Face Recognition Using Real-Time Registration Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition accuracy in video streams is lower than in static images due to changes in facial appearance over time and outdated registered images in databases, leading to decreased recognition rates.
Innovation Solution
A face recognition method using an electronic device with a processor and memory that compares face information from video streams against pre-registration and real-time registration information, updating the latter when a similarity threshold is met, and using a higher similarity threshold for more accurate recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pre-registration information from databases is used for face recognition, then recognition can be performed, but the recognition accuracy decreases over time as registered images become outdated
Solution Approach 1:
The system performs preliminary face recognition using pre-registration information before the database becomes outdated, and prepares real-time registration information in advance to replace it. This preliminary action ensures continuous accurate recognition without waiting for database updates.
Solution Approach 2:
The system establishes a feedback mechanism where recognition results are continuously monitored and used to update registration information. When recognition accuracy drops below a threshold, the system automatically triggers database updates or switches to real-time registration, creating a closed-loop feedback system that maintains accuracy.
2Productivity
If a lower similarity threshold is used to increase recognition success rate, then more faces are recognized, but recognition accuracy decreases
Solution Approach 1:
The recognition process is segmented into multiple stages with different threshold requirements. First, a lower threshold is used for initial matching to ensure high success rate, then a higher threshold is applied for verification to ensure accuracy. This segmentation allows both productivity and precision to be optimized at different stages.
Solution Approach 2:
The system performs partial verification by using a two-threshold approach: a lenient first threshold for broad matching and a strict second threshold for final confirmation. This partial application of stringent criteria only where necessary maintains high success rates while preserving accuracy for final decisions.
3Reliability
If multiple image frames are processed to improve recognition accuracy, then recognition reliability increases, but processing time increases
Solution Approach 1:
Instead of continuously processing all video frames, the system selectively processes periodic or interval frames for recognition. This periodic sampling reduces the total number of frames processed while still maintaining sufficient temporal coverage to capture facial variations, thereby reducing processing time while preserving recognition reliability.
Solution Approach 2:
The system performs preliminary filtering and preprocessing on video frames before full recognition processing. Frames are pre-screened based on quality metrics, motion detection, or relevance criteria, so that only suitable frames undergo comprehensive multi-frame analysis. This preliminary action reduces the effective number of frames processed while maintaining recognition accuracy.
Data Source
AI summary
The application provides a face recognition method and an electronic device using the method. The method includes: obtaining face information from an image frame in a video stream; determining whether a first similarity between pre-registration information and the face information is higher than a first similarity threshold; determining that face recognition is successful if the first similarity is higher than the first similarity threshold, and updating real-time registration information with the face information; and determining that face recognition fails if the first similarity is lower than the first similarity threshold, and then determining whether a second similarity between the real-time registration information and the face information is higher than a second similarity threshold, where the second similarity threshold is higher than the first similarity threshold.