This study introduces an embedded real-time driver monitoring system that synergizes deep learning with fuzzy inference techniques. The system utilizes two convolutional neural networks to analyze the driver’s eye, mouth, and head posture states. These measurements are subsequently fed into a Type-1 Mamdani Fuzzy module to classify the driving state as either safe or risky. Performance evaluation was conducted through two experiments using standard metrics from the literature. The results indicate that the proposed hybrid approach achieves an accuracy exceeding 92% in both experiments, demonstrating its robustness and suitability for real-world deployment in Advanced Driver Assistance Systems.