Domain-Adaptive Object Detection for Autonomous Vehicles Under Foggy Conditions Using Lightweight Convolutional Neural Networks

Key Points

• To enhance object detection accuracy for autonomous vehicles operating in foggy conditions.
• Developed a domain-adaptive YOLOv8 framework.
• Tested performance under dense fog compared to clear-weather training results.
• Achieved real-time inference speed during object detection.
• Demonstrated meaningful accuracy retention in fog with the adapted model.
• Real-time inference speeds maintained even under adverse weather conditions.