Research On Driving Fatigue Identification Method Considering Individual Differences Of Electrocardiogram

Driving fatigue is one of the main causes of road traffic accidents.Compared with other causes of traffic accidents,driving fatigue which is more likely to lead to death or serious injury has the characteristics of diverse individual representations and strong concealment of fatigue.Due to individual differences among drivers,the common features of driver fatigue are not clear,and the existing fatigue identification algorithms cannot meet the high prospective requirements of fatigue identification.Therefore,how to correctly deal with the individual differences of driver fatigue characteristics has become significant in the research of driver fatigue.Against such a backdrop,this paper explores the driver fatigue identification method based on the electrical signal of the driver’s heart,aiming to establish a driver fatigue identification algorithm with high performance as well as high generalization ability and solve the problem of individual differences in fatigue characteristics in the process of driver fatigue identification,in order to improve the accuracy and generalization ability of the driver fatigue identification method.Firstly,this paper takes driving fatigue as the research object and designs a driving fatigue induced test in a highway scenario,relying on the UC-win/Road driving simulation test platform.In the test,12 subjected drivers were asked to perform driving tasks in a monotonous driving environment for a long time.The data such as ECG signals and fatigue subjective scores during the transition from wakefulness to fatigue state were obtained by using BIOPAC polysomnography recorder and Karolinska Sleepiness Scale（KSS）subjective evaluation method.Secondly,by analyzing the research basis of driving ECG signal and driving fatigue,the collected raw ECG signal is continuously sliced and processed.The linear and nonlinear methods are used for in-depth feature mining to extract 17 typical feature indicators from frequency domain,time domain and nonlinear perspectives as candidate features for driving fatigue classification.And then the fatigue sample dataset including feature vectors and feature labels is obtained after feature normalization processing and KSS fatigue level reclassification.Thirdly,in order to improve the generalization ability of the driver fatigue classification identification model and reduce the negative impact of individual variability on the generalization model,this paper ranks the importance of fatigue features and creates a ranked list of the importance of candidate features in the fatigue sample for each subject driver,using the Support Vector Machine（SVM）discriminant function information as the evaluation criteria and the recursive feature elimination algorithm as the search strategy.In order to avoid evaluation bias,the correlation bias reduction algorithm was introduced.By constructing the driver fatigue feature importance evaluation matrix K and the feature importance index V_a,the optimal subset of features is extracted which can significantly characterize the drivers’fatigue state,including 8 common key features that can be used to train a generalized driver fatigue identification model.Finally,a binary classification evaluation model for the driver fatigue state was established by using the feature vectors of the 8 optimal features as input vectors and combining the SVM algorithm.Meanwhile,the penalty parameter C and the kernel parameterγof the support vector machine were optimized by choosing the kernel function that provides the highest classifier performance and using the grid search method and the cross-validation method.After testing,the average classification accuracy of the model is 87.04%,which is 9.20%higher than the accuracy of 79.71%of the model built in this paper without considering individual driver variability.The fatigue identification model considering individual differences of ECG in this paper can significantly improve the identification effect and demonstrate the importance of taking individual differences into consideration when building fatigue monitoring models.The findings of the study lays the foundation for improving the reliability of fatigue driving monitoring methods in the future and provides a reference for improving the effectiveness of fatigue driving identification methods.