Abstract: A research on longitudinal stimuli-based classification and recognition for driving style were carried out to make Advanced Driver Assistance System ( ADAS) work in a more human-like or personalized way and to im- prove the safety and comfort for intelligent vehicles． Six typical longitudinal driving stimuli of the leading vehicle were designed based on the periodicity and mutability of the leading vehicle's speed，and data collection for 64 dri- vers was conducted in field test． The corresponding driving style was defined and classified by combining particle swarm optimization clustering ( PSO-Clustering) method with subjective questionnaire． The optimal longitudinal stimulus set，the Sine NO． 3 and Step NO． 3，was obtained by comparing the classification results under different stimulus． The recognition model for driving styles based on the multi-dimension Gaussian hidden Markov process ( MGHMP) was modeled． And the optimal model input set was obtained based on the recognition accuracy and key parameters were optimized by the orthogonal test method． Ｒesults show that the longitudinal stimuli based classifica- tion and recognition for driving styles can achieve better classification and identification accuracy．

Key words: vehicle engineering, driving style, particle swarm clustering, multi-dimension Gaussian hidden Mar- kov process, advanced driver assistance system