Fatigue driving is one of the most important causes of traffic accidents. An efficient stable algorithm is a core for detecting non-fatigue state or fatigue state. Relevance vector machine (RVM), as a frontier method, provides a potential way to meet this requirement. In this study, the driver's pulse index and grip strength of steering wheel index were selected as multi-source dependent variables, and a previous classification foundation (objective survey) of driving non-fatigue and fatigue is established through literature and investigation. Based on this, a RVM model for detecting driving state automatically is proposed reasonably. A total of 996 groups of driving status data (half of non-fatigue and half of fatigue) were obtained, and the results of driving state recognition were analyzed by different RVM classifiers. Result shows that the recognition accuracy of the RVM driving state classifier with different kernel functions is higher than 90%, which indicates that pulse index and grip strength of steering wheel index selected in this paper are indeed closely related to driving state and the driving fatigue state identification method designed in this paper has certain significance for improving the accuracy of fatigue driving detection. More importantly, it provides a scientific theoretical basis for the development of dangerous driving state warning.

CICTP