Evaluating Short-Term Driving Style in Different Driving Scenes

Driving is a personalized activity. A driver’s psychological activity significantly affects behavior and driving style. Traditional research has divided driving styles into several types, giving the driver a constant style label. However, studies have not considered the influence of driver’s short-term psychological activities on behavior. This paper proposed a new concept called “driving proneness” to evaluate the driver behaviors with respect to whether they are prone to traffic accidents or not. Real data about vehicle dynamics were collected from sixteen drivers; the driving segments, or “scenes,” analyzed included car-following, lane-changing, and car overtaking. The time series data capturing vehicle dynamics were represented using symbolic aggregate approximation. The statistical results show that the parameters representing acceleration, braking, and steering behavior align with different distribution patterns during different driving scenes. Then, an evaluation model to assess “driving proneness” was established using the Hidden Markov Model (HMM)—Viterbi. The results provide an accident proneness score to each diver in different diving scenes.