Isophote-based Low-Computing-Power Eye-Detection Embedded-System

Abstract

The eye tracking field is an area highly studied by different researchers. In addition, there are companies that offer commercial eye trackers, but they are usually economically restrictive. However, current technology allows to design embedded eye tracking systems, capable of running on low-cost hardware. This paper presents the implementation of a real-time eye-detection method that uses the properties of isophotes, to achieve robustness against changes in illumination, eye rotation and pupil size. The method is implemented in a portable platform, reduced both in computing power and in RAM memory. The performance is evaluated using different eye tracking methods implemented by the authors on the same platform. For this purpose, a database generated on this hardware is used. This database is composed of a set of low-resolution near-infrared eyes-images created with the dark pupil technique, using a lightweight glasses head unit. Results show the robustness of the method and a significant accuracy improvement over the methods analyzed.