Driver Dozy Discernment Using Neural Networks with SVM Variants

A driver’s lack of concentration or distraction is one of the main reasons for causing road accidents. Thus, increasing the driver’s awareness at the ideal moment will reduce the possibility of an accident of any kind. There were around 155 thousand accidents in India, and around 40 percent of accidents were caused by driver distraction, mainly due to driver drowsiness. Detecting drowsiness or fatigue prior to an accident will help reduce these accidents. There are several ways we may execute this. One of the easiest and most effective ways is through artificial intelligence and machine learning algorithms. We consider both physiological and behavioral categories, such as face movement and eye closure movements, to detect drowsiness. Further, training a particular model with different types of eye movements helps in detecting driver conditions. Driver drowsiness detection can be improved by continuously monitoring the driver via video, which helps in real-world applications, and by expanding the dataset through training, we get high accuracy and unrecognizable losses. Therefore, in this paper, we use the MRL dataset, which contains images from every angle and in every shade. To train the existing model with this dataset, we use image processing techniques and classification techniques to distinguish images of open and closed eyes on the basis of accuracy and loss function, a comparison of SVM (Support Vector Machine) and CNN (Convolutional Neural Network) models has been performed. As a result, CNN is considerably better than SVM and it is an effective technique for dozy detection.