AATiENDe: Automatic ATtention Evaluation on a Non-invasive Device

Abstract

The study of student attention is an important topic in education because this type of analysis provides important information to teachers to potentially improve the quality of their classes. In this paper, we present AATiENDe, a system that uses emotion recognition, gaze direction approximation and body posture analysis as features to classify whether students are paying attention to their computer screens. To do this, we use a mixture of deep learning-based techniques and novel machine learning techniques applied to tabular classifiers to produce the final predictions. We also capture and label a customized dataset to train the models. Our approach provides over 90% accuracy using two cameras and over 80% accuracy using only the foreground camera.