Source-linked AI summary
A Survey on Eye-Gaze Tracking Techniques
H. R. Chennamma, Xiaohui Yuan
TL;DR
Eye-gaze tracking is a challenging computer-vision problem used in HCI and other fields. This paper reviews remote video-based gaze tracking and its methods. It concludes that the field remains unstandardized and identifies priorities for future development.
Problem
Method
The paper reviews non-contact video-based gaze tracking, including image-based estimation methods and single- and multi-camera systems.
Source: §2.4.1 · p. 3, §2.4.2 · p. 3, §3 · p. 4Results
The review concludes that eye-gaze tracking remains unstandardized, including the minimum duration for a fixation.
Source: §4 · p. 5Takeaways & Limitations
The paper identifies standardizing metrics, reducing equipment intrusiveness, improving capture robustness and accuracy, and lowering costs as future needs.
Source: §4 · p. 5The review is scoped to non-contacting video-based gaze tracking.
Source: §4 · p. 5
Abstract
from arXiv · showhide
Study of eye-movement is being employed in Human Computer Interaction (HCI) research. Eye - gaze tracking is one of the most challenging problems in the area of computer vision. The goal of this paper is to present a review of latest research in this continued growth of remote eye-gaze tracking. This overview includes the basic definitions and terminologies, recent advances in the field and finally the need of future development in the field.
1. Introduction
The introduction distinguishes measuring eye movement from analyzing eye-tracking data to determine gaze in a head or visual scene. It frames gaze tracking as useful across HCI and many other fields.
1. Introduction
- Eye tracking measures eye movement or activity, while gaze tracking analyzes that data relative to the head or visual scene.The paper says these terms are often used interchangeably.Source: §1 · p. 1
- Systems that attach devices to the skin or eye are intrusive; systems without physical contact are remote or non-intrusive.Gaze location is usually computed by integrating eye and head position.Source: §1 · p. 1
2. Methods of Eye Tracking
The section surveys eye-tracking approaches from contact sensors to remote video systems, then describes single- and multi-camera designs. Its focus is remote video-based gaze tracking.
2. Methods of Eye Tracking
- The paper groups eye-motion recording under oculography and describes four tracking methods.The supplied discussion covers electro-oculography, scleral search coils, infrared oculography, and video oculography.Source: §2 · p. 1
2.1. Electro-Oculography
- Electro-oculography estimates eye position from changes in skin potential measured by electrodes around the eyes.It is inexpensive and can detect movements with the eyes closed, but requires contact and its signal may change without eye movement.Source: §2.1 · pp. 1–2
2.2. Sceleral Search Coils
- Scleral search coils use wire coils embedded in a contact lens to measure eye position through their orientation in a magnetic field.The approach offers high accuracy and near-unlimited time resolution, but is invasive and, to the authors’ knowledge, had not been used for HCI gaze tracking.Source: §2.2 · p. 2
2.3. Infrared Oculography
- Infrared oculography estimates eye-position changes from differences in infrared light reflected by the eye.It can measure in darkness, but is sensitive to external-light changes and is limited to about ±35 degrees horizontally and ±20 degrees vertically.Source: §2.3 · p. 2
2.4. Video Oculography
- Video oculography uses one or more cameras to infer eye movement from captured images and is widely used in commercial trackers.The paper focuses on remote video systems, where gaze estimation uses changes in eye images as the eye rotates or translates.Source: §2.4 · p. 2
2.4.1. Single Camera Eye Tracker
- Single-camera systems face a field-of-view trade-off: capturing high-resolution eye images can constrain the area available for user movement.The paper describes designs using multiple light sources and reports a commercial system with about 1-degree accuracy and good tolerance to user movement.Source: §2.4.1 · p. 3
2.4.2. Multi-camera Eye Tracker
- Multi-camera systems combine cameras for eye imaging and head tracking to estimate gaze while accommodating a wide field of view.Reported configurations include stereo cameras for pupil position and additional cameras or movable optics for tracking the eye region.Source: §2.4.2 · pp. 3–4
3. Methods of Gaze Tracking
The paper organizes image-based gaze estimation into feature-based and appearance-based approaches. Feature methods use local eye cues or geometric and learned mappings, while appearance methods map image content directly to gaze.
3. Methods of Gaze Tracking
- Video gaze estimation detects the eye in image data, then estimates gaze direction using eye-region information and possibly head pose.Illumination, viewing angle, occlusion, and head pose make eye detection and tracking challenging.Source: §3 · p. 4
3.1. Feature-based Gaze Estimation
- Feature-based estimation uses local eye features such as contours, eye corners, and corneal reflections, and divides into model-based and interpolation-based approaches.The methods can lose accuracy outdoors or under strong ambient light, particularly when iris and pupil features are unavailable.Source: §3.1 · p. 4
3.1.1. Model-based approaches
- Model-based methods reconstruct optical and visual axes in 3D, then intersect the visual axis with scene geometry to estimate gaze.They generally require camera calibration and a geometric model of the light sources, camera, and monitor.Source: §3.1.1 · p. 4
3.1.2. Interpolation-based approaches
- Interpolation-based methods fit a mapping from image features to 2D or 3D gaze coordinates using calibration data rather than explicitly modeling eye geometry.Mappings may be parametric, such as polynomial functions, or nonparametric, such as neural networks.Source: §3.1.2 · p. 4
3.2. Appearance-based Gaze Estimation
- Appearance-based methods map image content directly to gaze or screen coordinates, typically without camera calibration or geometric data.The survey names morphable models, grayscale unit images, appearance manifolds, Gaussian interpolation, and cross-ratio methods.Source: §3.2 · pp. 4–5
4. Summary and Conclusion
The conclusion reviews non-contact video-based gaze tracking and identifies standardization, comfort, robustness, accuracy, and cost as areas for future development.
4. Summary and Conclusion
- The field lacks standardization of eye-movement metrics, their names, and their interpretation in interface design.The paper gives the minimum fixation duration as an example for which no standard exists.Source: §4 · p. 5
- Future systems should reduce equipment intrusiveness, improve capture robustness and accuracy, and lower costs for smaller agencies and research labs.The paper identifies these as needs for making gaze tracking a viable usability tool.Source: §4 · p. 5