Confidence Evaluation Algorithm for Infrared Small Target Tracking in Complex Backgrounds

In complex background scenarios, infrared small target tracking algorithms often face challenges such as cluttered backgrounds, noise interference, tracking drift, scale changes, and occlusions. To address these issues, an anti-interference approach based on tracking confidence evaluation is adopted. Traditional tracking confidence evaluation methods often rely on intermediate results, such as response maps or values of tracking algorithms, which have limitations in terms of compatibility, accuracy, and stability. To overcome these limitations, we propose a tracking confidence evaluation algorithm called Conf-Net, which is based on contrastive learning. The proposed algorithm employs an end-to-end neural network and self-supervised training methods, directly analyzing the tracking result image patches without relying on intermediate results. Moreover, the threshold setting for ConfNet is relatively loose, making it simple and easy to use. Experimental results on public datasets demonstrate that, compared to traditional methods, Conf-Net provides a significant enhancement in tracking performance with a short computational delay (0.283 ms), proving the effectiveness of this approach. Moreover, the loose threshold setting of Conf-Net makes it particularly user-friendly.