Online Discriminative Structured Output SVM Learning for Multi-Target Tracking

In this letter, we propose an online discriminative learning method for feature combination during multi-target tracking. Previous works utilize offline learned weights for fusion of multiple features, which is not always effective for different tracking contexts. Our work aims to update the weights adaptively in online tracking. We formulate the feature combination problem in data association using structured output SVM, and solve it by online learning algorithm. The constraints of discriminative appearance affinity are integrated to discriminate positive associations from disturbing ones, which makes association more reliable. By comparison with five state-of-the-art methods, our proposed online tracking approach outperforms the other online methods, and is competitive with the global optimal ones.