Red-eye Determination Using Black Eye Probability Density
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In low-light conditions, it is challenging to accurately detect the red eye of a vehicle occupant due to the presence of multiple red-eye candidates, including reflections from LED light, which can lead to errors in detection.
Innovation Solution
A red-eye determination device that learns the probability density distribution of black eye positions from image data collected during driving and uses this information to differentiate between true red-eye candidates and LED reflections by analyzing their behavior and position relative to the face direction changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If red-eye detection is performed using LED illumination in low-light conditions, then the ability to detect red eye is improved, but the detection accuracy deteriorates due to false candidates from LED reflections
Solution Approach 1:
The patent segments the red-eye detection process into multiple stages: first detecting red-eye candidates using LED illumination, then filtering these candidates by analyzing black eye position probability density distributions and face direction changes. This segmentation allows the system to benefit from LED illumination for detection while using additional criteria to eliminate false candidates.
Solution Approach 2:
The patent introduces an intermediary approach by using black eye position probability density distribution as a mediator between the red-eye candidate detection and the final red-eye determination. This intermediary model helps distinguish true red-eye candidates from false candidates caused by LED reflections.
2Adaptability or versatility
If multiple red-eye candidates are detected from image information, then the coverage of detection is improved, but the reliability of red-eye determination deteriorates due to inability to distinguish true red eye from false candidates
Solution Approach 1:
The patent implements feedback by continuously monitoring face direction changes and using this information to validate red-eye candidates. The system compares the position of detected red-eye candidates against the expected movement pattern based on face direction, providing feedback to confirm or reject candidates.
Solution Approach 2:
The patent introduces dynamic analysis by examining how red-eye candidates behave relative to face direction changes over time. True red-eye candidates should move consistently with face direction changes, while false candidates from LED reflections remain stationary or move differently, allowing the system to dynamically filter candidates.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables accurate identification of the red eye even when multiple candidates are present, improving detection reliability by leveraging learned black eye position probabilities and face direction changes.
Implementation Method 1
The LED light is reflected from the eye or eyelid of the driver and is inserted into image information
Data Source
AI summary
A black eye position existence probability density distribution learning unit that records a black eye position which is detected in the daytime to a black eye position existence probability density distribution, a red-eye candidate detection unit that detects red-eye candidates from the image of the driver at night, and a red-eye determination unit that determines the red eye from the red-eye candidates. The red-eye determination unit determines the red eye on the basis of the relationship between a change in the direction of the face and the behavior of the red-eye candidate and determines, as the red eye, the red-eye candidate disposed at the position of high black eye position existence probability density with reference to the black eye position existence probability density distribution.