Research On Non-contact Fatigued Driving Detection Method Based On Kernel Extreme Learning Machine

Fatigue driving has always been the main cause of car accidents.When the driver becomes fatigued,he will doze off involuntarily,which will eventually lead to the occurrence of a car accident and cause the loss of life and property.Therefore,research on a device that can quickly and accurately detect the driver’s fatigue level and accurately warn the driver of dangerous driving has very important practical significance for reducing the incidence of traffic accidents.Various detection methods have limitations such as low detection accuracy,high invasiveness,and poor stability.The fatigue detection method based on multi-source information fusion can complement the shortcomings of multiple fatigue detection technologies,which has gradually attracted the attention and research of scholars.This thesis designs a non-contact fatigue driving detection method based on multi-source information fusion of driver’s physiological parameters and behavior characteristics,which effectively enhances the accuracy and stability of fatigue recognition.This article first uses Doppler radar to detect the physiological parameters of the driver,and establishes the Physiological Parameter Data Set(PPDS)after filter separation and feature extraction.In order to verify the superiority of multi-source information fusion and avoid the shortcomings of poor anti-interference of a single signal,a highdefinition camera is used to locate and capture facial information,and formulas for eye aspect ratio,mouth aspect ratio and head Euler angle are proposed to extract the driver’s face Feature points,establish a Multi-source Fusion Data Set(MFDS)of physiological parameters and behavioral characteristics.Finally,designs optimization algorithms such as Extreme Learning Machine(ELM),Kernel ELM(KELM)and Wavelet Kernel ELM(WKELM)to train PPDS and MFDS data sets respectively Classification to obtain the optimal classification model to determine the driver’s physical fatigue level.The experimental results verify that the algorithm model based on multi-source information fusion data set can detect body fatigue state.It is increased by about 2% compared with the single physiological parameter data set,which reflects the superiority of the multi-source information fusion.In addition,in order to verify the advantages of WKELM is high recognition rate and strong stability in fatigue state detection,the traditional classification algorithm SVM(Support Vector Machine)and the BP Neural Network optimized based on Genetic Algorithm are analyzed and compared.The classification training on the MFDS finally proves that the WKELM algorithm has fast training speed,and the fatigue state recognition rate is up to 95.83%,which can better distinguish the driver’s fatigue level.