DeepEyeTracking - Robust eye tracking with deep neural networks in vehicles
Overview
Background & policy context:
The safe transfer of control between the vehicle and the driver is an important prerequisite for automated driving. Eye tracking can be used to assess whether the driver is capable of taking over control. It is of great importance that such systems are robust against rapidly changing lighting conditions and eyeglasses.
Objectives:
The project "DeepEyeTracking" aims to develop a novel system for determining the direction of vision of drivers. The new system will be tested in practice together with automotive suppliers and the functionality will be evaluated.
The planned system for direction of vision recognition makes an important contribution to the establishment of safe procedures for automated driving. In addition to this primary use, improved eye tracking systems can also be used in many other areas such as consumer electronics.
Methodology:
In contrast to conventional approaches in which the direction of vision is determined by light reflections from the pupil, retina and cornea using a three-dimensional physiological model of the eye, the project will use deep neural networks that determine the direction of vision from camera images of the head position, head orientation and eye sections. This innovative approach has fundamental advantages, as it is significantly less susceptible to interference from bright daylight or reflections, for example from spectacle lenses.
Share this page