基于稀疏表示和特征选择的LK目标跟踪

为了实现复杂场景中的视觉跟踪, 提出了一种以LK（Lucas-Kanade）图像配准算法为框架, 基于稀疏表示的在线特征选择机制。在视频序列的每一帧, 筛选出一些能够很好区分目标及其相邻背景的特征, 从而降低干扰对跟踪的影响。该算法分别构造前景字典和背景字典, 前景字典来自于第一帧的手动标定, 并随着跟踪结果不断更新, 而背景字典则在每一帧重新构造。同时, 一种新的字典更新策略不仅能有效应对目标的外观变化, 而且通过特征选择机制, 能避免在更新过程中引入干扰, 从而克服了漂移现象。 大量的实验结果表明, 该算法能有效应对视角变化、光照变化以及大面积的局部遮挡等挑战。

LK tracking based on sparse representation and features selection

In order to tracking object in complex scenes, the paper proposed an online features selection mechanism based on sparse representation in the Lucas-Kanade image registration framework. To reduse the impact of interference on the tracking, it selected the features that owned best discrimination betweem object and adjacent background in each frame of the video sequence. The algorithm was composed of forward dictionaries and background dictionaries, the former which would be created manually from the first frame and updated with the tracking results, the background dictionary would be reconstruct by evrey frame. Meanwhile, a new dictionary updating strategy not only can effectively cope with the appearance changes, but also handle drift. Experiment shows that the proposed algorithm can effectively deal with pose change, illumination change and large partial occlusion.