Research On Driving Fatigue Detection Based On Features Of Human Multi-physiological Signals

With the continuous improvement of China’s economy,the national living standards have been greatly improved.The original way of travel has gradually been replaced by driving.Therefore,car ownership in China is showing a trend of increasing year by year.The increase in the number of automobiles has also brought many problems to road traffic.Studies shown that tens of thousands of traffic accidents occurred every year in China,in which driving fatigue was one of the main causes of many traffic accidents.Consequently,effective detection of driving fatigue is of great significance for improving road traffic safety.In this paper,the multiphysiological signals characteristics of human body were used to comprehensively detect driving fatigue.The main research contents are as follows:（1）Feature extracting of Electroencephalogram（EEG）signal.EEG signals are widely used to detect driving fatigue,which can directly reflect the changes of driver’s brain activity.In this paper,EEG features were used to detect driver fatigue state.Firstly,the EEG signals were decomposed adaptively into multiple Intrinsic Mode Functions（IMF）components using Variational Mode Decomposition（VMD）algorithm.Secondly,the most optimal IMF component was selected according to the different fatigue components in each component.Then,the least square method was used to determine the optimal Scale factor for fatigue feature extraction.Based on the determined Scale factor,the Modified Multi-Scale Entropy（MMSE）of the selected optimal IMF component was calculated.Finally,the extracted MMSE features were used to detect driver fatigue changes in real-time.（2）Features extracting of electrocardiograph（ECG）signal and electrooculogram（EOG）signal.Since vast amount of information related to the human mental state was embedded in the human heart rate.Therefore,the ECG features were extracted to detect driver’s fatigue state in this paper.Firstly,the wavelet noise reduction method was used to denoise the ECG signal.Then,the Pan＆Tompkins algorithm was used to detect the R wave position of the ECG signal,calculated the R-R interval of the ECG signal.Finally,the approximate entropy（Ap En）of the R-R interval of the ECG signal was calculated by the Ap En algorithm.The driver’s mental state was detected by the Ap En feature of the R-R interval.In addition,the signal waveform changes significantly when driver perform the blinking action.Therefore,the back propagation（BP）neural network was used to extract the blink times,judging the driver’s fatigue state according to the change of blink frequency.（3）Driving fatigue features fusion.Due to the fatigue information contained in single feature is insufficient,the accuracy and reliability of driving fatigue detection is relatively low.Fusion features can effectively improve the accuracy and reliability of driving fatigue detection through the mutual verification and supplementation between different features.Therefore,on the basis of the single feature extracted,the fusion features containing EEG,ECG and EOG signals was formed by using factor analysis in this paper.Finally,the fusion feature was used to detect the driver’s fatigue state.In addition,the dispersion of EEG signal features,ECG signal features,EOG signal features and fusion features were calculated separately in this paper,and compared of each feature dispersion.The results showed that the dispersion of fusion features was smaller than single features,indicating that the data distribution of fusion features was more concentrated and less influenced by individual differences.