Elastic regularization networks for enhanced UAV visual tracking.

Discriminative correlation filter (DCF) algorithms have demonstrated promising performance in drone visual tracking tasks. However, most DCF-based trackers rely on predefined regularization terms and update their appearance models frame-by-frame, leading to increased computational complexity. This study aims to address these limitations. An elastic regularization network is introduced, enforcing sparsity and temporal smoothness in the objective function. This network is optimized using the augmented Lagrangian method to enhance the efficiency of the discriminative tracker. During feature extraction, color name (CN) features are combined with lower-dimensional fast histogram of oriented gradient (fHOG) features to enrich sample information. Principal component analysis (PCA) is employed for dimension reduction to mitigate time complexity and storage demands caused by high-dimensional data. Experiments conducted on multiple benchmark datasets validate the proposed approach, demonstrating both its effectiveness and robustness. On the DTB70 dataset, the proposed method achieves a precision of 0.747 and a success rate of 0.789, representing improvements of 1% and 2.9%, respectively, over the STRCF algorithm. The proposed tracker, leveraging elastic regularization networks, ensures high tracking accuracy and speed, making it suitable for real-time UAV applications.