SAMViTrack: A Search-Region Adaptive Mamba-ViT Tracker for Real-Time UAV Tracking

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Real-time object tracking is essential for Unmanned Aerial Vehicle (UAV) applications.
• Challenges include unpredictable target motion and rapidly changing environmental conditions.
• Existing trackers often struggle with accuracy, robustness, and computational efficiency.

Purpose of the Study:

• To propose a novel object tracking framework for real-time UAV applications.
• To enhance tracking performance under challenging conditions like occlusions and scale variations.
• To develop a tracker suitable for resource-constrained UAV platforms.

Main Methods:

• Introduced the Search-Region Adaptive Mamba-ViT Tracker (SAMViTrack).
• Combined Mamba attention for efficiency with Vision Transformer (ViT) for feature extraction.
• Implemented a dynamic search region adjustment based on target motion and environmental context.

Main Results:

• SAMViTrack demonstrated superior accuracy and robustness compared to existing trackers.
• The adaptive search mechanism significantly reduced computational overhead.
• Achieved real-time performance suitable for UAVs with limited onboard resources.

Conclusions:

• SAMViTrack offers a significant advancement in UAV object tracking.
• The integration of Mamba and ViT provides a balance of speed, accuracy, and adaptability.
• The proposed tracker sets a new benchmark for real-time tracking in dynamic environments.