Real-time monitoring system for attendance and attentiveness in virtual classroom environments

With the outbreak of the COVID-19 pandemic, classroom environments have been subjected to revolutionary changes via the employment of virtual classrooms and allied technological advancements. The traditional methodologies are proving to be inefficient in such an environment for teaching as well as managerial tasks. Also considering their cumbersome nature, the need for a newer, stronger, and better model is evident. As of now, many Deep Learning techniques have been employed for the purpose, ranging from the usage of standard object detection APIs or even CNNs and their variants. Our study proposes a model based on SVM embedded on top of embedding vectors combined with a Single-shot detector for real-time monitoring of attendance and attentiveness of students in a virtual classroom set up making use of video feed. A small comparative study between the proposed model and dlib, a standard library for the purpose as well is performed. The results show that our model outperforms dlib methodology significantly with high accuracy and performance efficiency. We had done experimentations on the fer2013 dataset particularly for emotion detection and custom datasets in general. Even though the model performs well in our experimentations, the need for a stronger and better dataset is high for evaluating the model and implementing it in real-life scenarios.