Twofold Structured Features-Based Siamese Network for Infrared Target Tracking

TL;DR

This paper introduces a twofold structured features-based Siamese network that enhances infrared target tracking by fusing multi-level features and updating templates to handle appearance variations, achieving improved accuracy and speed.

Contribution

The paper proposes a novel feature fusion network and a multi-template update mechanism specifically designed for infrared target tracking, addressing challenges of appearance changes and tracking drift.

Findings

Achieves a good balance between tracking performance and real-time speed.

Outperforms other state-of-the-art trackers on VOT-TIR 2016 dataset.

Effectively handles target appearance variations and reduces tracking failures.

Abstract

Nowadays, infrared target tracking has been a critical technology in the field of computer vision and has many applications, such as motion analysis, pedestrian surveillance, intelligent detection, and so forth. Unfortunately, due to the lack of color, texture and other detailed information, tracking drift often occurs when the tracker encounters infrared targets that vary in size or shape. To address this issue, we present a twofold structured features-based Siamese network for infrared target tracking. First of all, in order to improve the discriminative capacity for infrared targets, a novel feature fusion network is proposed to fuse both shallow spatial information and deep semantic information into the extracted features in a comprehensive manner. Then, a multi-template update module based on template update mechanism is designed to effectively deal with interferences from target appearance changes which are prone to cause early tracking failures. Finally, both qualitative and quantitative experiments are carried out on VOT-TIR 2016 dataset, which demonstrates that our method achieves the balance of promising tracking performance and real-time tracking speed against other out-of-the-art trackers.