Traditional attendance management systems in educational institutions are inefficient, time-consuming, and prone to errors, including proxy attendance and manual data entry mistakes. This research presents an innovative facial recognition-based attendance system that leverages YOLOv8 for face detection and Siamese networks for accurate student identification. The system utilises a comprehensive dataset of 70 students captured in real classroom environments to train deep learning models for robust face recognition under varying lighting conditions and facial orientations. The methodology integrates a Single Shot Scale-Aware Face Detector for multi-face identification, FaceNet-style embeddings for precise facial feature extraction, and a web-based platform supporting administrator, teacher, and student role-based access. The system addresses key limitations of existing approaches through dynamic venue mapping, dual-mode operation (institutional and standalone), continual learning capabilities, and a privacy-by-design architecture that stores only irreversible facial embeddings. Experimental validation conducted in authentic classroom settings with randomly positioned students achieved 98.5% detection accuracy and 81.9% recognition accuracy across 66 students, demonstrating superior performance compared to traditional attendance methods.