Correlation analysis between the EEG parameters and the parameters derived from ECG and Steering wheel related signals for driver drowsiness detection

Abstract

Physiological signals such as Electroencephalography (EEG), Electrocardiography (ECG) and nonphysiological signals such as steering wheel related parameters have been investigated for drowsiness detection in previous researches. EEG has been deemed as a reliable way to detect drowsiness; while the accuracy of using ECG or steering wheel related parameters for drowsiness detection is not as high as that of EEG’s but they have the advantages of low cost and non-intrusive. This work is devoted to find out the possibility to enhance the accuracy of drowsiness detection based on the ECG and steering wheel related parameters. The correlation between EEG and ECG parameters and the correlation between EEG and steering wheel parameters for drowsiness detection are analyzed. If strong correlation between them are found it is possible to use the ECG and steering wheel related parameters to represent the EEG parameters which means the accuracy of drowsiness detection based on ECG and steering wheel parameters can be improved. Several parameters were chosen for the correlation analysis, the EEG parameters are the alpha, beta and theta band power, the ECG parameters are heart rate, Heart Rate Variability (HRV) parameters and a parameter derived from Detrended Fluctuation Analysis (DFA), and the steering wheel related parameters are four variables derived from steering wheel movement, all of these parameters are most commonly used parameters for driver drowsiness detection in the literature. The results of the analysis showed that neither the ECG parameters nor the steering wheel related parameters have strong correlation with EEG parameters. Parameters composed of combination of ECG signal parameters and steering wheel related parameters also did not show strong linear correlation with EEG parameters, however close nonlinear relationships have been found by artificial neural network methods, the promising results have largely increased the possibilities to build driver drowsiness detection system inexpensively and intrusively.