Wireless Wearable Fatigue Driving Early-Warning System Based On 3D CNN

In recent years,fatigue driving has been a major cause of traffic accidents.At present,experts have proposed a number of physiological indicators for assessing driver's fatigue levels.Electroencephalogram(EEG)has proven to be one of the most reliable indicators for detecting fatigue levels.This paper proposes a classification model based on Three-Dimensional Convolutional Neural Network(3D CNN)to predict three fatigue levels,and designs a wireless wearable fatigue driving warning system for to detect the fatigue state of the driver in real time.The system consists of three parts:(1)wireless wearable EEG acquisition device;(2)EEG analysis system;(3)warning device.The first part conveniently collects the driver's EEG signal through the wireless wearable EEG acquisition device and sends it to the EEG analysis system through the wireless transmission device.In the second part,the EEG signals are preprocessed by a combination of linear filter,wavelet transform and independent component analysis.Then,the Power Spectral Density(PSD)is calculated by the Welch method and the EEG signal is converted into a feature matrix.Finally,the feature matrix combined with the lead spatial information is extended into 3D feature data,and as an input of the predictive model,and three fatigue levels are obtained by the 3D CNN algorithm.The third part is the warning device,which consists of intelligent bracelet and in-vehicle monitoring system.The wireless transmission device receives the predicted result through the wireless transmission device and reflects the driver's fatigue level in real time to remind the driver to improve the driving state.Finally,the paper conducted a performance test on the fatigue driving warning system.Ten healthy subjects performed a 90-minute driving task in a virtual driving environment,in which the first 60 minutes of EEG data was used to train the model and the remaining 30 minutes of EEG data was used to test the model.The results show that the fatigue level predicted by the classification algorithm is basically consistent with the actual fatigue level quantified by the response time,which proves the feasibility of our proposed fatigue detection system.