Robust and fast object tracking via co-trained adaptive correlation filter

Abstract

Recently, convolutional neural network (CNN) model based on correlation filter has achieved great success in object tracking. However, most of these methods relying on correlation filter train many classifiers individually, and fail to cope with significant appearance change of targets and complex challenging environment. In this paper, we propose a robust and fast object tracking method by training only two discriminative classifiers jointly. It poses a Tikhonov regularization on the ridge regression where each tracker tries to correct the other one during tracking. In addition, an online detector is adopted to re-detect objects in case of tracking failures and a target pyramid is builded to estimate the optimal scale of target. Extensive experiments on OTB-2013, OTB-2015 and VOT-2016 demonstrate that our method outperforms other state-of-the-art trackers with fast speed.