Comparative Study of Yawn Classification on CNN Architectures

This study explores the efficacy of convolutional neural networks (CNNs) in the context of yawning detection, employing five distinct architectures: DenseNet201, AlexNet, MobileNetv2, ResNet50, and VGG16. Leveraging the yaw-DD dataset, the models undergo rigorous training and testing, revealing compelling insights into their performance. The results underscore the competence of all five CNN models in effectively classifying yawning, with DenseNet201 and ResNet50 emerging as top performers, showcasing superior accuracy in both training and testing phases. Notably, the models demonstrate minimal overfitting and underfitting, highlighting their robustness in generalizing to new data. While acknowledging limitations, such as the use of public datasets and reliance on augmentation, the study lays the groundwork for future research. Recommendations include refining preprocessing techniques, diversifying datasets, and exploring architectural modifications to further enhance model performance. Transfer learning and fine-Tuning with larger datasets are suggested avenues for improved generalization. In conclusion, this research advances our understanding of CNNs for yawning detection, offering valuable insights and setting the stage for future investigations. The findings affirm the potential of deep learning in complex classification tasks, with implications for enhancing driver fatigue detection and bolstering road safety measures. © 2024 IEEE.