Research shows that webcam-based eye-trackers can achieve good accuracy and reliability, often matching traditional infrared systems in specific situations, particularly in laboratory environments [1].
RealEye collects data from standard webcams in various environments, providing flexibility and easy access for users. Consequently, the accuracy validation methodology needed to address the diverse range of conditions encountered in real-world settings.
Evaluating the Accuracy of RealEye on Desktops and Laptops
The Accuracy means how close the system’s prediction is to the actual point on the screen where a person is looking. To evaluate the Accuracy of RealEye on Computers, we performed a validation study that involved grid-based tasks (learn more in our Whitepaper). The measurements were taken in pixels – this is due to the fact that the distance of online users from the screen was not controlled, and screen size was self-reported, introducing uncertainty - and both these variables are crucial in calculating visual angle:
The results of the study showed that the average accuracy on Computers is ≈ 106 pixels. This means that, on average, the system’s prediction is within ~1/20 of the screen’s width and ~1/10 of the screen’s height.
For 72% of the cases, fixations landed within 150 pixels of the target center, 80% were within 200 pixels, and 91% were within 250 pixels. These results indicate that RealEye’s system performs reliably across a wide range of environments and user setups, making it a practical solution for remote eye-tracking research.
💡Keep in mind: the better the participant's hardware and software, the more accurate the results. Be sure to review the RealEye Hardware & Software Requirements.
The Importance of Accuracy When Creating AOIs
Areas of Interest (AOIs) are predefined sections of the screen that researchers focus on during an eye-tracking study. For example, AOIs might include buttons, menus, or specific sections of content.
The Accuracy of the eye-tracking system impacts the size of defined AOIs. To minimize misclassification errors, AOIs should be large enough to accommodate the system's accuracy, but not so large that they include irrelevant parts of the screen. Based on RealEye’s system performance on Computers, minimum AOI size cover approximately 5% of the width and 10% of the height of the displayed content to ensure reliable fixation detection.
[1] Wisiecka, K., Krejtz, K., Krejtz, I., Sromek, D., Cellary, A., Lewandowska, B., & Duchowski, A. (2022, June). Comparison of webcam and remote eye tracking. In 2022 Symposium on eye tracking research and applications (pp. 1-7).
💌 If you have any questions or need further assistance, feel free to reach out to our team at contact@realeye.io