A new deep learning model based on convolutional neural network and residual blocks for driver drowsiness detection

Abstract

Recognizing the pivotal importance of monitoring driver inattention in the quest to minimize accidents and enhance safety and security, it is essential to highlight the inherent danger posed by drowsiness-a specific form of inattention that can significantly contribute to accidents. To address this issue, several propositions involving artificial intelligence have been put forth to effectively monitor and identify instances of driver drowsiness. However, challenges persist in the form of real-time processing constraints, the intricate nature of model parameters, and the response time of the model. The proposed methodology focuses on using convolutional neural networks (CNNs) and Residual blocks as robust and effective deep-learning models for the real-time detection of driver drowsiness. The integration of CNNs and Residual blocks enhances the model's precision, striking a well-balanced synergy between computational efficiency and performance. Notably, this approach demonstrated an impressive accuracy of 96.09% along with a recall, f1-score, and precision all at 96% when evaluated on the publicly available eye_dataset. Furthermore, the runtime of the developed model is a mere 70 ms. To further validate the efficiency of our proposed model, we conducted a comparative analysis with various pre-trained residual neural networks, including ResNet152, ResNet50, ResNet50v2, ResNet101, and ResNet101v2.