Real-Time and Automated Student Attention Monitoring System in Classroom Using Deep Learning Techniques

Modern education system can be improved by using an automated classroom attention monitoring system, which can evaluate students’ attention in real time and reveal their involvement. Deep neural networks, computer vision, and other contemporary technologies can make automated real-time monitoring operations easier. In this paper, we represent a deep learning and computer vision based system for automatically assessing students’ attention in the classroom. To address the requirement of a larger dataset, we have gathered copious amount of real-time data from 120 undergraduate students on student behavior comprising approximately 7 hours of video footage. We have then used this data to assess two different categories of models: one that is object detection and classification models (YOLOv8, YOLOv9, and YOLOv10), which are applied to direct video images, and the other that is pre-trained deep learning classification models (VGG-19, ResNet-101, and MobileNet-v2),which are applied to cropped images of individual students. We have conducted two distinct analyses. The initial experiment involves assessing the performance of three variations of YOLO models: YOLOv8, YOLOv9, and YOLOv10 and the results show that YOLOv8 performs best with an accuracy of 97.61%. We have conducted another experiment on three different deep learning models: VGG-19, ResNet-101, and MobileNet-v2 and have compared their performance, where VGG-19 outperforms with an accuracy of 99.56%. YOLO is generally much better in terms of speed, lightweight design, and efficient for detecting objects in real time, but overall, for classification, VGG-19 performs better according to our experimental results.