中文摘要：

传统目标跟踪算法的模板更新方法易导致目标模型漂移,为此提出一种在线判别双字典学习算法更新目标模板.双字典由目标字典和投影字典组成,其中目标字典表示目标模板.根据目标和背景样本在线迭代学习双字典,保证获其对目标维持高度描述性.通过判别函数的约束,不但降低背景信息更新到目标字典中的概率,而且保证真实目标在投影近字典上的投影近似于在目标字典上的稀疏系数,背景在投影字典上的投影近似零.因为投影的运算量较低,所以利用投影字典选择与目标字典相近的候选目标可以降低算法整体运算量.实验表明,在各种复杂环境中,算法都具有较高的稳定性.

英文摘要：

Target template updating method of traditional target tracking algorithm is easy to cause the target model drift, a discrimina- tive double online dictionary learning algorithm is proposed to update the target template. Double dictionary is composed of target dic- tionary and projection dictionary, and target dictionary represents the target template. Double dictionary is iterative learned based on the target and background samples, which maintain it＇s highly descriptive to the target. The constraint discriminative function ,not only re- duce the probability of background information update to the target dictionary,but also ensure that projection of the real target on the projection dictionary nearly approximate sparse coefficient of real target on target dictionary, while projection of background on the projection approximation dictionary is nearly zero. Because projection computational cost is low, so using the projection dictionary to chosen candidate targets which are close with target dictionary can reduce the overall computation. The experimental results show that in the complex environment, the algorithm has higher stability.