An Infrared Maritime Target Tracking Algorithm Based on Temporal Context

In territorial waters where suspicious incidents frequently occur, continuous tracking of vessels exhibiting suspicious behavior is necessary. However, most existing infrared image target trackers not only fail to fully utilize temporal information but also have insufficient infrared feature extraction capabilities. To address these issues, this paper proposes an infrared maritime target tracking algorithm based on temporal context. Building upon the OSTrack target tracking network, we propose using temporal contextual information for target tracking and add a time information fusion branch to leverage this information. Additionally, the CoTattention module is introduced to enhance the network's feature extraction capabilities. Finally, by introducing <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\alpha-CIoU$</tex> to optimize the IoU loss calculation, the accuracy of target bounding box prediction is improved. Our method is compared with other mainstream trackers on the LSOTB-TIR-BOAT dataset. Experimental results show that this algorithm can improve the accuracy of maritime infrared target tracking and effectively address the issue of poor robustness during tracking. This method overcomes the limitations of traditional self-attention mechanisms in feature extraction capability and the finite nature of sinale-frame infrared image information.