Smart Access: A Multi-Biometric System with Advanced Spoofing and Deepfake Detection for Secure Access Control

Résumé fourni par la source

Smart Access is a biometric access control system that leverages artificial intelligence to provide secure, contactless, and cost-effective authentication for physical environments such as offices, hotels, hospitals, and research facilities. Traditional access control systems that rely on keys, PINs, RFID cards, or dedicated biometric hardware often suffer from high costs, vulnerability to spoofing, and usability issues. Smart Access (SA) addresses these challenges by replacing physical tokens and sensors with advanced AI-powered multimodal biometric verification, using the user's smartphone as the interface. The system combines facial recognition, voice authentication, and deepfake detection to verify user identity. Facial authentication is implemented using the Deep Face framework with a VGG-Face model, enhanced by liveness detection to prevent spoofing through images or pre-recorded videos. Voice authentication includes three distinct layers: speaker verification using Speech Brain, transcript verification using Whisper ASR, and deepfake detection via a fine-tuned Wav2Vec2 model. This multi-layered approach effectively counters replay attacks and synthetic voice manipulation. The system architecture includes a mobile/web application, a secure backend for AI processing, and an ESP32 microcontroller that controls physical access points. Once authentication is successful, a secure command is issued to the ESP32 device to unlock the door. The system also includes a comprehensive admin dashboard for managing users, rooms, permissions, and access logs, supporting multi-tenant configurations to isolate organizational data. Evaluation results demonstrate a true acceptance rate of 97.4% for facial recognition and 94.6% detection accuracy for deepfake audio, with an average end-toend authentication time of 2.4 seconds. A user study with 30 participants showed that over 90% found the system intuitive and more secure than traditional methods.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.