Deep Convolution Neural Network-Based Feature Learning Model for EEG Based Driver Alert/Drowsy State Detection

Driver state detection is an important feature of Advance Driver Assistance Systems (ADAS) of automotive. Accurate determination of the driver’s alert/drowsy condition avoids accidents and offers safety to both driver and vehicle. The electroencephalogram (EEG) based method of determining driver’s alert/drowsy condition is the proven most accurate direct measure of driver state. Researchers have attempted to extract significant features representing the driver’s state for a long time. However, extraction and selection of features from the large number of them is very difficult for EEG based systems. In this paper, a representation learning model using a deep convolution neural network (DCNN) is proposed that can automatically learn features from labeled data. The model was used to extract and learn features for publicly available EEG data sets and experimented for different classification results. The results showed that features extracted using DCNN based feature learning model proved better than conventional Principal Component Analysis (PCA) and Recursive Feature Elimination (RFE) techniques in terms of significant feature extraction, data dimension reduction, and classification accuracy. The features also can converge quickly and reduce training times for classifiers. The model can be very effectively applied to automotive application where speed is the criteria.