Discrete Wavelet Transform based statistical features for the Drowsiness detection from EEG

Drowsiness is a major patron to road accidents. Detection of drowsiness while driving is a challenging objective in accidents avoidance systems. This study reports a new index to assess the drowsiness state of drivers using Joint Time-frequency analysis of Electroencephalography (EEG). Fifteen healthy male participants proffered in this study by performing a tedious driving task on a static simulator for 60 min. The subjects are deprived of sleep for at least 18 hrs and sleep music (Delta waves) played in the background, induces sleep during the task. Acquisition of EEG signals was implemented by eight channel Octal Bio-Amplifier (AD Instruments) at a sampling frequency of 1000Hz. The electrodes positioned at the four lobes of the brain namely Frontal, Temporal, Parietal, and Occipital lobes and further analysis has been carried out in MATLAB T 2007b (Math works, Inc., USA). The EEG signals are de-noised by Chebyshev filter (0.5-40 Hz) and subsequently decomposed into various rhythms of EEG such as Beta (CD5: 14-30 Hz), Alpha (CD6: 8-13 Hz), Theta (CD7: 4-7 Hz), and Delta (CA7: 0.5-3.5 Hz). Two parameters viz., Relative Wavelet Packet Energy (RWPE) and Power within the RMSD (PRMSD) are computed in this study to analyse the driving performance of the subjects against the subjective assessment inferred from the video recordings. The parameters RWPE, PRMSD within Beta and Alpha has reduced relatively in the Parietal lobe, occipital lobe, and temporal lobe as the subjects fall into the drowsy stage. This analysis is clinically correlated as the cortical activity reduces slowly during the onset of sleep. It is clearly evident that these features are significant (p< 0.05) in the detection of drowsiness with Confidence Interval of 14. This study also reports a significant correlation (p< 0.05) between PRMSD of Total mean with Active mean and Drowsy mean with R2 value of 0.82.