Detection-Driven Gaussian Mixture Probability Hypothesis Density Multi-Target Tracker for Airborne Infrared Platforms

Area of Science:

• Remote Sensing Technology
• Computer Vision
• Artificial Intelligence

Background:

• Airborne infrared platforms face challenges with irregular imaging and poor textural features for target detection.
• Effective tracking of time-sensitive ground targets is crucial for early warning and surveillance systems.

Purpose of the Study:

• To develop a robust multi-object tracking system for weak-textured infrared targets from unmanned aerial vehicles.
• To enhance detection accuracy and tracking stability in challenging infrared imaging conditions.

Main Methods:

• Enhanced YOLOv10 model incorporating DSA, c2f_fasterblock, and NMSFree modules for improved weak-textured target detection.
• Integration of detection results with GM-PHD (Gaussian Mixture Probability Hypothesis Density) tracking for rapid and stable multi-object tracking.

Main Results:

• Achieved a 2.3% improvement in detection accuracy and a 3.8% increase in recall on public infrared tracking datasets.
• Demonstrated high performance with MOTA (Multi-Object Tracking Accuracy) of 90.7% and IDF1 score of 94.6%.

Conclusions:

• The proposed algorithm significantly outperforms existing methods in effectiveness, accuracy, and robustness for infrared multi-target tracking.
• The system meets the demanding requirements of airborne infrared target tracking tasks, enhancing surveillance capabilities.