Detection of driver drowsiness using transfer learning techniques

Major traffic accidents are often caused by driver drowsiness. Modern lifestyles reduce the amount of sleep an individual gets. Hence, this paper proposes to use Deep Learning to detect such scenarios and prevent mishaps. In the proposed experiment, 7 different deep learning models based on transfer learning are trained and tested for the driver drowsiness problem. The models include VGG19, ResNet50V2, MobileNetV2, Xception, InceptionV3, DenseNet169, and InceptionResNetV2. All of the experiments are conducted on the NTHU-DDD2 dataset. Evaluation and comparison of each model's performance are conducted. We identify the best-performing model and compare it with previous literature in the same field. The best-performing model came out to be VGG19. It achieved an accuracy of 96.51%, precision of 98.14%, recall of 95.36%, and f1 score of 96.73%.