Aerial infrared target tracking based on a Siamese network and traditional features

Abstract Due to the low signal-to-noise ratios and fast motion of aerial targets, tracking aircraft based on infrared sequences is still a challenging task. To develop a robust and fast aerial infrared target tracking method, a new framework based on a Siamese network is proposed. Traditional features, namely, the aspect ratio and mean contrast, are used to improve the performance of the tracker. We performed experiments on three simulated infrared sequences to compare the proposed method with nine state-of-the-art tracking algorithms. To make the experimental results more credible, we also performed experiments on six real shot sequences that were not used when training the network. These experiments clearly demonstrate that our algorithm can reliably track an aerial infrared target while running at a real-time speed. The accuracy of our approach is improved by 136.3% compared with CFNet.