On-Device Face Recognition Using Sound Source Direction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition technologies in mobile devices face challenges in achieving high accuracy and fault tolerance without network connectivity, as they often rely on cloud services that are computationally intensive and pose data security risks.
Innovation Solution
A face recognition method and apparatus that uses a sound receiving device and an image capturing device to detect and track faces, assigning image identifications and updating cluster identifications based on feature value comparisons, allowing for accurate face recognition within a single device without network dependency, thereby enhancing fault tolerance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If cloud services are used for face recognition, then recognition accuracy is improved, but data security risk increases and network dependency is created
Solution Approach 1:
The patent extracts the core face recognition functionality from cloud services and implements it locally on the electronic device. The processor performs face detection, tracking, and recognition operations using local computational resources, separating the critical recognition function from network-dependent cloud services while maintaining accuracy through sophisticated local algorithms including feature extraction and clustering techniques
2Measurement precision
If cloud services are used for face recognition, then recognition accuracy is improved, but network connectivity requirement increases
Solution Approach 1:
The electronic device performs face recognition operations autonomously using its own processor and local computational resources. The system captures images locally, processes them through face detection and tracking algorithms, performs feature extraction and clustering, and completes recognition without requiring external cloud service assistance, enabling operation in offline or restricted network environments
3Reliability
If continuous face tracking is performed, then fault tolerance rate is improved, but computational complexity increases
Solution Approach 1:
The system performs preliminary face detection and assigns temporary identifications to detected faces before full recognition processing. This preliminary tracking allows the system to maintain continuous monitoring of multiple faces, assign stable identifiers throughout the recognition process, and handle occlusions or temporary losses gracefully, improving fault tolerance while managing computational load through staged processing
4Measurement precision
If feature value comparison is performed for all cluster identifications, then recognition accuracy is improved, but processing time increases
Solution Approach 1:
The patent transforms face images into feature values (embeddings) that compactly represent facial characteristics in a reduced-dimensional space. This parameter transformation enables efficient comparison between detected faces and stored cluster identifications through mathematical operations on feature vectors rather than full image processing, maintaining high recognition accuracy while significantly reducing computational time and resource requirements
Data Source
AI summary
A face recognition method and apparatus are provided. The method is applicable to an electronic device including a sound receiving device and an image capturing device, and includes the following steps: detecting a direction of a sound source by using the sound receiving device, to capture an image in the direction by using the image capturing device; detecting and tracking a face in the image, to assign an image identification to the face according to a tracking result; determining whether the image identification exists in multiple cluster identifications recognized in a data set of face recognition; and in response to the image identification not existing in the data set, acquiring a feature value of the face and comparing the feature value of the face with a feature value of a face of each cluster identification to update a cluster identification to which the face belongs according to a comparison result.