In Morocco, road traffic accidents remain a significant public safety concern, with driver drowsiness being a critical contributing factor. This study presents a comprehensive approach to detecting driver drowsiness using the Driver Drowsiness Dataset (DDD) and advanced deep learning techniques. The dataset underwent preprocessing, augmentation, and splitting into training, validation, and testing subsets. Two models were evaluated: MobileNet, which achieved an accuracy of 94.15%, precision of 97.05%, and recall of 92.49%, and a newly proposed CNN architecture, which outperformed MobileNet with an accuracy of 99.98%, precision of 99.96%, and recall of 100%. The study also emphasizes the integration of these detection systems with Vehicular Ad hoc Networks (VANETs) to enable real-time communication and emergency response. This integration has the potential to enhance road safety by providing timely interventions to prevent accidents caused by driver drowsiness.