3D Cornea Position Estimation Using Kalman Filter
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eye tracking systems face inaccuracies and noise in cornea position estimation when the eye is moving, particularly due to challenges in identifying false glints and updating the cornea position effectively.
Innovation Solution
A method using a Kalman filter-based cornea movement filter to predict glint positions and iteratively refine the 3D cornea position estimation by selecting the most probable glint pairs, thereby continuously updating the filter with the latest position and velocity information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing cornea position estimation solutions are used, then the system can provide basic eye tracking functionality, but the accuracy deteriorates when the cornea is moving due to noise in distance measurements
Solution Approach 1:
The system performs preliminary actions by predicting the cornea position and glint positions before actual measurement using a Kalman filter model. This allows the system to prepare expected values that guide the identification process, preventing noise from derailing the measurement when the eye is moving.
Solution Approach 2:
The system implements feedback by continuously updating the Kalman filter with new measurements and using the filtered estimates to improve subsequent measurements. The predicted cornea position and glint positions are fed back into the identification process, creating a closed-loop system that reduces noise accumulation over time.
2Device complexity
If traditional glint identification methods are used, then the process is simple, but false glints cannot be accurately identified when the cornea is moving
Solution Approach 1:
The system performs preliminary prediction of glint positions based on the moving cornea model before identifying actual glints in the image. This preliminary action provides expected locations that guide the search process, making it easier to distinguish true glints from false ones even when the cornea is in motion.
Solution Approach 2:
The Kalman filter acts as an intermediary between the raw image data and the final glint identification. It provides predicted glint positions that mediate the matching process, allowing the system to accurately identify false glints by comparing actual detections against the predicted positions from the dynamic model.
3Measurement precision
If the system continuously updates cornea position, then tracking accuracy improves, but noise in distance measurements increases when the eye is moving
Solution Approach 1:
The system performs preliminary prediction of the cornea position using the Kalman filter before updating with new measurements. This preliminary estimate acts as a reference that prevents noise from causing large deviations in the updated position, allowing continuous tracking while filtering out measurement noise.
Solution Approach 2:
The system uses feedback from the Kalman filter's state estimation to guide the update process. The filter combines predicted position with new measurements in a weighted manner, where the weighting is determined by the estimated uncertainty. This feedback mechanism allows continuous updating while systematically reducing the impact of noise on the final gaze information.
Data Source
Figure 1~3
Figure 4~6
Figure 7
AI summary
There is provided a method, system, and non-transitory computer-readable storage medium for performing three-dimensional, 3D, position estimation for the cornea center of an eye of a user, using a remote eye tracking system (210), wherein the position estimation is reliable and robust also when the cornea center moves over time in relation to an imaging device (110) associated with the eye tracking system (210). This is accomplished by generating, using, and optionally also updating, a cornea movement filter, CMF, in the cornea center position estimation.