Application of multi-fractal features of driving performance in driver fatigue detection

Driver fatigue caused by long-term driving is usually accompanied by a decline in driving performance. Therefore, driver fatigue threatens the safety of the drivers and passengers seriously. The development of driver fatigue detection technology can help remind the drivers and take relevant measures against driver fatigue, thereby improving traffic safety to a certain degree. The purpose of this paper is to improve the classification accuracy of the driver fatigue detection model. To achieve this goal, this paper introduces multi-fractal features of the driver behavior data. Six male subjects participated in the driving simulation experiment and UC-win/Road driving simulation software was used to collect driver behavior data such as driving speed, acceleration, steering wheel angle and steering wheel angular velocity. Indicators of the mean, the standard deviation and several different multi-fractal features of each kind of data were calculated. The changes of the drivers' subjective fatigue were also measured with a time interval of 600 s. And the relationship between the indicators and driver fatigue were measured. Additionally, the accuracies of driver fatigue detection models based on support vector machine (SVM) considering different features and different time window were compared. The research shows: The correlation between the singularity strength of acceleration (A0) and driver fatigue is significant; Compared with the correlation between other indicators and driver fatigue, that between A0 and driver fatigue is less affected by the time window width; The singularity strength of acceleration can help improve the accuracy of SVM; Compared with the time window of 15 s, a wider time window of 30 s can make the improvement more obvious. Therefore, it has certain application value in driver fatigue detection technology. © 2021, Jilin University Press. All right reserved.