Towards Robust Visual Object Tracking for UAV With Multiple Response Incongruity Aberrance Repression Regularization

Benefiting from high computational efficiency and satisfactory tracking accuracy, discriminative correlation filters (DCF)-based tracking methods have garnered widespread attention in the unmanned aerial vehicle (UAV) tracking community. However, omnipresent appearance variations such as occlusion, deformation, and scale changes can easily cause aberrations, leading to a decline in tracking accuracy. Moreover, consistent responses may exacerbate model overfitting issues. To address these challenges, we propose a real-time UAV tracking method leveraging the multiple response incongruity aberrance repression (MRIAR) correlation filter based on the DCF paradigm. This method comprises a feature response incongruity aberrance repression (FRIAR) regularization module and a target spatiotemporal response incongruity aberrance repression (TSRIAR) regularization module. Tracking accuracy and reliability can be enhanced by implementing aberrance repression for incongruous response map variance rates. Furthermore, a dynamic spatiotemporal variance response map is introduced into the TSRIAR module to capture the variation information of the target response between spatiotemporal frames, thereby further strengthening aberrance repression. Comprehensive experiments performed on four prevalent UAV benchmarks demonstrate that our MRIAR achieves competitive performance in terms of precision and success rate, meeting practical aerial tracking requirements at a speed of $\sim$ 40 FPS.