Abstract
The proposed system measures the attentiveness of the driver using the developed wearable device with electrodes placed on the temples of the driver. By using EEG signals, the accuracy of attentiveness detection may be increased compared to standalone camera-based systems. Preprocessing was used to remove noise from the measured signals. Features were extracted from the signals, and an optimal feature set was determined by analyzing the feature's effectiveness. A Support Vector based machine-learning algorithm was used to classify the driver's states. To gather data for each state, experiments were conducted on an indoor driving simulator with four subjects. The proposed model's classification accuracy was 84.4% for the five-fold cross-validation of the data. In a subject-independent test, the accuracy was 84.2% for the four states (city attentive, city distracted, highway attentive, and highway distracted). When adding a third classifier category to detect drowsiness from inattentiveness, the proposed model's accuracy was dropped to 72.6%. In this preliminary study, abnormal conditions of the drivers were detected and classified using the proposed system.