Learning channel-selective and aberrance repressed correlation filter with memory model for unmanned aerial vehicle...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Unmanned aerial vehicle (UAV) tracking is crucial for military and civilian applications.
• Existing UAV tracking methods struggle with limited computing power, appearance variations, and environmental volatility.
• Human brain mechanisms offer inspiration for developing more intelligent and autonomous AI systems.

Purpose of the Study:

• To propose a novel, robust, and computationally efficient UAV tracking method.
• To enhance tracking performance by integrating human brain-inspired memory and cognitive processes.
• To address challenges of limited computational resources and dynamic environments in UAV tracking.

Main Methods:

• Developed a tracking method based on Discriminative Correlation Filter (DCF) and a memory model.
• Introduced dynamic feature-channel weighting and aberrance-repressed regularization into the loss function.
• Incorporated a historical model retrieval module for improved target appearance representation and utilized the ADMM method for optimization.

Main Results:

• The proposed tracker demonstrated superior performance compared to advanced algorithms on challenging UAV benchmarks.
• Dynamic feature-channel weighting enabled the filter to focus on more reliable features.
• Aberrance-repressed regularization effectively suppressed background clutter and target appearance changes.

Conclusions:

• The novel DCF-based tracker with a memory model significantly improves UAV tracking robustness and accuracy.
• The integration of cognitive processes and memory mechanisms enhances the AI's ability to handle complex tracking scenarios.
• The method offers a computationally efficient solution for real-time UAV tracking applications.