Breakthrough in Eye-Tracking Technology by University of Arizona Researchers
Eye tracking is an essential technology used in a range of fields including virtual and augmented reality headsets, entertainment, medical and behavioral sciences, automotive driving assistance, and industrial engineering. However, accurately tracking eye movement has always been a significant challenge.
Researchers from the University of Arizona’s Wyant College of Optical Sciences have made a groundbreaking advancement in this field. Their research, published in Nature Communications, reveals that integrating deflectometry, an advanced 3D imaging technique, with sophisticated computational methods could substantially enhance the precision of eye-tracking technology.
Jiazhang Wang, the study’s lead author and a postdoctoral researcher in Willomitzer’s lab, explained: “More data points enable better accuracy in estimating gaze direction, which is crucial for next-generation virtual reality applications. Our method increases the number of data points by over 3,000 times compared to conventional techniques.”
Deflectometry is a high-accuracy imaging method used to measure reflective surfaces, commonly applied in the inspection of telescope mirrors and precision optics. The research team at the University of Arizona has expanded deflectometry’s use, pairing it with computational techniques from computer vision. This novel approach, which Willomitzer refers to as “computational deflectometry,” has been applied in fields ranging from artwork analysis to skin lesion measurement and, notably, eye tracking.
“The combination of accurate measurement and advanced computation allows for ‘superhuman vision’—enabling machines to perceive things beyond human capabilities,” said Willomitzer.
In their study, the team conducted experiments involving both human participants and a realistic artificial eye model. They were able to track gaze directions with accuracy ranging from 0.46 to 0.97 degrees in real eyes, and an impressive 0.1 degrees on the artificial eye model.
The new technique departs from the traditional approach, which relies on a few infrared light sources. Instead, it uses a screen displaying structured light patterns, with each pixel on the screen acting as an individual light source. By analyzing how these patterns deform when they reflect off the eye’s surface, the team can generate precise 3D surface data from the cornea and sclera (the white part of the eye). This data is then processed using advanced algorithms to predict the gaze direction.
In earlier studies, the team demonstrated how this technology could be integrated into virtual and augmented reality systems. By using a fixed pattern embedded in the headset or reflected from visual content, the system could streamline complexity. The potential to use infrared light for future applications could further reduce distractions from visible patterns.
“This technique helps us capture gaze information from the eye’s surface with minimal ambiguity. Unlike other methods, our approach doesn’t make assumptions about the eye’s shape or surface, which can vary from person to person,” Wang explained.
Additionally, the new method generates detailed 3D reconstructions of the eye, which could pave the way for diagnosing and treating eye disorders.
While this is the first time deflectometry has been applied to eye tracking, the team’s initial results already show comparable—or better—accuracy than commercial systems. With a patent pending and plans for commercialization via Tech Launch Arizona, this research promises to usher in a new era of more accurate and reliable eye-tracking technology.
The researchers aim to further refine their technology, incorporating more 3D reconstruction methods and artificial intelligence to push the limits of eye tracking. Their ultimate goal is to reach the 0.1-degree accuracy demonstrated with the artificial eye models, enabling advanced applications in neuroscience, psychology, and other fields.
The study also included contributions from Oliver Cossairt, adjunct associate professor at Northwestern University, along with former students Tianfu Wang and Bingjie Xu, who were involved in the project.
