Iris Ellipse Gaze Estimation Without Head Pose Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current gaze direction estimation systems are unreliable due to errors in deriving user information, such as camera-head relationships and user dependency, leading to incorrect determinations.
Innovation Solution
A method using a convolutional neural network to detect the iris ellipse and determine two possible gaze directions based on geometric information, eliminating the need for head pose estimation and user-specific calibration, and providing probabilities for each direction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current gaze direction estimation systems use prior user data and head pose information, then they can provide personalized accuracy, but they become vulnerable to errors from missing or erroneous data and require complex calibration
Solution Approach 1:
The patent extracts and eliminates the dependency on error-prone prior user data, head pose information, and complex calibration procedures. By using only the current image and iris ellipse geometry, the system removes multiple potential failure points while maintaining accuracy through direct geometric computation of the gaze vector from the ellipse parameters.
Solution Approach 2:
The system performs self-calibration by automatically adapting to each user's eye geometry in real-time without requiring prior calibration data. The neural network learns user-specific characteristics on-the-fly from the current image, eliminating the need for separate calibration sessions or stored user profiles.
2Measurement precision
If conventional systems require multiple prior images and user calibration, then they can establish accurate baseline data, but they increase processing time and system complexity
Solution Approach 1:
The patent performs preliminary encoding of the gaze direction computation within the neural network architecture itself. By training the network to directly predict the gaze vector from the iris ellipse parameters, the system prepares the computational pathway in advance, enabling instant gaze estimation without requiring sequential processing of multiple images or calibration routines.
Solution Approach 2:
The patent merges multiple functions into a single neural network operation: iris detection, ellipse parameter extraction, and gaze direction computation are all performed in one integrated process. This consolidation eliminates the need for separate processing stages and reduces overall computation time while maintaining accuracy.
3Adaptability or versatility
If systems use head trackers and user authentication modules, then they can provide contextual information, but they introduce multiple error sources and reduce system robustness
Solution Approach 1:
The patent extracts and removes the dependency on external modules such as head trackers and user authentication systems. By computing gaze direction solely from iris ellipse geometry in the current image, the system eliminates the error propagation chains that exist when multiple interconnected modules are involved, while still providing the necessary contextual information for accurate gaze estimation.
4Measurement precision
If current systems require knowledge of camera-head relationships and user calibration, then they can achieve accurate gaze estimation, but they increase device complexity and reduce ease of operation
Solution Approach 1:
The system performs self-calibration by automatically adapting to each user's eye geometry and camera-relative positioning in real-time. The neural network learns the user-specific parameters from the current image without requiring manual calibration procedures or pre-stored camera-head relationship data, making the system equally simple to use for all users while maintaining high accuracy.
Data Source
AI summary
Disclosed is a method for determining a gaze direction of a user (5) by employing a neural network (220) operating on the basis of image data of an image of a face of the user (5). The neural network (220) detects an ellipse (20) representing an outer border of an iris (17) of the user's eye (15) and outputs a first vector (31), a second vector (32) and a respective first and second probability (p1, p2) representing the probability that the first vector (31) or the second vector (32) is the gaze direction Further disclosed are an apparatus configured to perform such method and a vehicle comprising such apparatus.