Eyelid Detection Using Geometric Constraints for Red-Eye Interference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eyelid detection systems face challenges in accurately determining whether a driver's eyes are closed, especially when the red-eye effect occurs or when the subject is wearing eye shadow, as these conditions lead to uniform pixel luma and obscure the edges of the eyelids, making precise detection impossible.
Innovation Solution
An eyelid detection device and method that calculates an evaluation value using a fitting curve defined by the inner and outer corners of the eyes, detects pixel groups affected by the red-eye effect, and sets reference positions for the eyelids based on characteristic curves, allowing for precise detection even in the presence of the red-eye effect or eye shadow.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional edge detection methods are used to detect eyelid edges, then the detection process is simple, but the detection accuracy deteriorates when red-eye effect or eye shadow is present
Solution Approach 1:
The patent segments the eyelid detection process into multiple stages: initial edge detection, candidate edge identification, verification through geometric constraints (distance relationships between inner/outer corners and eyelid edges), and final selection. This multi-stage segmentation allows the system to achieve high accuracy by systematically eliminating false edges caused by red-eye effect or eye shadow, while keeping each individual stage computationally manageable
Solution Approach 2:
The patent performs preliminary actions by first identifying candidate edges and then verifying them against geometric constraints before final detection. The system pre-calculates expected distance relationships between anatomical landmarks (inner corner, outer corner, and eyelid edges) and uses these as verification criteria. This preliminary verification step filters out false edges from red-eye effect or eye shadow before they compromise the final detection accuracy
2Measurement precision
If the red-eye effect or eye shadow is present, then the imaging process is unaffected, but the edge detection accuracy deteriorates due to uniform pixel luma
Solution Approach 1:
The patent converts the harmful uniform pixel luma regions (caused by red-eye effect or eye shadow) into beneficial verification targets. By calculating expected geometric relationships and using them to verify candidate edges, the system identifies that true eyelid edges will consistently satisfy these geometric constraints, while false edges from red-eye effect or eye shadow will not. This transforms the previously harmful uniform regions into opportunities for verification and false edge rejection
Solution Approach 2:
The patent introduces geometric constraint verification as an intermediary step between edge detection and final eyelid edge identification. This intermediary verification process uses pre-calculated distance relationships between anatomical landmarks as a mediator to filter out false edges. The geometric constraints act as a bridge that connects raw edge detection results with accurate eyelid edge identification, eliminating the harmful effects of uniform pixel luma regions
3Measurement precision
If multiple edges are detected near the eyelid due to red-eye effect, then the edge detection process finds more candidates, but the correct eyelid edge identification becomes difficult
Solution Approach 1:
The patent implements feedback by using geometric constraint verification to evaluate each candidate edge and provide feedback on its validity. The system calculates expected distance relationships between anatomical landmarks, compares actual measurements to these expectations, and uses this feedback to accept or reject candidate edges. This feedback loop systematically eliminates false edges from red-eye effect while preserving true eyelid edges, enabling accurate identification even when multiple candidate edges are present
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A secondary curve, the ends of which coincide with the inner corner and the outer corner of the eye, is determined successively, and the total of the edge values of the pixels overlapping the secondary curve is calculated as an evaluation value (S207). Next, a characteristic curve is generated on the basis of data made up of the calculated evaluation value and the Y-coordinate of the intersection between the secondary curve and a straight line passing through the center of a line segment whose ends coincide with the inner corner and the outer corner of the eye (S208). Then, the reference positions for the upper eyelid and the lower eyelid of the eye are set on the basis of the result of an attempt to detect a pixel group occurring because of the red-eye effect in a search area defined on the basis of peaks in the characteristic curve (S209). Accordingly, when an image is captured with an imaging device, it is possible to precisely set reference positions for the eyelids, even when the so-called red-eye effect or the like occurs and edges other than the edges of the eyelids are produced near the edges of the eyelids.