Improved Infrared Aerial Small Target Tracking Algorithm Based on Fully-Convolutional Siamese Networks

Abstract

In order to solve the background clutter interference, occlusion and other practical problems faced by infrared imaging guided missile in the process of tracking aerial small target, an improved algorithm based on fully-convolutional Siamese networks is proposed. This algorithm judges target tracking state by the average peak to correlation energy and the maximum peak of deep feature response map. When target is jammed by background clutter, the deep feature response value combined with the local contrast between target and clutter is used to select target. When target is judged to be obscured, the target position is predicted by Kalman filter. Compared with the benchmark algorithm SiamFC, the tracking success rate and accuracy of the improved algorithm are improved by 33.4% and 21.9% respectively, when tested on the infrared aerial dim small target dataset. Experimental results show that the proposed algorithm can adapt to complex and diverse infrared aerial scenes, and realize effective and stable real-time tracking for infrared aerial small target.

Keywords

• |missile|infrared aerial small target|target tracking|siamese networks|average peak to correlation energy|kalman filter