Researchers and students from Skoltech (part of the VEB.RF group) and Central University have developed a new method that significantly improves eye tracking accuracy in difficult circumstances. Their research was published in the proceedings of the 8th International Youth Conference on Radio Electronics, Electrical and Power Engineering, where it won the Best Paper Award.

The new approach proposed by the Russian team is twice as accurate as other similar methods. It will make eye trackers more efficient and applicable to various fields. For example, it can be used to measure gaze speed in patients undergoing rehabilitation, to evaluate behavioral patterns in neurology and eSports, and to improve the user experience in working with graphical interfaces.

Eye trackers are usually installed next to a monitor. They consist of an infrared emitter that illuminates the face and a camera that detects IR reflections in the pupil. The software then converts these reflections into screen coordinates, enabling the analysis of eye movements across the entire screen.

However, most widely available solutions tend to be less accurate. They often fail to focus when faced with obstacles such as glasses, contact lenses, or glare caused by bright light. The lack of accurate eye tracking prevents this technology from advancing further.

Researchers from Skoltech and Central University have found a solution: a new eye-tracking algorithm. This new approach ensures high speed and accuracy when determining the position of the pupil under complicated circumstances, such as glasses, bright overhead lighting, or slight head movements.

The proposed algorithm has two steps. First, the camera uses a special IR illuminator to capture two images: one with a bright pupil and the other with a dark one. Distinguishing between the two images helps differentiate the pupil from the glare, making subsequent processing of eye movements much easier.

In the second step, the selected objects are grouped using the K-means method, an iterative grouping algorithm that divides the dataset into a predetermined number of classes, such as “pupil”, “glare”, and “background”. The eye tracking algorithm then uses this data and the calibration procedure to calculate the gaze point on the monitor screen.

The new approach was tested under controlled conditions. Participants were positioned 50 to 60 centimeters from the camera and presented with three scenarios: one with overhead lighting and no glasses, one with overhead lighting and glasses, and one with no overhead lighting or glasses.

The test demonstrated significantly better eye-tracking performance in terms of both frame processing speed and accuracy under challenging conditions. Pupil detection accuracy increased by 64% with glasses and by 27% in bright light. In this study, the final gaze detection error was approximately 16 pixels on a Full HD screen, nearly half that of previous approaches.

Andrey Somov, an associate professor at the Skoltech Engineering Center, a professor at Central University, and the head of the Central University Laboratory of Intelligent Sensor Systems, commented: “This new approach to eye trackers will make the technology more accessible for mass adoption by minimizing errors. We’ve succeeded in increasing the accuracy of eye trackers in difficult conditions, which will help expand their functionality and usability in a wider range of applications. Importantly, the study involved young researchers who will continue working in the laboratory to refine and validate the eye trackers’ algorithms on real devices in challenging scenarios.”

Daria Pechenova, a student in the Skoltech “Engineering Systems: Aerospace, Robotics, and Digital Engineering” MSc program, added: “Imagine a neurologist assessing a patient with suspected Parkinson’s disease based on their gaze patterns. Or imagine a teacher noticing which words a child lingers on while reading so they can focus on them in future lessons. With the widespread use of eye trackers, our new algorithm could make this possible. This is my first research project in the field of intelligent engineering systems. It’s especially important that our results were presented at an international conference and received high praise, including a Best Paper Award.”

 Credit: Central University PR Office