YOLO-Ro-KCF: a lightweight gradient-guided real-time multi-object tracking framework for embedded UAV vision systems.

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Visual multi-object tracking (MOT) from unmanned aerial vehicles (UAVs) faces challenges like small targets, clutter, motion blur, and occlusions.
• Existing methods struggle in resource-constrained embedded systems.

Purpose of the Study:

• To develop a novel, lightweight, and real-time MOT framework for UAVs.
• To enhance tracking accuracy and robustness in challenging visual conditions.

Main Methods:

• Proposed YOLO-Ro-KCF framework integrating gradient-domain priors into detection and tracking.
• YOLO-Ro detector uses gamma-corrected gradient magnitude maps for improved small object detection.
• KCF-Ro tracker fuses gradient orientation cues with HOG features and employs multi-scale search.
• A dynamic fusion module adaptively reconciles detection and tracking hypotheses.

Main Results:

• Achieved state-of-the-art performance on VisDrone2019, UAVDT, and Anti-UAV benchmarks.
• Obtained 70.6% MOTA and 72.1% HOTA on VisDrone2019.
• Operated at 43 FPS on an NVIDIA Jetson AGX Xavier platform.
• Demonstrated consistent performance gains (2.1-4.5%) across cross-dataset tests.

Conclusions:

• YOLO-Ro-KCF provides a practical and deployable solution for high-accuracy, low-latency UAV-based visual tracking.
• The integration of gradient-domain priors significantly improves robustness against scale variations and occlusions.