A Study On EEG Based On-line Fatigue Monitoring Algorithms

Because trafc accidents caused by fatigue driving occur frequently in recen-t years, fatigue monitoring has become an important research topic. In the past re-searches often use the facial video signal, blood pressure, body temperature or otherphysiologicalsignals. Comparedtothesesignals,theelectroencephalogram(EEG)canrefect the brain’s activities more directly and objectively, has a higher temporal reso-lution, and can not be artifcially controlled and faked, therefore, we use EEG signalto study fatigue monitoring in this article. In the frst half of the article, we mainlyintroduce the common EEG processing processes, and in the second half we introducethe methods used in our research. Firstly, subjects are asked to complete task whichwill induce subject’s fatigue, and at the same time we record subject’s EEG signal andperformance. Then we use fast Fourier transform (FFT) to obtain the power spectraldensity (PSD) features of the original EEG signal in the respective frequency bands.In order to remove the fatigue-unrelated noise, we use linear dynamic system (LDS)to smooth features. Then we use principal component analysis (PCA) to reduce fea-tures’ dimension and discard those features which have bad correlations with fatiguelabels. Finally, we extend the remaining features to dynamic feature groups, use par-allel hidden Markov model (PHMM) and fuzzy integral to train and fuse classifers.Experimental results indicate that the accuracy of classifcation obtained by using ournew method are88.85%for classifying3states and83.09%for classifying4states,respectively.