A Time-Efficient Deep Learning Approach for Real-Time EEG-Based Driver Drowsiness Detection

Drowsy driving significantly contributes to road accidents globally, underscoring the need for reliable real-time drowsiness detection systems. While various approaches exist, Electroencephalography (EEG) based methods have proven particularly effective due to their ability to capture brain activity that precedes physical symptoms of drowsiness. This study presents a time efficiency analysis of an EEG-based driver drowsiness detection pipeline, integrating deep feature extraction via Convolutional Neural Networks (CNNs) and classification through a Bidirectional Long Short- Term Memory (BiLSTM) network. We achieved an accuracy of 83% in classifying EEG data into alert and drowsy states using the SEED-VI G dataset. The results reveal that the total processing time is 0.3482 seconds, confirming that the pipeline meets the demands of real-time classification with latency. Our findings contribute to the development of practical, resource-efficient drowsiness detection systems, offering insights for future improvements in algorithm optimization and real-world deployment.