Development of a Real-Time Lecturer Attendance and Student Check-in System
Từ khóa:
ArcFace, machine learning, digital transforma tion, dynamic QR code, Education, Face Recognition, fraud detection, QRCampus, Real-time Notifications, Real-time Communication, smart attendance system, ViT-Tiny, YOLOv8s, YOLOTóm tắt
Traditional attendance management in higher edu cation relies heavily on manual roll calls, consuming 6–10 minutes per session and remaining vulnerable to proxy attendance fraud, equivalent to a 10–15% [1] loss of instructional time. This paper presents QRCampus, an integrated smart attendance ecosystem that combines Dynamic Quick Response code verification with Artificial Intelligence-based computer vision to address these challenges in real-time educational environments. The system architecture employs a Node.js backend with PostgreSQL and Redis, leveraging WebSocket for real-time synchronization between a web dashboard and a React Native mobile application. A dual-verification mechanism– comprising Dynamic QR codes refreshing every 5 seconds, device identifier binding, and Global Positioning System geofencing– ensures physical presence validation at the correct time and location. For automated fraud detection, the system integrates the YOLOv8s object detection model, achieving a mean Average Precision (mAP@0.5) of 95.20% at 12
15ms inference speed, enabling cross-verification between actual headcounts and registered attendance records. Concurrently, a face recognition pipeline employing OpenCV Haar Cascade detection, heuristic accessory filtering, image normalization to 112×112 pixels, Vision Transformer Tiny with ArcFace loss ViT-Tiny ArcFace for 512-dimensional embedding extraction, and Cosine Similarity for identity verification enables reliable impersonation detection. Experimental evaluation demonstrates that QRCampus reduces attendance processing time by up to 90%, eliminates manual administrative errors, and effectively detects proxy attendance attempts. The main contributions are: The main contributions are: (i) a dual-verification architecture combining Dynamic QR codes, Device ID binding, and GPS geofencing for tamper-resistant presence validation; (ii) a custom trained YOLOv8s model on a 13,606-image classroom dataset,
achieving mAP@0.5 of 95.20% for real-time headcount cross verification; (iii) deployment of a pre-trained ViT-Tiny ArcFace
pipeline (5.5M parameters) attaining 99.67% on LFW, applied directly for impersonation detection without retraining; and (iv) a
full-stack real-time ecosystem unifying Node.js, PostgreSQL, Redis, WebSocket, and React Native on a single platform
