Fatigue Driving Recognition Using Improved YOLOv8

Abstract

Fatigue driving contributes to 21% of fatal accidents, leading to slower driver responses and increases crash risks. This study presents research to detect driver fatigue. Fatigue driving systems face challenges in real-time detection due to limited computing resources and poor accuracy detection. To address this, we introduce the YOLOv8-FCA model, enhancing the YOLOv8 algorithm to boost detection accuracy and efficiency in controlled environments. YOLOv8 was selected for its proven performance in object detection tasks. Our model integrates a feature pyramid network (RepFPN) to improve detection of subtle fatigue indicators and details, and utilizes the convolutional block attention module (CBAM) to focus attention on critical regions. Experimental results demonstrate a 1.3% increase in detection accuracy (mAP50), achieving 99.1%, while reducing computational costs by 40.73%, thus significantly enhancing detection and response capabilities.