Drowsiness Detection In Drivers Using Transform Learning

Drowsiness detection in drivers is a critical safety concern in the transportation industry, as it can lead to accidents and demise. In this study, we propose to use transfer learning to detect drowsiness in drivers. The proposed method uses a deep convolutional neural network (CNN) model that is pre-trained on a bundle of image datasets (52620) and fits it on a smaller driver image dataset. One of the main challenges in this area is the lack of large-scale data, the data is needed to train the model accurately. To overcome this limitation of data-set, transformation learning method is used, which is a technique that allows a model to be trained on first task and then switch to another related task. The proposed transform learning method is used to fine-tune a pre-trained CNN model on a smaller dataset of driver's images, which overcomes the limitation of limited data availability. The fine-tuned model is able to classify driver's images into drowsy and non-drowsy categories with higher accuracy (training accuracy: 94.18% ; validation accuracy: 90.56%; test accuracy: 94.55%). The proposed method can be easily integrated into existing transportation systems, such as vehicles, to detect drowsiness in real-time and have potential to help millions of lives worldwide.