EEG Based Feature Extraction and Classification for Driver Status Detection

摘要

Driver status determination is one of the important features present in today's automotive. EEG analysis based driver status indication is one of the effective ways to measure driver status. This paper discusses about EEG analysis using wavelet transforms for separating EEG signal frequencies and extracting time and frequency domain features for further classification. EEG is a non-stationary signal and analysis only in time or frequency domain is not preferred. Wavelet transforms analyze the signals in both time and frequency domain. EEG rhythms consisting of different frequency bands are separated using Daubitius DB8 wavelet transform. Sleep data sets from Physionet were used for the proposed study. The drowsy status is indicated with alpha and theta frequency rhythms. The features representing alpha and theta activity were extracted and can be used to classify the driver status. The statistical features in time and frequency domain are used classify alert and drowsy state of the driver. Variants of SVM models were used to classify the signals and cubic SVM is found to give highest classification accuracy of 93.9%. The proposed method can be used to analyze driver status and further to analyze different sleep stages.