Driving Style Recognition Considering Different Driving Scenes

Abstract

Correctly identifying the surrounding driving scenes and drivers' driving styles plays an essential role in improving driving safety. This paper presents a driving style recognition algorithm considering the differences in driving scenes. The driving data of 38 randomly recruited drivers are collected by a T-box device installed on a truck. The collected data is clustered into five categories according to different velocity and acceleration values, representing five different driving scenes. Some characteristics that can reflect driving styles are calculated in each scene, and they are as the input of SOM+KMeans and the GMM model for driving style recognition. And the classification results of the two methods are compared in each driving scene, which has an average 91.7% recognition consistency rate. Our driving style identification model combines the characteristics of different driving scenes and can reflect the impact of road scenes on driving style.