AI-Based CCTV Attendance Systems in Educational Institutions: AComputer Vision and Deep Learning Case Study on Automated Monitoringfor Classroom Engagement

Manual, name-based attendance procedures remain common in higher
education despite being time-consuming, susceptible to proxy attendance, anddisruptive to instructional time. This paper presents a computer vision and deeplearning case study of an artificial intelligence (AI)-based closed-circuit
television (CCTV) attendance system built on a FastAPI backend, OpenCV's
YuNet face detector and SFace face recognizer, a lightweight SQLite database, and a browser-based analytics dashboard. Rather than evaluating the systemthrough operational metrics alone, this study centers on a technical computer
vision analysis of the underlying recognition pipeline: model convergenceduring training, detector and recognizer discrimination performance, similarity- score separation and threshold selection, processing latency by pipeline stage, comparative accuracy against earlier generations of face detection andrecognition algorithms, accuracy under varying capture conditions, andclassification error structure. Results show stable training and validationconvergence, strong discrimination ability as measured by receiver operatingcharacteristic (ROC) and precision–recall analysis, clearly separated genuineand impostor similarity-score distributions supporting the selected operatingthreshold, a clear accuracy advantage for the deployed deep convolutional
pipeline over classical Haar cascade and HOG plus support vector machine(SVM) baselines, network upload rather than model inference as the dominant
source of end-to-end latency, and a confusion matrix indicating a modest false- negative rate consistent with occlusion and pose variation in real classroomfootage. These findings position deep learning-based face detection andrecognition as a substantially more capable foundation for classroomattendanceautomation than earlier handcrafted-feature approaches, while highlightingresidual error sources, deployment trade-offs, and ethical safeguards that merit
continued attention as such systems are adopted more broadly.