Analysis of Driver’s Predictive Characteristics using Eye Tracking and a Deep Learning Model that Mimics Human Vision

Abstract

A human predictive characteristic may affect the traffic accident factors such as prediction failure. In our latest research, we observed the area in front view where a human pays attention unconsciously while driving, using deep neural networks which reproduce human vision perception. In this paper, we analyzed the relation between the observed area and the eye movement of a person who is looking at front-view recordings. As a result, the hypothesis that large prediction errors attract stronger attention is validated by measuring the gaze of 14 participants.