EEG based Drowsiness Detection using Relative Band Power and Short-time Fourier Transform

Authors
Pranesh Krishnan1, Sazali Yaacob1, Annapoorni Pranesh Krishnan1, Mohamed Rizon2, *, Chun Kit Ang2
1Intelligent Automotive Systems Research Cluster, Electrical Electronic and Automation Section, Malaysian Spanish Institute Universiti Kuala Lumpur, Kulim, Kedah 09000, Malaysia
2Faculty of Engineering, Technology and Build Environment, UCSI University, No 1, Jalan Menara Gading, Taman Connaught, Kuala Lumpur 56000, Malaysia
*Corresponding author. Email: [email protected]
Corresponding Author
Mohamed Rizon
Received 6 November 2019, Accepted 16 April 2020, Available Online 11 September 2020.
DOI
https://doi.org/10.2991/jrnal.k.200909.001
Keywords
Drowsiness; polysomnography; band power; short-time Fourier transform
Abstract
Sleeping on the wheels due to drowsiness is one of the major causes of death tolls all over the world. The objective of this research article is to classify drowsiness with alertness based on the Electroencephalogram (EEG) signals using spectral and band power features. A publicly available ULg DROZY database used in this research. Algorithms are developed to extract the five EEG channels from the raw multimodal signal. By using a higher-order Butterworth low pass filter, the high-frequency components above 50 Hz are removed. Another bandpass filter bank separates the raw signals into eight sub-bands, namely delta, theta, low alpha, high alpha, low beta, mid beta, high beta and gamma. During pre-processing step, the signals are segmented into an equal number of frames. An overlap of 50% and a frame duration of 2 s using a rectangular time windowing approach segments the signal into frames. Then, the feature extraction algorithm extracts the relative band power features based on the short-time Fourier transform for each frame. The extracted feature sets are further normalized and labelled as drowsy and alert and then combined to form the final dataset. K-fold cross-validation method is used. The dataset is trained using K-Nearest Neighbor algorithm (KNN) and support vector machine classifiers, and the results are compared. The KNN classifier produces 96.1% (dataset 1) and 95.5% (dataset 2) classification accuracy.
Copyright
© 2020 The Authors. Published by ALife Robotics Corp. Ltd.
Open Access
This is an open access article distributed under the CC BY-NC 4.0 license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)