Skip to content

vision-toolkit#

tarsier

Summary#

vision-toolkit is a Python library for processing eye tracking data and extract from them:

  • Scanpaths: A macro trajectory of the gaze without taking into account all the little movements of the eyes around what they see
  • Areas of Interest: Zones the viewer is attracted to.

Below a video showing the gaze of multiple viewers watching a movie scene:

hollywood2-eye-tracking-example

vision-toolkit supports:

  • multiple kind of coordinates:
    • 2D: representing a point on a screen. Coordinates could be cartesian or angular.
    • 3D: an estimation of where both eyes gazes converge, a.k.a. point of gaze
  • multiple kind of frames:
    • static frame (the head stays still, e.g. medical settings)
    • dynamic frame (the head moves, e.g. glasses or VR sets)
  • using data from multiple viewers to analyse the same scene or event

Papers#

In order to have a broad picture of what this package is meant to implement, you can document yourself with this 4-part paper

  1. Neurophysiological Foundations and experimental paradigms in Eye-Tracking Research
  2. Features and Metrics for assessing Oculomotor Signal
  3. Scanpaths and Derived Representations for Gaze Behavior Characterization
  4. Area of Interest and Associated Algorithms

Credits#

This work was initiated by Quentin Laborde as a PoC related to his PhD in a WIP/draft shape while the efforts to industrialize it and bring it to a production ready state were passed on to Robert Dazi.