High Probability Group Database for Edge Facial Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Facial recognition systems face challenges in real-time processing at the edge of the Internet due to computational needs and connectivity latency, especially when dealing with large databases and varying angles of facial images, leading to reduced recognition accuracy and increased latency.
Innovation Solution
The implementation of a High Probability Group (HPG) database that stores facial images likely to appear in multi-stream videos, allowing for real-time recognition using local resources, and incorporating a facial network of related persons to enhance recognition accuracy by collecting images from multiple angles and reducing the need for extensive cloud connectivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large database of facial images is used for recognition, then recognition accuracy is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent segments the large facial database into multiple smaller databases organized by probability groups. The high probability group contains faces most likely to appear in current video streams, while lower probability groups contain other faces. This segmentation allows the system to search only relevant subsets rather than the entire database, reducing computational complexity while maintaining recognition accuracy for relevant faces.
Solution Approach 2:
The system performs preliminary actions by pre-calculating and organizing facial databases into probability groups based on historical appearance data before recognition is needed. This pre-organization allows the system to quickly identify and search only the high probability group during real-time recognition, avoiding the need to search the entire large database and significantly reducing processing time.
2Loss of time
If real-time facial recognition is performed at the edge of the Internet, then latency is reduced, but computational resources are limited
Solution Approach 1:
The patent segments the facial database into high probability and lower probability groups, enabling edge devices with limited computational resources to search only the smaller high probability group during real-time recognition. This segmentation reduces the computational burden on edge devices while maintaining fast recognition performance.
Solution Approach 2:
The system applies local quality by adapting the database search strategy to the specific context of each video stream. High probability groups are locally optimized for each stream based on historical data, allowing edge devices to perform efficient local searches without needing to process the entire global database, thus reducing latency and resource requirements.
3Measurement precision
If facial images from multiple angles are collected, then recognition accuracy for unseen angles is improved, but database size increases
Solution Approach 1:
The system performs preliminary action by pre-collecting and organizing facial images from multiple angles into the high probability group database before recognition is needed. This pre-collection ensures that when faces appear from unseen angles during real-time processing, the system can quickly find matching images from the pre-prepared multi-angle data without needing to search the entire large database.
Solution Approach 2:
The patent applies local quality by enhancing the high probability group with multi-angle facial images specifically tailored to each video stream's characteristics. This localized enhancement ensures that edge devices have access to diverse angle data relevant to their specific context without the overhead of storing all possible angle variations for all faces in the database.
4Measurement precision
If the high probability group database is updated frequently, then recognition accuracy is maintained, but processing overhead increases
Solution Approach 1:
The patent implements periodic action by updating the high probability group database at scheduled intervals rather than continuously. This periodic update strategy maintains recognition accuracy by ensuring the database reflects current appearance patterns while avoiding the excessive processing overhead of continuous updates. The system balances freshness of data with computational efficiency.
Data Source
AI summary
Techniques are provided for facial recognition using a high probability group database. One method comprises maintaining (i) a first database of facial images of individuals, and (ii) a second database of facial images comprising a subset of the individuals from the first database based on a probability of individuals appearing in sequences of image frames at a given time; applying a face detection algorithm to sequences of image frames to identify one or more faces in the sequences of images; and applying a facial recognition to at least one sequence of image frames using at least the second database to identify one or more individuals in the at least one sequence of image frames. The second database is comprised of facial images of: (i) individuals from multiple angles; (ii) individuals that appeared in prior image frames; and/or (iii) individuals that appeared in an image frame generated by a plurality of cameras.