FACE RECOGNITION SYSTEM

Attendance management remains a significant challenge in educational institutions due to proxy entries, human errors, and the absence of real-time monitoring. Conventional roll-call methods are time-consuming and unreliable, particularly in large classrooms. This paper presents a Hybrid Edge-Cloud Face Recognition-Based Smart Attendance Monitoring System implemented as a prototype and evaluated in a simulated deployment environment. The proposed framework integrates lightweight Convolutional Neural Networks (CNNs) for face detection and recognition within a virtualized edge computing setup. Model optimization techniques, including structured pruning and INT8 quantization, are applied to reduce computational complexity and enable deployment on resource-constrained edge devices. To minimize bandwidth usage between edge and cloud components, facial embeddings are compressed from 512 to 128 dimensions. A hybrid loss function combining ArcFace and Center Loss is employed during training to enhance robustness against variations in lighting, pose, and multi-face classroom conditions. The architecture distributes processing between simulated edge nodes and a centralized cloud server to evaluate latency, bandwidth consumption, and scalability. An adaptive multi-frame attendance validation mechanism further reduces false positives. Experimental results demonstrate improved efficiency and reduced computational and network overhead without compromising recognition accuracy.