Analysis Of Driving Behavior For Estimating Drivers Cognitive Performance

In this era of rapid development,automobiles have become an indispensable means of transportation,and traffic accidents caused by improper driving behaviors are also increasing.Therefore,it is important to analyze driving behavior and discuss the impact of driving behavior on driver’s cognitive behavior,thereby reducing the occurrence of accidents.At present,the definition of driving behavior is not very accurate,and the methods used are different,so the analysis may not be comprehensive enough.In this study,we have collected novel data corpus including driving data and cognitive score of drivers for analyzing driving behavior.We have collected 21 types of sensor data from driving behaviors(for example:steering_angle,speed,accelleration position,room temperature and so on)and collected cognitive behavior data for each driver.There are 93 types of cognitive performance(for example:Reaction time_emergency reaction,Reaction Unevenness(max-min)_Emergency Reaction,Road environment understanding_WSQ,Accurate operation_DSQ and so on)for each person.The purpose of this thesis is to analyze the relationship between driving behaviors and cognitive performance of driver.In this study,several different methods are used to predict the driver’s cognitive performance behavior.First,regression modeling is conducted.Regression modeling is a technique to develop a prediction models that enable us to analyze the relationship between the dependent variable(target)and the independent variable(predictor).This technique is commonly used in predictive analysis,time series models,and finding causal relationships between variables.On the other hand,Long Short Term Memory(LSTM)neural network is used for time-series prediction tasks,and LSTM can capture the time-series dependency of sequence data such as time-series data observed from driving data.In evaluation of regression modeling and LSTM,the cross-validation method is used to validate the prediction accuracy of cognitive ability,and the data is divided into a training set,a verification set,and a test set.In this way,it can improve the reliability of the experiment,increase the accuracy,and reduce overfitting Get more information in limited data.This thesis mainly proposes an analysis of driver’s driving behavior and estimating the driver’s cognitive performance.We used 21 groups of driving behavior data to predict 93 sets of cognitive behaviors with lasso regression and ridge regression.At the same time,21 sets of driving behavior data predict 6 groups of typical cognitive behaviors with LSTM machine learning algorithms.A cognitive performance behavior under the same conditions,we can get that the r2 of the regression method is 0.726;the result is 0.678 in LSTM model.The results show that the regression method is more accurate in predicting cognitive behavior.