Research on a Recognition Algorithm for Traffic Signs in Foggy Environments Based on Image Defogging and Transformer

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Systems

Background:

• Accurate traffic sign recognition is vital for autonomous and driver-assistance systems.
• Extreme weather conditions like fog significantly challenge existing traffic sign detection algorithms, leading to missed or false detections.
• Existing methods struggle with maintaining performance in adverse visibility scenarios.

Purpose of the Study:

• To develop a robust traffic sign recognition algorithm capable of overcoming the challenges posed by foggy traffic environments.
• To improve the accuracy and reliability of traffic sign identification for enhanced vehicle safety.
• To reduce false and missing detections in fog-affected traffic scenes.

Main Methods:

• A defogging model was generated using the pix2pixHD network.
• An enhanced YOLOv5-Transformer (YOLOv5-T) algorithm was proposed by integrating a transformer module into YOLOv5's backbone.
• The pix2pixHD defogging model was combined with the YOLOv5-T algorithm for traffic sign recognition in foggy conditions.

Main Results:

• The proposed algorithm effectively mitigates the impact of fog on traffic sign recognition.
• Comparative experiments showed significant improvements over standard YOLOv5 and YOLOv5-T in moderate fog.
• The algorithm achieved a precision of 78.5%, a recall rate of 72.2%, and an mAP@0.5 of 82.8% in foggy traffic scenes.

Conclusions:

• The pix2pixHD+YOLOv5-T algorithm offers a promising solution for reliable traffic sign recognition in adverse weather.
• This approach enhances the safety and dependability of active driving assistance and driverless vehicles.
• Further research can explore optimizations for varied fog densities and other challenging environmental conditions.