An Adaptive Vehicle Detection Model for Traffic Surveillance of Highway Tunnels Considering Luminance Intensity

Area of Science:

• Computer Vision
• Artificial Intelligence
• Traffic Engineering

Background:

• Vehicle detection is crucial for traffic surveillance and safety.
• Deep learning models excel at feature extraction but struggle with abnormal lighting, particularly in highway tunnels.
• Existing models often fail to maintain performance under varying luminance intensities.

Purpose of the Study:

• To develop an adaptive vehicle detection model robust to abnormal lighting conditions in highway tunnels.
• To improve the accuracy and reliability of vehicle detection systems in challenging environments.
• To enhance road traffic surveillance and active safety management.

Main Methods:

• An adaptive vehicle detection model was proposed, categorizing image data into normal and abnormal luminance scenarios.
• An improved CycleGAN with edge loss was employed as an adaptive luminance adjustment module for abnormal lighting.
• The adjusted images were then processed using YOLOv7 for vehicle detection.

Main Results:

• The adaptive model demonstrated effective vehicle detection in abnormal luminance scenarios within highway tunnels.
• The improved CycleGAN successfully mitigated edge generation distortion.
• Compared to original YOLOv7, the model achieved a 16.3% precision increase, 1.7% recall increase, and 9.8% mAP_0.5 increase under abnormal lighting.

Conclusions:

• The proposed adaptive vehicle detection model significantly enhances detection performance in challenging low-light tunnel environments.
• The adaptive luminance adjustment module is transferable and can improve other vehicle detection models.
• This research contributes to more reliable road traffic surveillance and active safety systems.