A lightweight tunnel vehicle re-ldentification model based on YOLOv11n and FaceNet

Area of Science:

• Computer Vision
• Artificial Intelligence
• Surveillance Systems

Background:

• Tunnel vehicle re-identification (Re-ID) is hindered by low-resolution images, lighting variations, and occlusions.
• Existing algorithms struggle with these complex tunnel environment challenges, limiting effectiveness.

Purpose of the Study:

• To develop a novel, intelligent framework for vehicle monitoring in tunnels.
• To enhance vehicle detection and Re-ID capabilities using lightweight and efficient models.

Main Methods:

• Integrated YOLOv11n for lightweight vehicle detection.
• Optimized FaceNet for Re-ID by replacing its backbone with MobileNetV3 and adding a Coordinate Attention module.
• Utilized a joint loss function combining IoU-based hard triplet mining and Center Loss.
• Created a tunnel-specific dataset with 12,000 images and applied data augmentation.

Main Results:

• YOLOv11n achieved 98.63% mAP at 242 fps.
• The improved Re-ID model reached 94.18% accuracy at 25.43 fps (0.81 GFLOPs, 3.51M parameters).
• Ablation studies confirmed component effectiveness, with a 2.44% improvement in AUC.

Conclusions:

• The proposed framework offers a robust solution for real-time tunnel vehicle monitoring.
• The system effectively addresses challenges like low-resolution imagery and lighting variations.
• Potential extensions include multi-modal fusion and cross-tunnel transfer learning.