RealEye is an online platform that enables researchers and businesses to conduct eye-tracking studies using standard webcams without the need for expensive infrared sensors or specialized hardware (see: Eye-tracking study - offline vs. online). With the power of computer vision and machine learning, RealEye processes webcam footage in real-time to predict where a participant is looking is looking at the screen using a traditional webcam, which is built into most modern devices.
To learn about RealEye's accuracy and its validation methodology in detail, we encourage you to read our White Paper.
RealEye webcam eye-tracking technology
To conduct a study using RealEye, a webcam and an internet connection are required. Built-in cameras on most laptops and smartphones are generally sufficient. However, for desktop computers, it is recommended to use a webcam with a resolution of at least 1080 pixels at 30 frames per second, ensuring a minimum frame rate of 20 frames per second. RealEye platform supports eye-tracking tests at a sampling rate of 60 Hz, requiring a webcam capable of recording at 60 FPS to achieve this. For smartphones, studies must be conducted with the phone held vertically and with a front-facing camera. It is advisable that the device runs an up-to-date version of the operating system (Windows, macOS, or Linux for computers, and iOS or Android for smartphones), along with a modern web browser such as Google Chrome, Microsoft Edge, Mozilla Firefox, Opera, or Safari. Keeping both the operating system and browser updated ensures compatibility with the latest features and security protocols, which is crucial for optimal performance.
For further details on the recommended setup, refer to the RealEye System Requirements.
To try RealEye on Computers, participate in our Demo Study for Desktops and Laptops. To experience RealEye on Smartphones, visit Demo Study for Smartphones or scan the QR code below:
Fixation estimation pipeline
RealEye’s fixation estimation pipeline operates through two modules: Image Analysis and Gaze Estimation. The former uses a machine learning model to analyze the incoming images captured by the participant’s webcam, detecting eye positions by identifying key features, such as pupils and corners of the eyes. These isolated regions serve as input data for the subsequent Gaze Estimation Module, allowing the system to establish a baseline for where the participant’s gaze is directed.
Once the eye positions are identified and key areas isolated, the Gaze Estimation Module employs linear regression and other predictive algorithms to estimate the participant’s gaze position with greater accuracy. During calibration, the system uses predefined points and backgrounds to fine-tune the model. Calibration ensures the system can adapt to individual differences in eye movements and visual focus, enhancing the overall accuracy of gaze prediction.
The fixation estimation pipeline runs directly on the participant’s device, ensuring all data processing occurs locally and reduces delays in obtaining gaze predictions. This localized approach benefits both performance and privacy, as it eliminates the need to transmit raw video data to external servers, keeping all sensitive information securely on the participant’s computer. Only the calculated gaze predictions are stored in a secure text format like: Timestamp: 10, GazePointX: 76, GazePointY: 24, with gaze positions estimated as percentage values of the displayed item. RealEye’s solution therefore complies with [EU regulations (GDPR)].
Fixation filter
A Fixation is a series of Gaze points that are closely clustered in time and space, where the user's gaze stays long enough (typically around 200-300 milliseconds) to allow them to focus and process visual information. To determine whether a set of gaze points qualifies as a fixation, specific [Fixation Filter] parameters must be defined. RealEye employs a method similar to the I-VT (Velocity-Threshold Identification) fixation filter for analyzing the gaze data.
The default fixation filter settings in the RealEye Analytics Dashboard are shown below. These parameters are preconfigured but can be adjusted to suit your specific needs.
Virtual chinrest
Although participants are encouraged to keep their heads still for optimal eye-tracking accuracy, this is not strictly necessary and can be challenging to control in real-world, non-laboratory settings. To allow for more natural behavior, RealEye developed a "virtual chinrest" – a software solution that adjusts eye-tracking calculations to accommodate subtle head movements. It provides real-time feedback by displaying green dots on the screen whenever there's a noticeable shift in the participant's head position, indicating the recommended eye positions for maintaining stable head alignment.
The Virtual Chinrest initializes after a change in the participant's position compared to the Calibration process. The study continues once the participant's eyes remain fixed on the green dots for ~5 seconds. Consequently, the virtual chinrest enhances the overall user experience by increasing comfort levels while maintaining accuracy.
💌 If you have any questions or need further assistance, feel free to reach out to our team at contact@realeye.io