Research on imaging method of driver's attention area based on deep neural network

Abstract

Abstract In the driving process, the driver's visual attention area is of great significance to the research of intelligent driving decision-making behavior and the dynamic research of driving behavior. Traditional driver intention recognition has problems such as large contact interference with wearing equipment, the high false detection rate for drivers wearing glasses and strong light, and unclear extraction of the field of view. We use the driver's field of vision image taken by the dash cam and the corresponding vehicle driving state data (steering wheel angle and vehicle speed). Combined with the interpretability method of the deep neural network, a method of imaging the driver's attention area is proposed. The basic idea of this method is to perform attention imaging analysis on the neural network virtual driver based on the vehicle driving state data, and then infer the visual attention area of the human driver. The results show that this method can realize the reverse reasoning of the driver's intention behavior during driving, image the driver's visual attention area, and provide a theoretical basis for the dynamic analysis of the driver's driving behavior and the further development of traffic safety analysis.