MACTrack: Spatiotemporal context propagation with motion compensation for anti-UAV tracking.

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Unauthorized unmanned aerial vehicle (UAV) activities pose significant security threats.
• Existing anti-UAV visual tracking systems struggle with agile motion, complex backgrounds, and platform-UAV motion coupling, leading to tracking drift.

Purpose of the Study:

• To develop a robust anti-UAV tracking framework that addresses the limitations of current methods.
• To improve target localization accuracy and tracking stability in challenging anti-UAV scenarios.

Main Methods:

• Proposed MACTrack framework integrating motion awareness with spatio-temporal context modeling.
• Asymmetric spatial context module utilizing self-attention and cross-attention for enhanced discrimination.
• Temporal context propagation module with memory-guided attention for adaptive historical information aggregation.
• Motion compensation module for aligning feature representations to mitigate platform motion effects.

Main Results:

• MACTrack achieved 88.9% accuracy and 66.5% success rate on the ANTI-UAV410 benchmark.
• Outperformed a competitive baseline by 7.9% in accuracy and 7.4% in success rate.
• Demonstrated significant improvements in handling performance degradation in anti-UAV tracking.

Conclusions:

• The proposed MACTrack framework effectively addresses performance degradation in anti-UAV tracking.
• Joint integration of motion awareness and spatio-temporal context modeling is crucial for robust UAV tracking.
• MACTrack offers a promising solution for enhanced security against unauthorized UAV activities.