In order to avoid erroneous braking responses when vehicle drivers are faced with a stressful setting, a K-order propagation number algorithm-Feature selection-Classification System (KFCS) is developed in this paper to detect emergency braking intentions in simulated driving scenarios using electroencephalography (EEG) signals. Two approaches are employed in KFCS to extract EEG features and to improve classification performance: the K-Order Propagation Number Algorithm is the former, calculating the node importance from the perspective of brain networks as a novel approach; the latter uses a set of feature extraction algorithms to adjust the thresholds. Working with the data collected from seven subjects, the highest classification accuracy of a single trial can reach over 90%, with an overall accuracy of 83%. Furthermore, this paper attempts to investigate the mechanisms of brain activeness under two scenarios by using a topography technique at the sensor-data level. The results suggest that the active regions at two states is different, which leaves further exploration for future investigations.
Keywords: K-order structure entropy; brain network; brain-computer interface technology (BCI); braking intention detect; electroencephalogram (EEG); pattern recognition.