Detection of Driver Drowsiness Based on Eye and Mouth Movements Using Convolutional Neural Networks

Abstract

The Increasing of road mobility triggers the increasing of the number of traffic accidents. One of the main factors of the accidents is human errors which are heavily influenced by the driver conditions. Fatigue, drowsiness, and loss of concentration are among the common driver conditions that could cause traffic accident in addition to high-speed driving behavior. This could be minimized if early warning systems of driver conditions existed. This research aims to develop an early detection system for driver conditions using Convolutional Neural Network (CNN) method. Here, we investigate the effect of the depth of CNN and other hyper-parameters and observe their performance. We used eye movements and mouth conditions to be an indicator driver conditions. We evaluate the method using public dataset that contains image data of drivers on the highway in a state of yawning, not yawning, eyes open, and eyes closed. The experiment showed the best parameters with a learning rate of 0.001 and an epoch of 100. The resulting accuracy reached 99.31%.