Research On Fatigue Driving Detection Technology Based On EOG Signals

In recent years,people’s pursuit of a fast and comfortable living environment has led to the rapid development of the automobile industry.The increase in the number of cars has led to frequent traffic accidents,which have caused huge casualties and property losses to people.The survey found that fatigue driving is one of the main reasons among many factors leading to traffic accidents.It is very meaningful to explore a fatigue driving detection method for protecting life safety and reducing property damage.The electrooculogram（EOG）can reflect the state of eye movement.When a person enters a fatigued state,the movement state of the eye will change,so fatigue driving can be detected by collecting the driver’s EOG signals.The specific work in this paper is as follows:（1）A method is proposed to detect fatigue driving by extracting the approximate entropy（Ap En）,permutation entropy（Pe En）and fuzzy entropy（Fuzzy En）features of EOG signals,and then using genetic optimization of generalized regression neural network based on genetic optimization（GA-GRNN）for fatigue driving detection.The mean,standard deviation,root mean square features and approximate entropy,permutation entropy,fuzzy entropy features of the EOG signals are extracted respectively,and then the decision tree（DT）,fuzzy neural network（FNN）and GA-GRNN are used for fatigue driving detection.The experimental results show that extracting the approximate entropy,permutation entropy and fuzzy entropy features of the EOG signals,and then using the GA-GRNN for fatigue driving detection has the best effect.（2）The Relief F algorithm is used to improve the classification accuracy of GA-GRNN’s fatigue driving detection and reduce the running time of GA-GRNN’s fatigue driving detection.The approximate entropy,permutation entropy and fuzzy entropy features of the EOG signals are extracted,and then GA-GRNN is used for fatigue driving detection without feature dimension reduction,multiple dimensional scaling（MDS）algorithm for feature dimension reduction,and Relief F algorithm for feature dimension reduction.The experimental results show that using Relief F algorithm to reduce the feature dimension of approximate entropy,permutation entropy and fuzzy entropy can improve the classification accuracy of GA-GRNN’s fatigue driving detection and reduce the running time of GA-GRNN’s fatigue driving detection.