基于局部稀疏表示的目标跟踪算法

Object tracking algorithm based on local sparse representation

根据局部稀疏表示的特点,文章提出了一种基于局部稀疏表示的目标跟踪算法,该算法利用图像的局部稀疏系数作为训练样本,在贝叶斯分类器的框架下完成跟踪任务。首先,使用字典来提取局部图像块的稀疏系数,作为图像特征;然后通过训练简单的贝叶斯分类器来区分目标与背景;最后使用两步搜索策略对目标进行准确跟踪;此外,该算法还使用了一种能够去除遮挡干扰的鲁棒性更新策略。对比实验结果表明,该算法具有较为稳定的跟踪效果。

According to the characteristics of the local sparse representation, an online object tracking algorithm based on local sparse representation is proposed. The algorithm uses image patches local sparse coefficient as training samples and completes tracking task in the framework of Bayesian classifier. Firstly, it extracts image patches local sparse coefficient as image feature by dictionary. Secondly, it distinguishes the target and background by training simple Bayesian classifier. Finally, it tracks the target accurately using two-step search strategy. It also uses a robust update strategy which can remove occlusion disturbance. The comparative experimental results show that the proposed algorithm can track objects more stably.