LBCM: A Robust Image Based Driver Drowsiness Detection Approach Using Learning Based Blink Classification Methodology

Driver drowsiness is a significant contributor to road accidents, causing numerous fatalities and injuries each year. This paper presents the Learning-based Blink Classification Methodology (LBCM), a robust image-based approach for detecting driver drowsiness using blink classification and eye movement analysis. The system leverages real-time video input to detect facial landmarks, compute the Eye Aspect Ratio (EAR), and classify blinks into alert, drowsy, and asleep states. Using a hybrid method combining Haar Cascade and deep learning-based face detection, the system achieves accurate blink detection across varying lighting conditions. Blink classification is performed using Support Vector Machines (SVM) and Recurrent Neural Networks (RNN), achieving a precision of 96.5% for alert states, 93.2% for drowsy states, and 90.5% for asleep states. The system's overall blink detection accuracy in real-time is above 94<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> across different driving environments, including low-light and overexposed conditions. Additionally, the system performs efficiently with a frame rate of 50–60 FPS when utilizing GPU acceleration.