Vision-based Monitoring of Student Attentiveness in an E-Learning Environment

Due to the global spread of COVID-19, the world's educational institutions had been ordered to close. As a direct result of this, the time-tested method of acquiring knowledge by visiting classes is gradually being replaced by online education. In virtual classrooms, teachers had difficulty detecting student postures and determining whether or not students were comprehending the material. This research suggests using a computationally efficient method based on computer vision and machine learning to determine the attention levels of e-learning students. The method extracts characteristics using HoG and SIFT. Using K-means and PCA, the resulting feature vector is optimized for dimension reduction. The attentiveness is classified using the classifiers Decision Tree, KNN, Random Forest, and SVM. Random Forest yielded the best accuracy at 99.2% with a dataset of 15000 images.