Optimization of Neural Network Models of Computer Vision for Biometric Identification on Edge IoT Devices.

Area of Science:

• Computer Science
• Biometrics
• Internet of Things (IoT)
• Artificial Intelligence (AI)

Background:

• Biometric systems are crucial for secure identification.
• Integrating IoT and AI offers advanced capabilities for intelligent systems.
• Existing systems may face limitations in spoofing resistance and deployment efficiency.

Purpose of the Study:

• To develop an intelligent biometric system leveraging IoT and AI.
• To explore personal identification using facial images and palm vein patterns.
• To optimize the system for edge IoT deployment and enhance security.

Main Methods:

• Analysis of state-of-the-art computer vision and neural network architectures.
• Development of independent approaches for facial and palm vein feature extraction and comparison.
• Implementation of ResNet-50 backbone with Triplet Loss for metric learning and optimization.
• Deployment of the system on edge IoT devices using Dockerized FastAPI with JWT.

Main Results:

• Experimental results demonstrating effective feature extraction and comparison for both biometric modalities.
• Successful optimization for edge deployment, improving convergence and generalization.
• Demonstration of a functional web interface and server infrastructure.
• Validation of the system's potential for high reliability and spoofing resistance.

Conclusions:

• The proposed intelligent biometric system effectively integrates IoT and AI for robust personal identification.
• The system demonstrates practical optimization for edge devices, enhancing usability and efficiency.
• Facial and palm vein biometrics can be synergistically employed for advanced security applications.
• The solution holds significant potential for access control, smart buildings, and educational institutions.