Driver Drowsiness Detection Using Convolutional Neural Networks: A High-Accuracy Approach for Real-Time Monitoring

Driver drowsiness detection is one of the most critical tasks in improving road safety and avoiding accidents. This paper presents a deep learning model based on the Convolutional Neural Network to detect driver drowsiness. The model proposed here uses a very rich dataset available on Kaggle that encompasses real-time video data for drivers operating a car. This work will analyze the facial features of drivers, such as the pattern of eye closure and yawning that depicts a driver as drowsy or alert. The preprocessing was conducted with the intention of image augmentation, normalization for increasing the performance of the model, avoiding overfitting. The CNN architecture has been devised to capture the features in the space efficiently by using multiple convolutional layers followed by max-pooling and dense layers for classification. Hyperparameter tuning was also done by changing the learning rate, batch size, and dropout rates to come up with a promising model. The developed model was trained and validated using an 80/20 data split, which ∼ 0.0515, along with a validation accuracy of 97.92%, showing very strong generalization capability. This paper presents work in which CNN can perform drowsiness detection with high precision to develop real-time driver monitoring systems. This model is further integrable into ADAS, which could warn the driver long before any dangerous event may happen. Further research might include expanding the dataset and adding more physiological signals to enhance detecting accuracy and reduce false positives in complex real-world driving conditions.