SmartVision: A Zero-GPU Multi-Student Live Biometric Attendance Engine Using ResNet-34 Deep Embeddings, WebRTC Real-Time Face Counting, Tri-Tier RBAC, and Proactive Retention Risk Analytics

Abstract Most classrooms still attend manual roll calls or paper sign-in rosters. The cost is easily underestimated: such routines can consume up to 15% of the lecture time, invite proxy attendance fraud, and accumulate administrative overhead. The computer vision systems built to remedy this come with their own burdens—prohibitive GPU dependencies, latency bottlenecks in scanning groups of faces, rigid operational roles for single users, and no insight into academic retention risk. SmartVision fills those gaps. It is a lightweight, enterprise-grade, zero-GPU biometric attendance portal based on a native WebRTC live camera streaming engine, continuous real-time face counting (Ndetected), and automatic student identity resolution (Srecognized). Under the hood, a ResNet-34 deep convolutional neural network (DCNN) aligns 68 facial landmarks via Dlib and projects each face into a 128- dimensional L2-normalized Euclidean metric space, re- ducing identity matching to a trivial distance comparison. Recognition runs on vectorized NumPy matrix operations with calibrated confidence thresholds (τ = 0.50), and the system resolves up to 60 students from a live feed or an unconstrained classroom snapshot in under 0.3 seconds per face - no GPU required. SmartVision also separates its three user roles through a tri-tier Role-Based Access Control architecture - Student, Teacher (scoped to Class/Subject assignments), and Administrator - and adds an automated 5-consecutive-day Zero-Attendance Retention Risk Algorithm that flags at-risk students early. We validated the system on 500 real-world classroom test images, where it achieved an aggregate 97.40% recogni- tion accuracy with sub-second processing latency, and it held up against basic static photo spoofing attempts. Index Terms Facial Recognition, Deep Embeddings, ResNet-34, We- bRTC Live Detection, Face Counting, Zero-GPU Accel- eration, Smart Attendance, Multi-Student Identification, RBAC, Retention Analytics.