Bioinspired Motion Information Guided Tracking Algorithm for Small Targets in Infrared Images

Researchers proposed the one-stream tracking framework to solve the problem that the features extracted by the two-stream tracking framework have limited target-background discriminability. OSTrack is a representative one-stream object tracking algorithm that unifies feature learning and relation modeling by bridging template-search image pairs with bidirectional information flows. Although OSTrack can improve target awareness, as the target is getting smaller and smaller, merely relying on the appearance information of the target is insufficient to track the target. In this paper, we propose an infrared small target tracking framework guided by bioinspired motion information, which introduces the motion information extracted by the magnocellular pathway. The spatial feature transformation is used to integrate the motion features with the appearance features. The proposed method achieved an accuracy rate of 84.70^% and the success rate of 82.22% on the Anti-UAV dataset. The accuracy and success rate on small targets (targets less than 10×10 pixels) in the Anti-UAV data set are improved by 35.88% and 27.25% compared with the benchmark algorithm. The tracking performance of this algorithm is higher than mainstream tracking algorithms such as SiwnTrack, TransT, and MixFormer. The results indicate that introducing bioinspired motion information can effectively improve the tracking performance on small targets.

1. Introduction

The infrared images are obtained from radiation emitted from objects. The infrared sensors can be used in challenging environments, including low light, dust, or smoke. Infrared imaging technology is widely used in military detection, modern traffic, public security investigation, intelligent security, and other fields. The infrared sensor is an irreplaceable method to detect and track UAV targets.