Driver Drowsiness Detection using CNN with attention mechanism of Transformer model

Driver drowsiness is a major threat to road safety, causing innumerable accidents globally. To effectively detect and analyze driver drowsiness, we employ a multi-stage approach leveraging the MediaPipe Face Mesh model and deep learning techniques. We apply a transformer attention mechanism in conjunction with convolutional layers to capture both spatial and temporal characteristics of the extracted facial features. This hybrid deep learning architecture allows for enhanced feature extraction and attention modeling, providing a thorough analysis of the driver’s state. We train and assess our model using the YAWDD dataset. By combining the robust facial landmark detection capabilities of the Face Mesh model with the powerful feature extraction and attention mechanisms of the proposed deep learning framework, our system can accurately identify driver fatigue. This comprehensive methodology ensures reliable drowsiness detection, ultimately contributing to improved road safety.