Eye Tracking Research

Following my final year dissertation, I continued working at the University of St Andrews alongside researchers to advance the project.

In this role, I collaborated with academics working on three papers, two of which have been published:

• TinyGaze: Lightweight Gaze-Gesture Recognition on Commodity Mobile Devices

  Lei, Cho, Buchanan et al., CHI 2026

• Quantifying the Impact of Motion on 2D Gaze Estimation in Real-World Mobile Interactions

  Lei, Wang, Buchanan et al., 2025

Use Motion

Keep your head still and take a photo of your eyes from below, now from above. We quantified how motion degrades gaze estimation and proposed corrections for real-world mobile use.

Transform

Eye tracking models need large, varied datasets but collecting real gaze data is expensive. By using head and phone position, we apply geometric transformations to existing eye images, synthetically generating new viewpoints and greatly expanding training data without additional capture.

Identity

Without a depth sensor, we use eye saccades as a form of identification. Saccade patterns are unique to each individual and near-impossible for deep fakes to replicate, making them a promising biometric for authentication.