A Distortion-Aware Dynamic Spatial-Temporal Regularized Correlation Filtering Target Tracking Algorithm

The discriminative correlation filtering target tracking algorithm can achieve a good balance between tracking accuracy and speed, and therefore has attracted much attention in the field of image tracking. The correlation of response maps can be efficiently calculated in the Fourier domain through the input discrete Fourier transform (DFT), where the DFT of the image has symmetry in the Fourier domain. However, most algorithms based on correlation filtering still have unsatisfactory performance in complex scenarios, especially in scenarios with similar background interference, background clutter, etc., where drift phenomena are prone to occur. To address these issues, this paper proposes a distortion-aware dynamic spatiotemporal regularized correlation filtering target tracking algorithm (DADSTRCF) based on Auto Track. Firstly, a dynamic spatial regularization term is constructed based on color histograms to alleviate the effects of similar background interference, background clutter, and boundary effects. Secondly, a distortion perception function is proposed to determine the degree of distortion of the current frame target, and the Kalman filter is integrated into the relevant filtering framework. When the target undergoes severe distortion, the Kalman filter is switched for tracking. Then, the alternating direction multiplier method (ADMM) is used to obtain the optimal filter solution, reducing computational complexity. Finally, comparative experiments were conducted with various correlated filtering target tracking algorithms on the four datasets of OTB-50, OTB-100, UAV123, and DTB70. The experimental results showed that the tracking precision of DADSTRCF improved by 6.3%, 8.4%, 2.0%, and 6.4%, respectively, compared to the baseline Auto Track, and the success rate improved by 9.3%, 9.3%, 2.5%, and 3.9%, respectively, fully demonstrating the effectiveness of DADSTRCF.