ANN based Joint Time and frequency analysis of EEG for detection of driver drowsiness

Abstract

Drowsiness detection plays a vital role in accidents avoidance systems, thereby saving many precious lives. Many attempts were made to detect the drowsiness by the physiological features such as Electroencephalogram (EEG), Electrooculogram (EOG), and Heart Rate Variability, etc., but a reliable index to determine the drowsiness is not yet a reality. This study contributes in identifying the drowsiness levels by an index called Drowsiness Index (DI) from the EEG signal analysis of the drivers. In this report, the EEG signal is processed to detect the behavioural patterns of the brain and drowsiness state of the drivers while performing monotonous driving for long distances. An eight-channel EEG data acquisition system is used to acquire the EEG data from thirteen male volunteers. The EEG signal is pre-processed and decomposed into various rhythms by applying a digital filter in MATLAB 2007b. Time-Frequency domain analysis has been done to extract certain features namely power within spectrogram, power within the root mean square deviation which are statistically significant (ρ < 0.05) in the detection of drowsiness. The driving profile is classified into active and drowsy by a separable marker with a range of 0.4-0.6, and linear regression analysis has been performed on the features extracted. A Drowsiness index is proposed stating a positive correlation (0.8-0.9) between the Total mean and the drowsy mean of the subject. The final features extracted from the data are classified using an ANN-based classifier system and has achieved a sensitivity of 99.82 per cent and specificity of 99.78 per cent.

Analysis of Driver Drowsiness Detection using EEG and EOG

“…The β activity corresponds to the frequency band [12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30] Hz. It is present on both sides of the brain symmetrically.…”

“…The main difference between all approaches using neural networks lies in the nature of the inputs used as well as the type of network used. Authors of [25]suggested applying a PCA to a three-channel EEG spectrum to drive a multilayer neural network on the main component spectrum.Authors of [26]estimated the EEG spectrum by both STFT and WT to drive their network. Their technique has not been tested on real signals.Authors of [27] used a single parieto-occipital EEG channel (P4-O2) to educate two different networks.…”