Driving Style Recognition of Interacting Vehicles in Multiple Scenarios

Abstract

For autonomous vehicles in interactive scenarios, one of the necessary prerequisites for making decisions is to identify the driving style of interactive vehicles accurately. One of the future trends in the field of driving style recognition is to select a set of appropriate driving features to represent driving style and use a machine learning algorithm to process the selected features. This paper subtly designed a set of driving features suitable for multiple scenarios and proposed a driving style recognition method based on K-means clustering and a naive Bayesian classifier. The inputs are the trajectory data of the ego vehicle and the interactive vehicle in the past two seconds, and the output is the driving style recognition result of the interactive vehicle. The proposed algorithm is not only tested in the public dataset but also combined with a decision-making algorithm. The results show that the algorithm shows the ability of accurate driving style recognition, adaptability to multiple scenarios, and potential in practical applications.