Eye-tracking
Lexplore is a company that offers an innovative tool for reading assessment in schools to identify dyslexia at an early age. By combining eye tracking with machine learning, Lexplore enables an accurate assessment of a student’s reading ability in minutes. The results can then be used by teachers for individualized intervention and progress monitoring.
Most eye-trackers rely on dedicated, expensive hardware to track eye movements. Eye-tracking on mobile devices using their built-in camera would be beneficial on a large scale since additional, expensive, equipment would not be needed. In this project I implemented a mobile device eye-tracker using AR to evaluate the accuracy and precision of such system.
The eye-tracker was implemented in Swift using ARKit to track the head and eyes. The methods used for gaze estimation was a ray-plane intersection in which the mobile device was modelled as a 2D plane (see image). Including a system-controlled calibration using linear scaling and translation to adjust the future gaze positions.
The accuracy and precision were evaluated by two validations using a point pattern to collect gaze data on fixations detected by the system. The conclusions from the study are that mobile device eye-tracking is feasible, easy to implement and delivers reasonable results in terms of accuracy and precision.
