Estimating vigilance in driving simulation using probabilistic PCA.

In avoiding fatal consequences in accidents behind steering wheel caused by low level vigilance, EEG has shown bright prospects. In this paper, we propose a novel method for discriminating two different vigilance states of the subjects, namely wake state and sleep state, during driving a car in a simulation environment. After filtering the EEG data into a specific frequency band, we use probabilistic principle component analysis (PPCA) to reduce the data dimension. Then we model each vigilance state as a lower dimension Gaussian random variable by applying PPCA again. The feature related to class posterior probability is calculated for classification. The experimental results show satisfying time resolution (< or = 5s) and high accuracy (> or = 96%) across five subjects on both common frequency bands beta (19-26 Hz) and gamma (38-42 Hz), and a broad band (8-30 Hz).