A LIME-Enhanced SVM Framework for Driver Drowsiness Detection in Nighttime Driving Scenarios

Nighttime traffic accidents caused by driver fatigue remain a critical issue as most visual-based detection systems find it challenging to interpret facial cues under poor lighting conditions. Key obstacles include decreased accuracy in dark settings, difficulties in detecting eye and mouth features, and the impractical nature of real-time approaches that rely on physiological sensors. This study introduces a vision-based drowsiness detection framework that integrates the Adaptive Low-light Image Enhancement (LIME) method with a Support Vector Machine (SVM) classifier employing an RBF kernel. The dataset comprises 11.566 images of eyes and mouths, which are analyzed to extract features like Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR), and blink frequency. Evaluation results show that the SVM model with the RBF kernel attained 90.94% accuracy, 91.22% precision, and 91.82% recall. This system is effective in detecting drowsiness under low-light conditions and has the potential to be implemented as an early warning feature in vehicles. Keywords: Drowsiness Detection, SVM, EAR, MAR, Adaptive LIME