Lightweight Drowsy Driving Recognition System

To address the issues of excessive parameters and high computational load in fatigue driving detection models, a lightweight detection algorithm based on YOLOv8n is proposed. The AIFI module from RT-DETR is employed to enhance and optimize the SPPF layer in the backbone network, improving both speed and computational efficiency in target detection. The FasterBlock module from FasterNet is integrated to optimize the C2f module, reducing computational demands and increasing detection speed. Additionally, MPDIoU is introduced to redefine the loss function, accelerating model convergence and enhancing accuracy. Results indicate a 1.5% improvement in accuracy, a 23% reduction in computational load, a 26% decrease in parameters, and a 25% reduction in model size compared to the original YOLOv8n. This ensures accurate, real-time detection while maintaining a lightweight structure, providing robust support for vehicle-mounted systems.