Research on Driving Obstacle Detection Technology in Foggy Weather Based on GCANet and Feature Fusion Training

Area of Science:

• Computer Vision
• Autonomous Driving Systems
• Artificial Intelligence

Background:

• Foggy weather degrades visual sensor image quality, challenging obstacle detection for autonomous vehicles.
• Existing defogging methods can lead to information loss, further complicating accurate detection.

Purpose of the Study:

• To propose a robust method for detecting driving obstacles specifically in foggy weather conditions.
• To enhance the accuracy and reliability of autonomous driving perception systems under adverse visibility.

Main Methods:

• A novel approach combining the GCANet defogging algorithm with an edge and convolution feature fusion detection algorithm.
• Utilizing the YOLOv5 network trained on clear day and edge feature images for fused feature detection.
• Ensuring algorithm compatibility by considering target edge feature characteristics post-defogging.

Main Results:

• The proposed method demonstrated a 12% improvement in mean Average Precision (mAP) and a 9% increase in recall compared to conventional training.
• The system effectively identifies image edge information post-defogging, enhancing detection accuracy.
• The approach maintains time efficiency, crucial for real-time autonomous driving applications.

Conclusions:

• The integrated defogging and detection method significantly improves obstacle detection in fog.
• This technique enhances the safe perception of driving obstacles in adverse weather, contributing to autonomous driving safety.
• The method offers a practical solution for real-world autonomous driving challenges posed by fog.