Fatigue Driving Detection Methods Based on Drivers Wearing Sunglasses

Abstract

During daily driving, many drivers choose to wear sunglasses to mitigate the glare from sunlight. However, conventional visual detection methods encounter challenges in discerning fatigue among these individuals due to the obstructive nature of sunglasses. This paper presents an innovative approach that integrates Yolov8n with transfer learning to devise a precise fatigue detection system tailored for sunglasses-wearing drivers. Utilizing onboard infrared cameras, videos of such drivers were recorded, and essential facial features were extracted to construct a specialized dataset. Annotations were meticulously applied to classify three distinct states: normal, closed eyes, and yawning. Through the amalgamation of Yolov8n and transfer learning, a fatigue driving classification model was developed by integrating thresholds based on the proportion of closed-eye frames, yawning frames, and consecutive closed-eye frames for sunglasses-wearing drivers, achieving an impressive detection accuracy surpassing 98%. Experimental findings showcase the system's capability for real-time monitoring, accurately identifying instances of fatigue driving at both per-minute and per-second intervals, thereby significantly enhancing detection efficacy. This study yields valuable insights for prospective investigations in fatigue driving detection among sunglasses-wearing drivers and contributes substantively to the advancement of traffic safety technology.