Research on Driver Abnormal Behavior Detection Method with Efficient Multi-Scale Attention Module Based on YOLOv8

At present, a significant issue in global traffic safety is guaranteeing the safe behavior of drivers during driving, especially to monitor and prevent abnormal behaviors of drivers, such as smoking, drinking water and playing with mobile phones. Although car cameras and other sensors are currently relied on to monitor driver behavior, these methods are often difficult to accurately detect in complex driving environments due to limited coverage and real-time constraints. Aiming at the problems of drivers' abnormal behaviors in the driving process, such as complex background, light changes, excessive occlusion, motion blur and size parallax, a driver abnormal behavior detection method model is proposed based on improved YOLOv8. The EMA was incorporated into the C2f module of the backbone network to enhance the attention of the network. The improved model can effectively focus on important features at different scales. The experimental results indicate that the enhanced model with EMA's attention mechanism can improve the performance of our collected driving behavior data sets, which includes smoking, drinking water and playing mobile phone. The Precision, Recall and mAP50 indexes achieved an improvement of 2.959%, 2.526% and 3.198% respectively, which proved the effectiveness of the method.