Research On Driver Behavior Analysis Based On Machine Learning

With the development of social progress and technology,the number of Chinese automobiles is increasing continuously,followed by the increasingly serious traffic safety problem,caused huge human casualties and economic losses,therefore,the prevention of traffic accident has the great research significance.While it is difficult to make a breakthrough in the prevention of traffic accidents by improving laws and regulations and improving vehicle performance,the analysis of driver behavior has attracted more and more attention.At present,there is still a lack of quantitative analysis of driver behavior,and the selection of evaluation index of driver behavior is not unified.Therefore,the machine learn-based analysis of driver behavior proposed in this paper has important research significance.Considering the characteristics of different driving styles of drivers,this paper innovatively proposed a driving style classification method based on lane change scenarios and acceleration scenarios,and quantified the behavior indicators of lane change scenarios and acceleration scenarios.Driving scenes can better reflect the characteristics of different drivers,and drivers’ driving style orientation can be intuitively seen.Based on the characteristics that drivers drive vehicles for a long time and their driving habits tend to be stable,drivers’ driving styles are divided into aggressive drivers and robust drivers by using the collected driver data.The main research work of this paper is as follows:(1)Data collection.This paper builds a data acquisition tool chain based on Vi CANdo(Intelligent Driving Data Acquisition and data analysis tool)and ROS(Robot Operating System),intercepts data signals such as distance between the vehicle and surrounding vehicles,running speed,running lateral acceleration and running longitudinal acceleration.(2)Through the extraction study of driver behavior indicators,the algorithm for identifying acceleration scene and lane change scene is designed to extract the influence indicators under the scene respectively.In the lane change scenario,drivers’ driving habits are judged from eight dimensions,including average lane change speed,average lateral acceleration,average longitudinal acceleration,longitudinal acceleration variance,lateral acceleration variance,speed variance,the distance between the lane and the nearest vehicle in front,and the distance between the lane and the nearest vehicle behind the lane change side.In the acceleration scenario,the driver’s behavior is studied quantitatively from five dimensions,namely the initial speed of acceleration,the end speed of acceleration,the average acceleration,the variance and the average speed.(3)The extracted index stability is judged by observation method and time series method respectively,indicating the characteristics of driving behavior habits tending to be stable.(4)KNN(K-nearest neighbor)algorithm,decision tree algorithm and support vector machine algorithm are applied to classify driving behavior index,analyze three algorithms from the evaluation index,and explain the suitable algorithms for this research.Finally,the driver recognition experiments were carried out with or without considering the distance parameter,and the driver recognition results were compared.The results show that considering the distance parameter is better.Through the study of driver style classification in this paper,it can be seen from the research results that the driver behavior indicators are distinguishable from the acceleration scene and lane change scene.The results show that by scenarioizing the driver behavior,the recognition results of the SVM algorithm can reach about 78% in both scenarios,and the effect is good,indicating that it can be used to distinguish the robust driver from the aggressive driver.This study provides theoretical support and new ideas for analyzing driver behavior.