Remote Eye Tracking With Pupil–Glint Candidate Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eye tracking systems face challenges in accurately identifying eye positions, particularly in cases where machine learning data is unavailable, such as with non-human primates or mannequin heads, and struggle to distinguish between true and false glints due to reflections from non-illuminator light sources.
Innovation Solution
A method and system using off-axis infrared illuminators and image sensors to capture dark pupil images, determining pupil and glint candidates through brightness thresholds, forming pupil-glint candidate groups, and calculating a score value to identify eye positions reliably without relying on machine learning data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning methods are used to determine eye position, then accuracy of gaze position estimation is improved, but the system becomes impractical for exceptional cases such as non-human primates or subjects with only one eye due to lack of training data
Solution Approach 1:
The patent uses simple geometric models (pupil as dark region, glints as bright regions) that can be quickly instantiated and discarded for each image, replacing complex machine learning models that require extensive training data. This allows the system to handle exceptional cases without needing subject-specific training data.
Solution Approach 2:
The patent changes the approach from learning-based parameter estimation to physics-based parameter estimation using geometric relationships. By changing from data-driven parameters to physics-driven parameters (light reflection geometry, pupil-cornea relationships), the system becomes applicable to all subjects regardless of training data availability.
2Measurement precision
If active illumination is used to generate glints for eye detection, then pupil and glint identification is improved, but false glints from other light sources or reflective surfaces complicate the detection
Solution Approach 1:
The patent applies different analysis criteria to different regions of the image. True glints are identified by their specific geometric relationship with the pupil and cornea, while false glints are rejected based on their inconsistent spatial relationships. This localized quality assessment distinguishes true from false glints.
Solution Approach 2:
The system uses the detected pupil position and cornea position as feedback to validate glint candidates. Only glints that conform to the expected geometric relationships with the already-detected pupil and cornea are accepted as true glints, creating a feedback loop that filters false detections.
3Loss of information
If full face images are used in remote eye trackers, then comprehensive facial information is captured, but additional algorithms are required for pupil detection and eye feature extraction
Solution Approach 1:
The patent extracts only the essential features needed for eye tracking (pupil as dark region, glints as bright regions, cornea position) from the full face image, discarding unnecessary information. This extraction approach simplifies the processing pipeline while maintaining the benefits of using full face images.
Solution Approach 2:
The patent segments the image processing into distinct stages: first detecting the pupil as a dark region, then detecting glints as bright regions, then using their geometric relationships to determine eye position. This segmentation reduces overall algorithmic complexity by breaking down the problem into manageable steps.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Provides a computationally efficient and reliable method for determining eye positions in various subjects, including non-human primates, by distinguishing true glints from false reflections, enhancing accuracy and robustness in eye tracking systems.
Implementation Method 1
illuminating a face of a subject using at least one infrared (IR) illuminator
Implementation Method 2
active illumination of the subject's face is needed to generate the at least one glint in the resulting image
Data Source
AI summary
The invention is related a method and a remote eye tracking system for determining a position of at least one eye of a subject. The method comprises the steps of: illuminating a face of a subject using at least one infrared (IR) illuminator off-axis from at least one image sensor; capturing at least one image of the face using the at least one image sensor at a time instant; and using a processing circuitry of the remote eye tracking system, performing processing steps of: determining at least one pupil candidate associated with the at least one image; determining at least one glint candidate associated with the at least one image; determining at least one pupil-glint candidate group, comprising at least one pupil candidate and at least one corresponding glint candidate; generating a score value for each of the at least one pupil-glint candidate groups; and determining an eye position of at least one eye of the subject in the at least one image based on the pupil-glint candidate group with the highest score value.