Answers are grounded in this paper and cite their sources.
Accurate tracking of small infrared drone targets is essential for reliable perception in challenging imaging environments. However, existing transformer-based trackers are mainly developed for general visual tracking and often suffer from insufficient representation of weak target structures, complex background interference, and inaccurate localization in infrared small-drone scenarios. In this paper, we present TranSDT, a transformer-based tracker designed for small-drone tracking in infrared videos. To enhance the representation of tiny targets with limited structural information, a pyramid convolutional structural self-attention mechanism is introduced to capture multi-scale structural cues and improve target-background discrimination. Furthermore, target-background feature interaction is incorporated into the attention computation to enhance robustness against complex infrared backgrounds. A deformation loss is also adopted to reduce localization errors by maintaining shape consistency between predicted and ground-truth bounding boxes. Experiments on the Anti-UAV410 benchmark demonstrate that TranSDT achieves a precision of 81.7%, outperforming most existing transformer-based trackers. The results verify the effectiveness of the proposed method and demonstrate its potential for small infrared drone tracking in practical applications such as infrared surveillance and low-altitude security monitoring.
No takes yet. Share an insight, caveat, or question.
Ma et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: