Contagem e identificação de pessoas em sala de aula através de visão computacional e internet das coisas

Abstract

School attendance is essential for efficient educational management. Traditional methods seem less effective than technologies that, in addition to saving teachers’ time and effort, allow the extraction of new insights from the data generated. Artificial Intelligence, with techniques such as Machine Learning, Natural Language Processing, and Computer Vision, has become part of people’s daily lives. However, complex tasks like these require computational offloading to more powerful devices, such as in Cloud Computing, due to the limitations of Internet of Things devices. This work adopts a Computer Vision approach for facial recognition of students in the classroom, aiming to automate the school attendance process. The proposed system is based on a Raspberry Pi with a camera that captures images, which will be processed by the device itself or sent to an edge device with greater computational power. Once the set of pictures or characteristic data is on the edge device or the Raspberry Pi, the system will offer various facial detection and recognition models to perform attendance and subsequent logging into a database. This work used the Single Shot MultiBox Detector and ArcFace models, which work together to detect faces and perform facial recognition, respectively. The experiments were conducted as a proof of concept and a computational stress test. The results show that the system achieved 85% accuracy in recognitions. Computational offloading proved a valid option, although the Raspberry Pi presented an average time of 4.05 seconds to complete the task.