Eye Detection via Projection Filtering and Temporal Consistency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Eye detection and tracking algorithms perform poorly in real-world conditions with varying lighting and occlusions, such as shadows, eyeglasses, sunglasses, or makeup, leading to incorrect grouping of pixel clusters and increased processing complexity.
Innovation Solution
The method employs horizontal and vertical projection filtering combined with appearance-based testing, spatial-geometric, and anthropomorphic considerations to identify and confirm eye candidates, using a digital signal processor to analyze video images and track eye locations across frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If appearance-based testing is used to identify eye candidates, then detection accuracy under controlled conditions is improved, but reliability deteriorates under real-world conditions with lighting variations and occlusions
Solution Approach 1:
The detection process is divided into multiple independent filtering stages: projection filtering to identify candidate regions, appearance-based testing to evaluate visual characteristics, spatial-geometric filtering to verify anatomical plausibility, and temporal consistency checking to confirm stability across frames. Each stage processes specific aspects of the data independently, allowing the system to maintain high accuracy while improving reliability through cumulative verification.
Solution Approach 2:
The system dynamically adjusts detection parameters based on lighting conditions and occlusion detection. When shadows or occlusions are detected, the appearance-based testing thresholds are modified to accommodate variations in eye appearance. The projection filtering parameters are adjusted based on the detected facial region characteristics, enabling the system to maintain both accuracy and reliability across diverse real-world conditions.
2Reliability
If multiple filtering techniques are applied to improve detection reliability, then eye detection reliability is improved, but processing throughput deteriorates
Solution Approach 1:
Projection filtering is applied as a preliminary stage before appearance-based testing to pre-identify and filter candidate regions. This preliminary action reduces the number of pixels and regions that require more computationally intensive appearance analysis, thereby maintaining high reliability through comprehensive filtering while improving processing throughput by avoiding unnecessary complex computations on all image data.
Solution Approach 2:
The system applies filtering techniques selectively rather than uniformly across the entire image. Appearance-based testing is performed only on candidate regions identified by projection filtering, and spatial-geometric filtering is applied only to clustered candidates. This partial application of filtering actions maintains detection reliability for actual eye regions while significantly improving processing throughput by avoiding excessive computations on non-eye areas.
3Reliability
If complex detection and tracking routines are used to handle occlusion and lighting variations, then detection reliability is improved, but device complexity increases
Solution Approach 1:
The complex detection routine is segmented into distinct functional modules: projection filtering module, appearance-based testing module, spatial-geometric filtering module, and temporal consistency module. Each module handles a specific aspect of the detection problem independently, making the overall system more manageable and easier to implement while maintaining high reliability through the coordinated operation of these segmented functions.
Solution Approach 2:
The projection filtering technique serves multiple functions simultaneously: it identifies candidate eye regions, filters out non-eye pixels, and provides spatial information for subsequent clustering. The appearance-based testing performs both classification and validation functions. This multi-functionality reduces the need for separate dedicated algorithms for each task, thereby improving reliability while controlling device complexity.
Data Source
AI summary
A human eye is detected in a video image by identifying potential eye candidates using horizontal and vertical projection filtering on a specified portion of the image, coupled with rudimentary appearance-based testing. The identified eye candidates are clustered and winning eye candidates are selected based on appearance, spatial-geometric and anthropomorphic considerations. The winning eye candidates are subjected to additional appearance-based testing, and high confidence eye candidates are accepted if they favorably compare with identified eye candidates from previous video image frames.