Efficient Face Recognition System for Operating in Unconstrained Environments.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Facial recognition is crucial for real-time security systems.
• Deep neural networks have advanced facial recognition performance.
• Need for efficient, low-cost facial recognition systems in unconstrained environments.

Purpose of the Study:

• To propose an efficient and low-cost real-time facial recognition system.
• To combine deep learning (FaceNet) with traditional classifiers (SVM, KNN, RF).
• To operate effectively in unconstrained environments with moderate hardware.

Main Methods:

• Utilized YOLO-Face (based on YOLOv3) for high-speed, real-time face detection.
• Employed FaceNet for feature extraction in the recognition stage.
• Integrated FaceNet with Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF) classifiers.

Main Results:

• YOLO-Face demonstrated superior performance with partial occlusions, pose variations, and small faces.
• The face detector achieved over 89.6% accuracy on the Honda/UCSD dataset at 26 FPS.
• FaceNet+SVM achieved 99.7% accuracy on the LFW dataset, outperforming other combinations.
• The complete system achieved 99.1% recognition accuracy with a 49 ms runtime.

Conclusions:

• The proposed system offers a robust and accurate solution for real-time facial recognition.
• The combination of YOLO-Face and FaceNet+SVM is highly effective for unconstrained environments.
• The system provides a practical and efficient approach for security applications.