This research paper presents a powerful, hybrid fatigue detection system designed to be used in real, time driver monitoring on resource, limited edge devices. To improve accuracy and save computational power, the system combines the use of lightweight Convolutional Neural Network (CNN) models with landmark, based geometric analysis. Two separate CNNs are employed by the system for eye state detection and yawning detection. Both of them are working in parallel with geometric feature extraction to get the values of the EAR (Eye Aspect Ratio) and the MAR (Mouth Aspect Ratio) from the images. Hybrid decision, making reasoning at its heart is the integration of deep learning inputs with geometric methods to give more reliable predictions of fatigue. Experiments confirm the system’s very high accuracy as the eye state detection model achieves a score of 98.34%, and the yawn detection model reaches 96.84%. Such a hybrid technique constitutes a cost, effective and efficient way of raising the level of safety on the road through the use of intelligent driver assistance systems, which can be implemented on embedded hardware.