IRSDT: A Framework for Infrared Small Target Tracking with Enhanced Detection.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Infrared Imaging

Background:

• Small target detection in infrared imagery is difficult due to low resolution and lack of features.
• Complex backgrounds further challenge the accurate identification and tracking of these targets.

Purpose of the Study:

• To develop a robust framework for real-time infrared small target detection and tracking.
• To improve the accuracy and reliability of tracking small vehicle targets in challenging environments.

Main Methods:

• A three-component framework: full-image detection, cropped-image detection/tracking, and trajectory prediction.
• Utilized a CNN-based real-time detection model for initial object region identification.
• Employed the KCF algorithm and a lightweight CNN for precise target locking and an optimized Kalman filter for trajectory estimation.

Main Results:

• The proposed framework demonstrated effective real-time detection and tracking of infrared vehicle small targets.
• The system showed adaptability in scenarios with temporary target disappearance and interference from other vehicles.
• Validation on a public dataset confirmed the steady tracking performance.

Conclusions:

• The developed framework offers a reliable solution for infrared small target detection and tracking.
• The method successfully addresses challenges posed by low resolution and complex backgrounds.
• The system's adaptability makes it suitable for real-world applications requiring robust vehicle tracking.