Event-Based Driving Style Analysis

Abstract

Driving styles have been analyzed for different purposes, such as building intelligent road systems, ensuring fuel economy, improving road congestion, vehicle automation, and road safety. Nevertheless, driving analysis is an exciting and evolving topic to study in terms of enhancing Driver Assistance Systems (DAS) and mitigating vehicle accidents. Therefore, the primary goal of this paper aims to develop a model that can classify the drivers based on their driving styles given the behavioural events that the drivers make while driving. Hence, we used an open source UAH-DriveSet dataset; the data were logged by many drivers who performed several trips on motorways and secondary roads with different driving styles (aggressive, normal, drowsy) in a real experiment. Furthermore, we propose the extraction of statistical and behavioural features that are used to represent different driving trips based on the driving events which occur during these trips. Then, we used these features to create a supervised learning model, i.e. using Support Vector Machines (SVM), to classify the trips according to the driver’s style. Moreover, the developed model can better classify the styles for different drivers, especially after selecting the best features that clearly distinguish between the different styles. One further application of the proposed features and methodology is to classify the road type according to the driver’s style.