CamCheck (Izri Vision): Secure AI Vision with Anti-Spoofing and Smart Space Management

Abstract

This thesis describes the design, implementation, and experimental validation of Izri Vision Pro, an automated biometric access-control system built around a closed loop control architecture. A real-time video stream is captured via a network camera, faces are localized by a YOLOv8-Face detector, and each detected face passes through a multi-feature anti-spoofing gate (LBP variance, FFT Moiré response, specular highlight analysis) before a 512-dimensional identity embedding is extracted by an InsightFace (ArcFace) model running in ONNX Runtime. Access decisions rely on cosine-similarity matching against a local Faiss vector index, keeping all biometric data on-premises to satisfy privacy constraints. The prototype runs on a Dell Lat itude 7320 laptop (i7-1185G7, 16GB RAM, integrated Iris Xe GPU) paired with a POCO M4 Pro smartphone serving as a wireless NDI camera. Single-subject cali bration experiments show that the YOLOv8-Face and InsightFace pipeline maintains reliable recognition up to 4m, outperforming classical detectors (dlib CNN, MTCNN) which cap at 2m. A classroom-scale validation conducted at the University of Abou Bekr Belkaid Tlemcen, involving M2 Automatique et Informatique Industrielle stu dents, confirms that the system operates as a functional automated attendance tool, producingstructuredpresencereportswithoutmanual intervention.