Attention-Driven Convolutional Neural Network for Real-Time Driver Drowsiness and Inattention Awareness

The driver drowsiness aspect has become a significant cause of road accidents in most countries across the globe, and therefore, real-time monitoring and control systems are required to address this issue. In the current research, a CNN Attention-based architecture of high accuracy multi-class driver behavior recognition is suggested, relying on the use of spatial feature representation and attention to pay attention to critical areas of the face, hands, and eyes. This model is trained and tested on the Kaggle Driver Inattention Detection Dataset which has 22,611 images of ten classes, namely, safe driving, texting, talking on the phone, operating the radio, drinking, and other inattentive behaviors. Several experiments prove that the proposed model has a high classification accuracy of 96 % and high F1-scores in comparison with the base CNN, ResNet50, and Autoencoder models. The indicators of the model, its strengths, discriminative capacity and reliability are confirmed by confusion matrices, ROC-AUC and precision-recall analysis under the variable driver behaviors and lighting conditions. The combination of attention processes is very effective in enhancing the relevance of the features, and it makes the framework appropriate to the in-vehicle applications in real-time, which is helpful to the intelligent driver-monitoring and ensure the enhancement of road safety.