Computer-Vision Based Attention Monitoring for Online Meetings

Due to the appearance of COVID-19, virtual video conferencing platforms like Zoom and Google Meet have become one of the main alternative ways to conduct virtual meetings and presentations. While the virtual platforms are cheaper and more flexible, presenters and meeting hosts are likely less efficient at assessing audience attention and engagement due to the lack of body language. In this paper, we propose a system for estimating and monitoring participant attention in virtual meetings by using computer vision. Our approach mainly focuses on changes in a person’s presence, gaze direction, and head orientation as a computer camera has a limited field of view. We first created a module to detect and extract participant video cells to isolate users and process their attention individually. Using those videos, we then monitored the user’s presence, using YOLOv3 and DeepSORT, and their gaze direction and head orientation, using PTGaze. Through this monitoring, the system is able to record and graph a user’s attention over the total amount of frames and return a collective attention level graph for the entire meeting. We believe that our system has potential usage in settings where attention is critical, such as academic lectures or collaborative business meetings.