Redundant Eye Tracking System Using Dual Feature Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eye tracking systems face inaccuracies due to software or hardware issues, and machine learning-based methods require extensive training data, limiting their performance across various scenarios.
Innovation Solution
A redundant eye tracking system that runs multiple independent eye tracking procedures on the same input image set, determining eye positions only when both procedures yield similar results within a threshold value, using different feature sets such as glint positions and head pose/pupil positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single eye tracking procedure is used, then the system is simple, but the accuracy and reliability are reduced due to software or hardware issues
Solution Approach 1:
The patent combines multiple independent eye tracking procedures (first and second procedures using different feature sets) into a single integrated system. The results from both procedures are merged and compared to determine the final eye position, thereby improving reliability through redundancy while managing complexity through a unified decision-making framework.
Solution Approach 2:
The system prepares for potential failures by having backup eye tracking procedures ready. When the first procedure fails or produces unreliable results, the second procedure serves as a pre-prepared cushion to maintain reliable eye position determination, preventing system failure before it occurs.
2Adaptability or versatility
If machine learning-based eye tracking is used, then the system can handle various scenarios, but extensive training data is required which takes time and resources
Solution Approach 1:
The patent segments the eye tracking task into multiple independent procedures, each using different feature sets (e.g., pupil center corneal reflection, head pose, iris detection). This segmentation allows each procedure to be simpler and require less training data, while collectively they provide versatility across different scenarios through diversity of approaches.
Solution Approach 2:
The system implements multiple eye tracking procedures that can each handle different types of scenarios independently. By making the system multi-functional with several procedures that use different features and algorithms, it achieves adaptability across various eye conditions and scenarios without requiring any single procedure to be trained on all possible cases.
3Loss of time
If training data is limited to certain scenarios, then collection time is reduced, but the algorithm performs poorly on unseen scenarios
Solution Approach 1:
The patent divides the eye tracking problem into multiple procedural segments, each specialized for different feature types and scenarios. This allows training data to be segmented and collected separately for each procedure, reducing the time burden while ensuring each procedure is well-trained for its specific scenario type, collectively covering broader adaptability.
Data Source
AI summary
There is provided mechanisms for eye position determination of a subject depicted in an image set. A method comprises obtaining first information indicating a first set of eye positions of the subject by applying a first eye tracking procedure on an image set depicting the subject. The first eye tracking procedure uses a first set of features extracted from the image set for obtaining the first set of eye positions. The method comprises obtaining second information indicating a second set of eye positions of the subject by applying a second eye tracking procedure on the image set depicting the subject. The second eye tracking procedure uses a second set of features extracted from the image set for obtaining the second set of eye positions. The method comprises determining the eye position of the subject in the image set based on the first information and the second information only when the first set of eye positions, as indicated by the first information, and the second set of eye positions, as indicated by the second information, differ from each other less than a threshold value.