Eye Pose Detection with Calibrated 3D Model Fitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eye pose detection methods using 3D models are either oversimplified and computationally inefficient or overly complex, and require cumbersome analytical solutions, especially when more elaborate models are used.
Innovation Solution
A parameterized 3D model of the eye is calibrated by acquiring images with known gaze directions, identifying characteristic features, fitting them with a 2D projection, and numerically solving a set of equations to determine the eye's pose, utilizing glints and optical properties for precise alignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If an oversimplified 3D model is used with analytical solutions, then computational efficiency is improved, but measurement precision deteriorates
Solution Approach 1:
The patent transforms the eye pose detection problem from an analytical parameter-solving approach to a numerical optimization approach. By formulating the problem as minimizing the difference between projected 3D model features and detected 2D image features through numerical optimization, the system can handle complex, realistic eye models with multiple parameters (cornea curvature, pupil size, iris patterns) without requiring simplified analytical solutions, thereby maintaining both computational efficiency and high measurement precision.
2Measurement precision
If a complex 3D model is used, then measurement precision is improved, but device complexity worsens
Solution Approach 1:
The patent replaces complex mechanical/geometrical analytical modeling with numerical computation and optimization. Instead of deriving closed-form analytical solutions for complex 3D model projections, the system uses iterative numerical optimization to minimize the error between model projections and actual image features. This substitution allows complex realistic eye models to be used without proportionally increasing computational complexity, as the numerical approach handles model complexity efficiently through standardized optimization routines.
3Ease of operation
If analytical solutions are used for simple models, then ease of operation is improved, but measurement precision deteriorates
Solution Approach 1:
The patent replaces analytical solution methods with numerical optimization methods. The system formulates eye pose detection as an optimization problem where the objective is to minimize the difference between projected 3D model features and detected 2D image features. This numerical approach maintains computational simplicity through standardized optimization algorithms while enabling the use of complex, realistic eye models that provide superior measurement precision compared to simplified analytical models.
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
This approach allows for more computationally effective and accurate eye pose detection, enabling precise determination of the eye's position and orientation in six degrees of freedom, even with complex models, by using numerical methods and calibrated parameters.
Implementation Method 1
The tracking system includes a light source for illumination of the target area, so that the tracking image includes a refection in the cornea, a so called 'glint'.
Implementation Method 2
The set of equations includes equations based on the glint position in the tracking image, a known geometrical relationship between an image sensor and a light source, and a modeled cornea normal direction in the glint position.
Data Source
Figure 1
Figure 2
Figure 3a~3b
AI summary
A pose of an eye of a user is determined by providing a parameterized 3D model of the eye, said model including a set of parameters which have been calibrated, acquiring (step S11) at least one tracking image of the eye, identifying (step S12) a plurality of characteristic features in the acquired tracking image, fitting (step S13) said characteristic features with corresponding features of an optical projection of the calibrated 3D model, thereby forming a set of equations, and numerically solving (step S14) the set of equations to determine the pose of the eye.