Robust Eye Tracking Methods

This project aims at the development of robust and accurate 2D/3D eye tracking methods. We are particularly interested in eye tracking solutions that are suitable for the integration with head-mounted displays. We have been developing feature-based eye tracking methods, using multiple near IR LEDs to improve the robustness of eye illumination, the accuracy of pupil/cornea tracking, the range of eye movements tracking, as well as the simplification of tracking algorithm for speed concerns. We are also developing 3D tracking methods to obtain point of gaze information. This project has been funded by a three-year NSF HCI grant.

This project aims at the development of robust and accurate 2D/3D eye tracking methods. We are particularly interested in eye tracking solutions that are suitable for the integration with head-mounted displays. We have been developing feature-based eye tracking methods, using multiple near IR- LEDs to improve the robustness of eye illumination, the accuracy of pupil/cornea tracking, the range of eye movements tracking, as well as the simplification of tracking algorithm for speed concerns. We are also developing 3D tracking methods to obtain point of gaze information. More details are reported in Prasanna’s MS thesis.

Recovered Eyelid

Eyelash

Extended Tracking Range

Eye tracking algorithm

This project aim to build a real-time system for robust gaze direction estimation and tracking. Efforts are on to develop a pupil/corneal reflection based tracking method exploiting the optical properties of the human eye. We are working on computer vision techniques to extract features from eye images and track them in a video sequence. To this end, we have developed a real-time connected components labeling algorithm as a quick way to initialize pupil detection by zeroing in on the region of interest. We are currently experimenting with robust ellipse fitting techniques for pupil boundary detection and particle filters for subsequent tracking of the boundary through a video sequence.

The following shows a sample pupil boundary detection result.

pupil_detection

The following is a sample sequence of tracked pupil boundaries through the frames of a video.

pupil_tracking

Reference

  1. Hong Hua, C. Pansing, and J. P. Rolland, “Modeling of an eye-imaging system for optimizing illumination schemes in an eye-tracked head-mounted display,” Applied Optics, 46(32): 1-14, November 2007.