LAUSR

Accurately describing and classifying driving style is crucial for driving safety intervention strategies in the design of advanced driver assistance systems (ADASs). This paper presents a novel driving style classification method based on constructed driving operational pictures (DOPs) which map sequential data from naturalistic driving into 2-D pictures. By using the nested time window method, 798/1683/1153 DOPs sized 42 (features) $\times60$ (seconds) were generated for three different driving styles (low-risk, moderate-risk, and high-risk), respectively. The three kinds of neural network algorithms, i.e., convolutional neural network (CNN), long short-term memory (LSTM) network, and pretrain-LSTM were applied to recognize driving styles based on DOPs. The results showed that CNN performed the best with an accuracy of 98.5%, better than the traditional support vector machine (SVM) method. This study provides a new perspective to classify driving style which may help design ADASs operating characteristics to improve driving comfort and safety.