Drowsiness detection of EEG signals using image-based convolutional neural network

Abstract

With the continuous development of autonomous driving technology, more drivers relax their attention while driving, resulting in a continuous accumulation of the number of vehicle accidents. The detection of driver's attention can be used as a condition for whether to execute autonomous driving, to help to reduce the probability of major accidents. This study compares the performance of image-based and traditional drowsiness detection trained on the same dataset. Total 5 participants were told to control a computer-simulated train with a 12-channel real-time EEG acquisition device on their heads. We transferred their EEG signals into images with the coordinates of every useful electrode as the training and testing data for our Convolutional Neural Network (CNN). Our results indicated that the CNN model trained on image-based EEG signals achieved much higher accuracy (83.49%) compared with SVM model trained on raw EEG signals, providing an effective method for improving driving safety.