Calibration

 

  • Calibration is the starting point for an accurate eye tracking. However, depending on your use case, you may not need this. We offer a confidence level of 95% without calibration, which is suited in high traffic scenarios (e.g. retail). This section is more accurate for users that want to build a personal profile with the software.

 

  • Types of calibration
  • 1. Face Calibration – make slow and small rotational movements of the head in circles, around the center of the camera with your head, exposing all sides of your face to the camera for it to record it (i.e., follow the borders of your screen, clockwise and counter-clockwise). Keep moving until the video stops. You should be able to see how the face model, in blue in 3D View Mode, becomes increasingly more accurate, by filling the missing gaps. When the system has gathered enough data, it will start creating the face model automatically. This is a computationally intensive task, and your computer may suffer a brief framerate drop for about 10-20 seconds. This is normal behaviour.
  • Animated GIF - Find & Share on GIPHY

 

  • 2. Gaze Calibration
  • Points calibration – This setup corresponds to calibrating the eye tracking by looking at a virtual target drawn on the computer screen. This mode is only available when Screen Interaction is selected in the Attention Context Control menu. In this mode, under the Gaze Calibration Points menu, you can select how many sequential points on the screen to use: whether 1, 5 or 9 points. It is crucial for the screen-cam settings parameters have been set correctly, before using this calibration procedure, otherwise the gaze calibration will lead to unexpected results.
  • Animated GIF - Find & Share on GIPHY
  • Screen calibration – This setup corresponds to calibrating the eye tracking system by looking at the camera itself. This is advantageous as it does not depend on external objects or specific parameters to be well defined. This function is best used to calibrate when using API Mode.

Powered by BetterDocs