Image Analyzer Mask Detection Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques, such as template matching, struggle to accurately detect subjects wearing masks due to limitations in training machine learning systems for all face patterns, leading to potential erroneous detection.
Innovation Solution
An image analyzer is developed with multiple detectors and a determination unit that evaluates detection results from both face-without-mask and face-with-mask detectors, using past determination results to improve accuracy and reliability, and reduces computational complexity by narrowing the search area based on coarse candidate detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If separate machine learning systems are used for subjects with and without masks, then detection capability for different states is improved, but training completeness for all face patterns deteriorates leading to erroneous detection
Solution Approach 1:
The detection system is segmented into multiple specialized detectors: a first detector for subjects without masks and a second detector for subjects with masks. Each detector is trained on specific face patterns, allowing them to specialize in their respective domains while the overall system maintains comprehensive coverage through the combination of these specialized components.
2Reliability
If multiple detectors are used to improve detection accuracy, then reliability of mask determination is improved, but device complexity increases
Solution Approach 1:
Multiple detectors (first detector for maskless subjects and second detector for masked subjects) are merged into a unified determination unit that integrates their outputs. The determination unit combines the detection results from both detectors and uses past determination results to evaluate reliability, achieving accurate mask determination while managing system complexity through coordinated integration.
Solution Approach 2:
The determination unit utilizes past determination results as feedback to evaluate the reliability of current detection results. This feedback mechanism allows the system to learn from previous detections and adjust its assessment, improving accuracy while providing a structured approach to managing complexity through iterative refinement.
3Reliability
If comprehensive search for subject areas is performed, then detection coverage is improved, but computational load increases
Solution Approach 1:
A coarse detector performs preliminary detection to identify candidate areas before the more computationally intensive first and second detectors are activated. This preliminary action filters out unlikely regions, allowing the specialized detectors to focus only on promising areas, thereby reducing overall computational load while maintaining comprehensive coverage.
Solution Approach 2:
The search process is segmented into multiple stages: coarse detection to identify candidate areas, followed by selective fine detection using the first and second detectors only in relevant regions. This segmentation of the search process enables comprehensive coverage while significantly reducing computational requirements by avoiding unnecessary processing in all image regions.
Data Source
AI summary
An image analyzer repeatedly obtains a captured image including a subject to be detected, and detects, in the captured image, a first area likely to include the subject without a cover, detects, in the captured image, a second area likely to include the subject with a cover, and determines whether the subject is covered by the cover based on a detection result of the first area and a detection result of the second area and a determination result for the captured image obtained previously.