Visual target tracking via weighted non-sparse representation and online metric learning
Jingdi Duan, Baojie Fan, Yang Cong · 2013
In this paper, we propose online metric learning tracking method that consider visual tracking as a similarity measurement problem, and incorporates adaptive metric learning and generative histogram model based on non-sparse linear representation into the target tracking framework. We propose a generative histogram model based on non-sparse linear representation, which make full use of the non-sparse coefficients to discriminate between the target and the background. The similarity metric is adaptively learned online to maximize the margin of the distance between the foreground target and background. A bi-linear graph is defined accordingly to propagate the label of each sample. The model can also self-update using the more confident new samples. Numerous experiments on various challenging videos demonstrate that the proposed tracker performs favorably against several state-of-the-art algorithms.