<![CDATA[Vehicle accidents globally are heavily caused by driver fatigue, highlighting the need for automated fatigue detection systems. This research introduces an efficient method to detect driver fatigue using facial geometric features. The proposed method integrates Haar Cascade to locate faces, followed by the extraction of Eye Aspect Ratio (EAR) together with Mouth Aspect Ratio (MAR) metrics, which are then classified through the K-Nearest Neighbor (KNN) algorithm. The experiment utilized a total collection of 2,468 facial samples structured across Open, Closed, Yawn, and No Yawn categories, which were subsequently grouped into “Drowsy” and “Not Drowsy” classes. Model evaluation was performed via Stratified 10-Fold Cross-Validation. The results show that the classifier obtained a mean accuracy rate of 88.2% ± 2.6% with an AUC of 0.93, confirming robust classification effectiveness. These findings position the integrated geometric-feature approach as a systematically validated classical baseline for driver monitoring research. While this study does not empirically measure hardware latency, the stable classification performance suggests that EAR and MAR features provide a reliable foundation for evaluating the performance-to-complexity trade-offs of more advanced models in resource-constrained environments.]]>
Copyrights © 2026
