EEG-based emotion recognition for road accidents in a simulated driving environment

Encountering unexpected events with different levels of danger can cause different levels of emotional changes in the driver, and identifying the driver's mental state can assist in determining whether he or she can control the vehicle. Electroencephalogram (EEG) is an essential component of the basis for emotional classification. Current driving-related EEG research has focused chiefly on fatigue driving and road rage. In order to explore the connection between panic emotion and accident-avoidance ability, this paper proposes an EEG acquisition and emotion classification scheme in a simulated driving environment. The scheme uses vehicle speed as a variable to simulate obstacle avoidance at different danger levels. It uses graph neural networks (GNN) with functional connectivity and attention mechanisms to simulate the physiological structure of the brain to process the data. In addition, various experiments were conducted to compare the features from entropy and power perspectives. The three-class classification result reached 75.26%, with a single label's highest F1 score of 76.7%. The binary classification result reached 91.5%, with a single label's highest F1 score of 91.86%. The experimental results show that the solution can effectively simulate different dangerous situations, capture the driver's EEG signals, and effectively monitor the emotional state in combination with deep learning models.