Driver drowsiness detection in low light

Driver drowsiness; Many effects such as micro-sleep while driving, delay in reaction times, distraction, leaving the lane, missing traffic warning signs are seen. Most of the time, traffic accidents occur as a result of these behaviors and cause many loss of life and property. In this case, one of the most important elements to prevent accidents is to detect the drowsiness of the driver early and warn the driver. The data set used is as important as the method used in order to determine the driver's drowsiness in the most accurate way. In this study, firstly, Sivas University of Science and Technology Driver Drowsiness Dataset (SUST-DDD), which includes real driving moments, was collected in order to detect driver drowsiness. The dataset consists of a total of 2074 videos, each 10 seconds long, belonging to two classes, drowsy and 'not drowsy'. Secondly, using the created dataset, the sleep state of the drivers was estimated with deep learning methods such as AlexNet, LSTM, VGG16, VGG19, VGGFaceNet and hybrid deep networks. As a result of the study, the created dataset and the applied hybrid deep network model estimate drowsiness with 98,42% precision, 98,76% recall, 98,47% f1 score and 98,48% accuracy values.