Driving Style Recognition Based on Multi-source Information and Multi-stimulus Paradigm

Aiming at the correlation between sudden stimuli and driver behavior in vehicle driving environment, this paper starts with the correlation between stimuli and driving style recognition and proposes the most significant multi-dimensional stimulus paradigm that affects the accuracy of driving style recognition. Specifically, the single mode stimulus is selected in the typical vehicle driving environment to create a multi-dimensional stimulus paradigm scene. The EEG signal and vehicle data of drivers under the multi-dimensional stimulus paradigm were collected and pre-processed by the brain-computer-environment synchronous experimental platform. Based on the collected EEG signal and vehicle data, the characteristic parameters of driving style were obtained by principal component analysis and the driving style was classified by K-means clustering algorithm. The neural network recognition algorithm is used to recognize the driving style. The results showed that under the multi-dimensional stimulus paradigm, such as driving straight on a sunny day with vehicle cue light flashing, driving straight on a sunny day with vehicle cue light flashing and intelligent voice interaction stimulus, driving straight on a rainy day with intelligent voice interaction stimulus, driving straight on a rainy day with vehicle cue light flashing and intelligent voice interaction stimulus, driving straight on a snowy day with voice interaction stimulus. The driving style recognition accuracy of the model is 87.5%, 75%, 83.3%, 70.8% and 87.5%, respectively, which significantly decreases compared with no stimulus, which lays a foundation for further optimization of the intelligent vehicle driving style recognition algorithm considering specific environment.