A High-Precision Vehicle Detection and Tracking Method Based on the Attention Mechanism

Area of Science:

• Computer Vision
• Artificial Intelligence
• Intelligent Transportation Systems

Background:

• Vehicle detection and tracking are crucial for intelligent transportation.
• Existing methods face challenges with small target vehicles.
• Attention mechanisms offer potential for improved accuracy.

Purpose of the Study:

• To propose a novel, high-accuracy vehicle detection and tracking method for small targets.
• To enhance real-time intelligent transportation management and control systems.
• To leverage attention mechanisms for improved detection and tracking performance.

Main Methods:

• Developed YOLOv5-NAM, integrating the normalization-based attention module (NAM) into YOLOv5s for vehicle detection.
• Proposed JDE-YN, a real-time small target vehicle tracking method with embedded feature extraction in the prediction head for joint training.
• Validated the methods on the UA-DETRAC dataset.

Main Results:

• The YOLOv5-NAM model showed a 1.6% increase in mAP compared to the original YOLOv5s.
• The JDE-YN tracking method achieved a 0.9% improvement in MOTA compared to the original JDE method.
• Experimental results confirm effective real-time detection of small target vehicles.

Conclusions:

• The proposed YOLOv5-NAM and JDE-YN methods significantly improve small target vehicle detection and tracking accuracy.
• The attention-based approach is effective for real-time intelligent transportation applications.
• The study demonstrates a viable solution for enhancing vehicle monitoring systems.