Eyeball Localization Using Visible and Obscured Facial Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Human facial recognition systems fail to achieve precise eyeball locating when users wear masks, obscuring facial features.
Innovation Solution
An image processing system and method that detects unobscured facial features, estimates obscured features, and locates the eyeball position using a combination of facial feature detection, restoration, and estimation techniques, including machine learning and simulation models to handle masked faces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional facial recognition is used, then it works well for unobscured faces, but it fails when facial features are obscured by masks
Solution Approach 1:
The patent segments the facial recognition task into two parts: detecting unobscured features (eye regions) and estimating obscured features (facial landmarks below the mask). This allows the system to process available information and infer missing information separately, improving reliability under mask obstruction.
Solution Approach 2:
The patent introduces an intermediary estimation process that uses detected unobscured features as input to predict obscured features. This intermediary step bridges the gap between visible and hidden facial regions, enabling the system to adapt to mask obstruction while maintaining recognition accuracy.
2Device complexity
If only unobscured features are used for recognition, then the system is simpler, but eyeball localization precision deteriorates
Solution Approach 1:
The patent performs preliminary detection of unobscured features first, then uses these detected features as a basis for estimating obscured features. This preliminary action provides a foundation that improves subsequent eyeball localization precision without requiring complete facial visibility.
Solution Approach 2:
The patent transitions from using only directly detected features (2D detection) to incorporating estimated features that extend the solution into another dimension of inference. This allows the system to achieve higher precision by combining detection and estimation in a multi-dimensional approach.
Data Source
AI summary
An eyeball locating method, an image processing device, and an image processing system are proposed. The method includes the following steps. A human facial image of a user is obtained, wherein the human facial image includes an unobscured human facial region and an obscured human facial region, and the unobscured human facial region includes an eye region. At least one unobscured human facial feature is detected from the unobscured human facial region, and at least one obscured human facial feature is estimated from the obscured human facial region. Next, an eyeball position is located according to the unobscured human facial feature and the obscured human facial feature.