Gaze tracking algorithm: a bachelor thesis

rockNroll87q

Study of techniques for tracking the position and eye movements: a bachelor thesis

The gaze tracking devices can be defined as the equipment designed to measure the direction of the look of an individual. Are devices used primarily in assisting disability that means you can not motor.

The main objective of the thesis was the creation of a system able to recognize the area observed by the user in a grid of cells (5×2). You tried to create a system that is compatible with a multitude of users with different faces and, more importantly, that it required the least possible contact with the user.
As a result of these objectives, the choice fell on the use of a technique based on white light or requiring only the use of a standard built-in camera.
In particular, it is tried to realize an algorithm that can detect and track the limbo. The basic steps are
  • face detection;
  • the detection of the eye area;
  • the evaluation of the gaze.

To achieve this last point are exploited six different features. These features are divided into two categories: lateral and vertical. The features lateral measure the amount of sclera to the sides of the irises, while the vertical ones measure the white part above them. Careful combination of these values ​​allows the estimation of the direction of gaze with respect to the grid boxes.

The program has made ​​an initial calibration phase, where the user is encouraged to look at each of the boxes for a given period. Later, the user can form words and sentences. In the program have been implemented, a predictive dictionary and other aspects of writing such as cancellation of a letter or the selection of another word from the dictionary.

Most of the operations carried out have been completed with the OpenCV libraries. Take a look at the thesis (link – ITA) or the presentation (link – ITA).

 

References

[1] – Armanini, A., & Conci, N. (2010, April). Eye tracking as an accessible assistive tool. In 11th International Workshop on Image Analysis for Multimedia Interactive Services WIAMIS 10 (pp. 1-4). IEEE. (link)