Research On Driving Fatigue Detection Method Based On Information Fusion

Driving fatigue is a physiological phenomenon that often occurs during the driving process.If the driver enters a fatigue state,he will be distracted,unresponsive,and easily lead to traffic accidents.Driver fatigue detection methods based on a single source of information are less stable in a specific driving environment and have significant limitations.Current researchers tend to fuse multi-sources information to make comprehensive judgments on driving fatigue.This paper aims to improve the accuracy,stability and environmental adaptability of driving fatigue detection,and takes the driver's facial image and electroence-phalogram(EEG)as research objects.This article studies the following critical issues such as the multi-source fatigue features extraction,single-modality driving fatigue detection,and information fusion based driving fatigue detection.Finally,we conducted a simulated driving experiment to evaluate the effectiveness of various methods of driving fatigue detection.The main research content is as follows:(1)This paper investigated current research status and research trends of driver fatigue detection,and summarized the advantages and disadvantages of various research angles such as image information,physiological information,and driving behavior information.For the deficiency of single-mode driving fatigue detection,we determined a driver fatigue detection scheme that fuses driver's facial image information with brain electrical information to improve the detection accuracy,stability and environmental adaptability.(2)Then the paper extracted eye movement characteristics such as PERCLOS,degree of eyelid closure,and blink frequency from the driver's eye information.and trained a single-modality driving fatigue detection model based on eye movement characteristics.The fatigue/non-fatigue classification accuracy reached 86%.Then we extracted the frequency band energy ratio of each rhythm from the driver's EEG information,and trained the single-modality driving fatigue detection model based on EEG features.The fatigue/non-fatigue classification accuracy rate reached 78%.(3)For the limitation of the single-mode driving fatigue detection method,we use the MKL and MCCA to fuse multi-source heterogeneity features and train the driving fatigue detection model respectively.The accuracy rate reaches 89%and 93%.Based on the D-S evidence theory,the image information and EEG information are combined at the decision-making level to build a driving fatigue detection model with an accuracy of 95%.(4)Finally,the paper used a simulation driving experiment platform to conduct a simulation driving experiment based on sleep deprivation and reaction time,and synchronously collected driver's facial image information and brain electrical information.The data is then annotated to get the driver fatigue data set,and the data set is used to train and test various driving fatigue detection models.Then we designed a simulation driving experiment in an interference environment to evaluate the stability of various driving fatigue detection models.