Target tracking via two-branch spatio-temporal regularized correlation filter network

Abstract

In this paper, a spatio-temporal regularized correlation filter for object tracking method based on two-branch Siamese fully convolutional network learning. Firstly, a correlation filter layer is added into the Siamese fully convolutional network to achieve end-to-end learning representation; secondly, the semantic feature is combined with the appearance feature to further enhance the discriminative ability of Siamese fully convolutional network; finally, the spatio-temporal regularized correlation filter is utilized to reduce the training time and improve the tracking performance. Extensive experiments conducted on VOT2017 and OTB2015 dataset demonstrate the superior performance of the proposed approach over the examined state-of-the-art approaches.