采用.两个阶段所采用的目标模型分别为动态字典和静态模板.大量的实验结果表明,本文算法能有效应对外观变化、局部遮挡、光照变化等挑战,同时具有较好的实时性.

Lucas-Kanade tracking based on sparse representation

In this paper, we propose a new object tracking algorithm applying sparse representation in the Lucas-Kanade image registration algorithm. The object state parameters are solved to realize precise tracking by minimizing the L₁-norm of the alignment error. The object appearance is represented by the static template and the dynamic dictionary. The dynamic dictionary is obtained by updating the tracking result in each frame. The object can be rebuilt by the sparse representation of the templates in the dynamic dictionary. To deal with tracking drift caused by dictionary update, a two-stage iteration with the static template and the dynamic dictionary respectively is included in our method. Numerous experimental results show that the proposed method is quite effective to partial occlusions, appearance changes and illumination changes. Meanwhile the system is computational efficient and works in real time.