PISTTN: Profile-Aware Infrared Small Target Tracking Network Using Spatiotemporal Context Information

Infrared small target detection and tracking play an increasingly important role in both military and civilian applications. However, challenges persist due to the small target size and low signal-to-noise ratio. For single-target detection and tracking, most existing methods require annotation in the initial frame. For multi-target detection and tracking, detectors often need to perform detection on each frame before tracking, which loses temporal features and struggles to handle occlusion effectively. Moreover, in some scenarios, the target often degenerates into a single point, posing significant challenges for detection and tracking. To address the challenges, we reformulate the infrared small target tracking task as a spatiotemporal profile detection problem, and proposes a novel infrared small target tracking network that unifies tracking and detection into a single end-to-end trainable architecture, termed the Profile-aware Infrared Small Target Tracking Network (PISTTN). Specifically, to address the loss of spatiotemporal information caused by single-frame detection in traditional tracking algorithms, we introduce a spatiotemporal tensor encoding module. This module automatically constructs sparse tensors based on target characteristics and employs 3D sparse convolution to extract profile-aware To address the challenges in detecting point-like targets, we propose a small target query module that integrates multi-scale features to enhance adaptability and generalization across varying target appearances, while generating distinct queries for different targets. In addition, we incorporate a profile detector to predict the spatiotemporal profile of targets, enabling accurate trajectory estimation through an efficient tracking strategy. Experimental results on multiple datasets demonstrate that the proposed network outperforms existing state-of-the-art methods in terms of visual and quantitative assessment.