Online non-negative discriminative dictionary learning for tracking

In this paper, online non-negative discriminative dictionary learning for tracking is proposed, which combines the advantages of the global dictionary learning model and the class-specific dictionary learning model. The previous algorithm based on general dictionary learning does not take into account the inter-class relations between classes and make full use of tag information. In order to improve the classification ability of dictionaries, the class correlation was proposed to guide the learning of discriminant dictionaries, which makes full use of the correlation and difference between the atomic classes of dictionaries and introduces the tag information of the categories to improve the discriminant ability of dictionaries. For this purpose, the Huber loss function and the Fisher weight coefficient is used in the discriminative term to improve computational efficiency. In addition, non-negative constraints is added on dictionaries to enhance the performance. The OTB50 and OTB100 datasets are used to evaluate our tracker and compare with related algorithm. The experimental results show that our method performs much better than the tracking method compared in this paper.