Extreme Edge Computing for Secure and Private Multimodal Biometric Identification in Intelligent IoT Systems

Area of Science:

• Computer Science
• Electrical Engineering
• Cybersecurity

Background:

• The proliferation of Internet of Things (IoT) devices necessitates decentralized computing architectures to overcome cloud limitations.
• Edge computing offers solutions but faces security and privacy challenges in data transmission.
• The Compute Continuum paradigm extends computation to the extreme edge for enhanced efficiency.

Purpose of the Study:

• To develop a secure and privacy-preserving multimodal biometric authentication system for resource-constrained embedded devices.
• To leverage TinyML for pushing intelligence to the extreme edge within the Compute Continuum.
• To demonstrate the feasibility of standalone, privacy-by-design biometric sensors for intelligent IoT environments.

Main Methods:

• Implementation of a hierarchical processing chain on embedded devices.
• Utilizing an ultra-lightweight person-detection filter as a wake-up mechanism.
• Integration of robust facial and voice recognition modules for multimodal authentication.

Main Results:

• Achieved a combined False Acceptance Rate (FAR) of 0.12% through a strict hierarchical pipeline.
• Demonstrated exceptional energy efficiency, requiring only 0.15 J per inference cycle on an ESP32 microcontroller.
• Showcased autonomous operation for over 39 hours on a 600 mAh battery.

Conclusions:

• The proposed TinyML-based multimodal biometric system is viable for secure and private authentication in edge computing.
• The system's low power consumption and high accuracy validate its use in standalone, privacy-centric IoT applications.
• This approach effectively addresses security and privacy concerns in decentralized IoT ecosystems.