Research On Driver's Fatigue Lane Departure Recognition Method

Through collecting and sorting out the statistics of road traffic collision for recent years,we find that fatal accidents were usually caused by fatigue driving.This dissertation has proposed a Fatigue Lane Departure(FLD)recognition method,which combines driver's operating behavior and vehicle motion state.Main work is as follow:An experiment about FLD has been taken through the co-simulation platform based on the Car Sim and Lab VIEW software.In the experiment,we select twelve drivers and relevant parameters which can reveal driver's characteristics.We propose a method to optimize the intent time window by ROC system which analyzed at various times preceding the lane change maneuver.Independent-samples T test were used to quantify the effects of FLD and NLC on characteristic parameters of driving behavior.Finally,the optimal driving behavior characteristic parameters sample set which could distinguish the FLD and NLC state.Via GM-HHM,set up the FLD recognition model,base on the driving behaviors can gauge vehicle control situation.To prove the superior effect of FLD recognition model based on GM-HMM as built above in comparison with other typical model establishment,we concurrently evaluate the influence of quantity,type of characteristic parameters toward the effect of the model.Step1: We build two recognition models: recognize the FLD based on Support Vector Machine(SVM)and driver's fatigue based on Time Series Analysis of steering wheel angle velocity(TSA).Based on theoretical basis of SVM to establish FLD recognition model and the use of training sample set must be similar to those in GM-HMM model.We put the characteristic parameter groups and individual characteristic parameter groups into the model to carry out to train and get the parameters of corresponding recognition model.The driver's operating behavior under the state of fatigue is analyzed,followed by the determination of the temporal recognition window;and then,the data series of the steering wheel angle velocity in the temporal recognition window is selected as the recognition feature.If the detection feature satisfies the extent constraint and the variable constraint in the temporal window,the state of fatigue is recognized accordingly.Step2: To evaluate the effectiveness of GM-HMM model,we compare it to SVM model and TSA model and by itself with different characteristic parameters.Firstly,we figure out three performance parameters of corresponding recognition model.After that compare these values with each other,which model has bigger feature parameter value that model is more effective.The result shows that GM-HMM model with characteristic parameter groups of driving behavior has feature parameter values are highest,so it is the most optimal model.This proves that,the research results provide theoretical and technical support to the related field of active safety assistance systems.