Face Recognition
Face recognition is a biometric authentication technology that detects a person’s face from images or video captured by a camera and identifies the individual. It consists of four stages: face image collection/detection, preprocessing, feature extraction, and matching/verification.
Algorithm Overview
The face recognition system operates in the following flow:
- Face Detection: Detects face regions from images
- Preprocessing: Normalization of face images (tilt correction, brightness adjustment)
- Feature Extraction: Extraction of facial feature vectors using deep learning models
- Matching/Identification: Identity verification through similarity calculation between feature vectors
Performance Metrics
| Dataset | Recognition Accuracy |
|---|---|
| LFW | 99.80% |
| IJB-C (E4) | 97.12% |
Edge AI Board (RV1126B) Execution Efficiency
| Algorithm | Processing Time |
|---|---|
| face_detect (Face Detection) | 28ms |
| face_recognition (Feature Extraction + Matching) | 12.4ms |
Key Features
- High-precision recognition: LFW benchmark 99.80%, IJB-C 97.12%
- High-speed processing: Face detection + recognition in approximately 40ms (on edge AI board)
- Low-light support: Operates even at night when combined with a night vision camera
- Edge-complete: Local processing without the cloud, effective for privacy protection
Use Cases
- Entry/exit management and attendance systems
- Identity verification (eKYC)
- Suspicious person detection in surveillance cameras
- Resident identification in nursing care facilities
- Smartphone face unlock
- VIP detection at event venues
Edge AI Board Implementation
Leveraging the RV1126B NPU, the face detection (28ms) + feature extraction (12.4ms) pipeline is completed locally. Same-person determination is based on a similarity score (-1 to 1), with values of 0.4 or above indicating the same person.