Research on driver behavior detection based on an improved YOLOv8 algorithm

In view of the shortcomings of YOLOv8 algorithm in small target detection and real-time performance in the driver behavior detection task, we propose a driver behavior detection method based on improved YOLOv8 algorithm. The main improvement measures include lightweight network, replacing the backbone network of YOLOv8 with MobileNetV3-Large, reduce the model complexity, improve reasoning speed, make it more suitable for deployment on resource-limited devices; attention mechanism, introduce SE attention mechanism, learn the dependence between channels, adaptively reset channel characteristics, improve the sensitivity of the model to the channel features; spatial attention, introduce CA attention mechanism, capture the correlation between channels and spatial long-distance dependence, and improve the ability of the model to characterize image features. The experimental results show that the improved model significantly reduces the model complexity, reasoning faster, and is more suitable for driver behavior detection tasks. Compared with the original YOLOv8 algorithm, the improved model achieved about 6% better accuracy in the distracted driving behavior detection task, reduced model size by about 55%, and inference speed improved slightly.