Robust Visual Tracking via Weighted Extreme Learning Machine

In this paper, a model based on weighted extreme learning machine (weighted ELM) is proposed for visual tracking. The weighted ELM considers different class distributions both of the positive and negative classes, where extra weights are utilized in the framework. The proposed model simultaneously trains a certain number of weighted ELMs with different feature blocks. Moreover, the weighted multiple instance learning (weighted MI-L) scheme is utilized in choosing the network with the greatest feature block, which has the most discriminative ability. In addition, a special method of calculating global output weights in weighted ELMs is introduced into the tracking framework. Experimental results on some video clips illustrate the good tracking ability of our tracker in many challenging circumstances.