Differential Gaze Estimation With Reference Eye Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gaze estimation methods struggle with accuracy and robustness, particularly in low-resolution images, due to issues like eye alignment, eyelid closure, and illumination variations, and are sensitive to head pose variability.
Innovation Solution
A regression model-based deep neural network is trained to estimate gaze differences between multiple images, using a differential approach that includes convolutional neural networks and batch normalization to enhance accuracy and robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If appearance-based regression methods are used for gaze estimation, then the method can handle low to mid-resolution images, but the accuracy is limited to around 5 to 6 degrees with high variances and biases between subjects
Solution Approach 1:
The method segments the gaze estimation problem into two independent components: subject-specific parameters (eye shape, size, position) and gaze-related parameters (gaze position, head pose). By training separate regressors for each component using different training strategies, the method achieves both high accuracy and cross-subject robustness without the trade-off present in traditional appearance-based methods.
2Measurement precision
If geometric models with eye feature extraction are used, then accurate gaze can be inferred, but the method requires high resolution eye images and is prone to noise and illumination variations
Solution Approach 1:
The method extracts and isolates the subject-specific eye parameters from the appearance data, separating them from the gaze-related information. This extraction allows the system to compensate for variations in noise and illumination by focusing on stable anatomical features while using robust regressors that are less sensitive to image quality issues.
Solution Approach 2:
The method changes the parameter representation from direct appearance-based regression to a two-stage parameter estimation process. By first estimating subject-specific parameters and then using them to guide gaze estimation, the system transforms the problem into one that is more robust to noise and illumination variations while maintaining accuracy.
3Ease of operation
If single-image prediction methods are used, then the process is simple, but the method is sensitive to eye alignment, eyelid closing, and illumination perturbations
Solution Approach 1:
The method performs preliminary estimation of subject-specific eye parameters before conducting the actual gaze prediction. This preliminary action creates a personalized model for each user that compensates for individual variations in eye anatomy and appearance, making the subsequent gaze estimation more robust to perturbations like eyelid closing and illumination changes while maintaining operational simplicity.
Data Source
Figure 1~2
Figure 3~4b
Figure 5
AI summary
The invention concerns a method for estimating a gaze at which a user is looking at. The method comprises a step of retrieving an input image and a reference image of an eye of the user and/or an individual.The method comprises then a step of processing the input image and the reference image so as to estimate a gaze difference between the gaze of the eye within the input image and the gaze of the eye within the reference image. The gaze of the user is the retrieved using the estimated gaze difference and the known gaze of the reference image.The invention also concerns a system for enabling this method.