Eyelid Detection via Hierarchical Face Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eyelid detection technologies face challenges in accurately detecting eyelid positions due to influences from ambient light, glasses frames, and require extensive processing operations, leading to potential errors and high computational loads.
Innovation Solution
An eyelid detection apparatus that captures face images at regular intervals, extracts edge lines based on gray levels, identifies candidate edge line pairs for upper and lower eyelids, and detects the centerline of the face contour to accurately identify eyelid positions with reduced processing operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If the face detection apparatus detects a narrow region such as the area between the eyes in a wide face region, then the detection coverage is improved, but the processing load increases significantly
Solution Approach 1:
The patent divides the face detection process into multiple stages: first detecting the entire face region, then identifying the area between the eyes as a sub-region, and finally detecting glasses within that sub-region. This hierarchical segmentation allows the system to maintain comprehensive detection coverage while reducing the computational burden by focusing intensive processing only on relevant sub-regions rather than the entire face area.
2Difficulty of detecting and measuring
If the apparatus judges whether a subject wears glasses based on gray level histogram or edge information of sub regions, then the detection capability is improved, but the determination accuracy deteriorates due to influences of glass frame design
Solution Approach 1:
The patent introduces the area between the eyes as an intermediary region that mediates between the face detection and glasses detection. By first identifying this intermediate sub-region and then performing glasses detection within it, the system improves determination accuracy by isolating the glasses detection from confounding factors in the rest of the face, while maintaining detection capability through the hierarchical approach.
3Measurement precision
If the open and closed detection apparatus approximates the eyelid by a multidimensional curve and calculates the curvature, then the eyelid state detection capability is improved, but the processing operations become enormous
Solution Approach 1:
The patent applies partial action by focusing the curvature calculation only on the detected eyelid region rather than the entire face image. After identifying the eyelid's position and shape through preliminary detection, the system performs the computationally intensive curvature approximation only on this localized region, thereby maintaining detection accuracy while significantly reducing the overall processing complexity compared to analyzing the entire image.
4Measurement precision
If the apparatus detects the upper eyelid in the face image, then the eyelid position detection is improved, but error detection occurs due to influence of ambient light and glasses
Solution Approach 1:
The patent extracts the eyelid detection from the general face detection process by first identifying the eye region as a separate sub-region, then performing eyelid detection specifically within this extracted region. This extraction approach isolates the eyelid detection from harmful factors such as ambient light variations and glasses frames that affect the broader face region, thereby improving detection accuracy by focusing only on the relevant area.
Data Source
AI summary
An eyelid detection apparatus includes a face image storing means storing a plurality of face images of a subject captured at different timings, a change detecting means detecting changes in each face image caused by blinking after processing the plurality of face images stored by the face image storing means, and an eyelid detecting means detecting a position of an eyelid of the subject based on the changes detected by the change detecting means.