A Comprehensive Survey of Students’ Engagement Detection for Online Classroom Management

Abstract Online classes have become a mainstream source of learning over the past few years, especially during the COVID-19 pandemic and its post-period. The convenience and ease of use of online learning have attracted many learners to online learning, alongside traditional learning modes. In the online learning approach, continuously tracking each learner’s affective state or level of engagement often becomes difficult, especially in a large classroom. Consequently, there is a need for an automatic method to detect the learners’ affective state. Over the last few years, several works have been targeted at addressing this problem. Among different techniques reported in the literature, machine/deep learning-based approaches used for image analysis are mostly common. This paper reviews the existing computer vision-based engagement detection methods in online learning to understand and summarize the process of identifying the affective state using emotion recognition. To carry out this survey, we used a different indexed dataset of literature published in the last ten years. Using four search strings and six inclusion-exclusion criteria, fifty-one research papers were considered in this study. To categorically analyze these works, nine research questions were formed. It was observed that existing works can identify student emotions with high accuracy. However, there is a need to focus on the multiple occluded images, images in low light conditions. There is also a need for further research on the association of emotion with the engagement detection of the students.