A Comparative Study of Computer Vision-Based Attendance Recognition Systems

Manual attendance management is time-consuming, error-prone and open to proxy attendance. This work studies automated computer vision based student attendance and classroom engagement monitoring systems for real-time classroom environments. It reviews seventeen recent papers (2025–2026) and builds on a base paper that uses multi-face recognition with FaceNet-based 128-dimensional embeddings. The proposed methodology integrates image quality assessment and CLAHE enhancement to improve facial image quality under varying illumination, followed by YOLOv5 for multi-face detection. Dlib 68-point facial landmarks are used for face alignment, while MobileFaceNet generates lightweight facial embeddings for student identification. Liveness detection using blink and head movement reduces spoofing, and quality-aware temporal verification improves recognition reliability across consecutive frames. In parallel, classroom engagement is analysed using eye/gaze analysis and CNN-based emotion classification to estimate students' attention and emotional states. A duplicate-check mechanism ensures that each student is marked only once per session, attendance and engagement information is stored with student and time details, and automated notifications can be sent to students. The report describes the existing methodologies, the architecture, the eight algorithms with flowcharts and worked examples, the datasets and the performance metrics used in the literature. The framework is to be evaluated using accuracy, precision, recall, F1-score, recognition latency, liveness performance and engagement classification performance under different classroom conditions. The hybrid framework aims to provide accurate, contactless, real-time and scalable attendance and classroom engagement monitoring.