Adaptive Eye Tracking Using Attribute-Based Tracker Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current eye tracking methods face challenges in accurately tracking eyes in images with varying light sources, glasses, sunglasses, and reflections, leading to limited tracking performance when using a single eye tracker for diverse input attributes.
Innovation Solution
An adaptive eye tracking method that detects the attribute of the eye area using an attribute classifier and selects the appropriate eye tracker from a plurality of specialized trackers based on the determined attributes, including the presence of glasses, sunglasses, and reflections, to improve tracking accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single eye tracker is used for diverse input attributes, then the device complexity is reduced, but the measurement precision deteriorates due to limited tracking performance across varying conditions
Solution Approach 1:
The patent divides the eye tracking system into multiple specialized eye trackers, each designed to handle specific input attributes (e.g., images with glasses, sunglasses, reflections, or different lighting conditions). This segmentation allows each tracker to be optimized for its specific condition, thereby improving measurement precision without requiring a single complex universal tracker.
Solution Approach 2:
The system changes the parameter of tracker selection based on the detected attributes of the input image. By analyzing image attributes (such as presence of glasses, lighting conditions, reflections) and dynamically selecting the appropriate eye tracker based on these parameters, the system achieves high precision across diverse conditions while maintaining manageable overall complexity through modular architecture.
2Measurement precision
If multiple specialized eye trackers are used for different attributes, then the measurement precision is improved, but the device complexity increases
Solution Approach 1:
The system performs preliminary attribute detection and analysis on the input image before selecting an eye tracker. By pre-classifying the image attributes (such as detecting presence of glasses, sunglasses, reflections, or lighting conditions) and preparing the appropriate tracker in advance, the system achieves high precision without requiring complex real-time switching or configuration during the actual tracking process.
Solution Approach 2:
The patent introduces an attribute detection module as an intermediary between the input image and the multiple eye trackers. This intermediary analyzes the image attributes and mediates the selection process, determining which specialized tracker should process the given input. This intermediary layer simplifies the overall system architecture by centralizing the decision-making logic and preventing direct complex interactions between multiple trackers.
3Measurement precision
If attribute detection and tracker selection are performed, then the measurement precision is improved, but the productivity decreases due to additional processing steps
Solution Approach 1:
The system performs partial attribute detection, focusing only on the most critical attributes that significantly impact tracking performance (such as presence of glasses or extreme lighting conditions). By detecting only the essential attributes rather than analyzing all possible image characteristics, the system achieves sufficient precision improvement while minimizing the additional processing time required for attribute detection and tracker selection.
Data Source
AI summary
Provided is a method and apparatus for eye tracking. An eye tracking method includes detecting an eye area corresponding to an eye of a user in a first frame of an image; determining an attribute of the eye area; selecting an eye tracker from a plurality of different eye trackers, the eye tracker corresponding to the determined attribute of the eye area; and tracking the eye of the user in a second frame of the image based on the selected eye tracker, the second frame being subsequent to the first frame.