An Edge-Optimized Vision-Based System for Smart Classroom Attendance Automation and Dynamic Crowd Analytics Using YOLOv8 and Deep Face Embeddings

Abstract

This work focuses on the development of an intelligent vision-based framework that would enable automating the process of classroom attendance and environment monitoring using edge computing. The current attendance procedures face problems caused by delays in processing and insufficient speed due to the absence of real-time information. In order to address these problems, the proposed project is aimed at combining high-speed object detection and deep face verification. For this purpose, the authors use the state-of-the-art YOLOv8 architecture to detect students' locations, as well as embeddings provided by the ArcFace network to enable precise identification of each individual automatically through local edge-based processing, without transmitting raw video or facial data to the cloud. In order to prove the efficiency and reliability of the framework, the researchers prepared a custom dataset comprising more than 4,500 video frames with 100 unique identities that account for all possible difficulties, including changing lighting conditions and dense groups of people. To ensure that the model could be deployed in the academic setting effectively, the architecture was optimized to operate on edge-computing-enabled workstations, such as a workstation powered by an Nvidia graphics processor and an Asus Vivobook 15 computer. As shown by the results, it significantly reduces processing time and preserves data confidentiality.