Robust Object Tracking Based On Deep Feature And Dual Classifier Trained With Hard Samples

Visual tracking has attracted more and more attention in recent years. In this paper, we proposed a novel tracker that is composed of a feature network, a dual classifier, a target location module, and a sample collecting and pooling module. The dual classifier contains two classifiers, called long-term classifier and short-term classifier, in which the long-term classifier is to maintain the long-term appearance of the target and the short-term classifier is for prompt response to the sudden change of the target. The training samples are divided into positive samples, negative samples, hard positive samples and hard negative samples and are used to train the two classifiers, respectively. Furthermore, in order to overcome the unreliability in locating the target by highest score, a density clustering method is introduced into the target locating process. Experimental results conducted on two benchmark datasets demonstrate the effectiveness of the proposed tracking method.