Frequência escolar por meio de reconhecimento facial utilizando redes neurais residuais

Abstract

The practice of recording and monitoring student attendance is a fundamental action in various contexts, especially in the school environment, although it often consumes a significant part of class time due to its manual process. This study proposes the development of a facial recognition model with the aim of automating school attendance control. Based on the contributions of King (2009) and Convolutional Neural Network (CNN) techniques, the model is designed to overcome the challenges related to the lag of image capture devices and variations in the classroom environment.The database consists of five class sessions from the study conducted by Mery et al. (2019) and one class session collected at a school unit of the Serviço Social da Indústria (SESI), totaling almost 100 images for analysis. Compared to the Facenet512, Facenet and ArcFace models, in terms of final frequency marking accuracy, the model showed superior performance. With an increase of around 51% in accuracy compared to the others, reaching a satisfactory 76% accuracy mark, as well as demonstrating superior specificity. It is important to note that, at this stage of development, the model has the potential to benefit both teachers and students, helping to save time and reduce errors in school attendance management.