Driver's Gaze Estimation Algorithm Based on Deep Learning

Abstract

In traffic safety accidents, 80% of accidents are caused by driver distraction,there is an important mapping relationship between the direction of human sight and the focus of attention,studying the driver's gaze estimation algorithm can reduce the incidence of vehicle traffic accidents.In this paper, a driver's gaze estimation system based on deep learning technology is designed.For human eye feature localization, a multi-level localization method is adopted,firstly, the detection network is used to locate the face position, and then the key points of 68 faces are located based on the face image.In the aspect of the expression of the human eye's line of sight, a method of the expression of the human eye's line of sight based on the central point of the pupil and the direction vector of the human eye's line of sight is adopted,a convolution neural network is used to return the line of sight direction.On the MPIIGaze public data set and self-made data set, the experimental results show that the method can accurately estimate the human eye line of sight.