YOLO-v5 Based Single Step Student Affect State Detection System

Online education has increased tremendously due to vast availability of internet recently. Student emotion and engagement is directly related to learning goals and productivity. The existing computer vision based student engagement analysis techniques require two steps for engagement detection. In this paper, single step student affect state detection method is proposed using recent deep learning algorithms. Also a learning centered affect state dataset is curated from public databases. The YOLO-v5 deep learning algorithm is trained on the curated database to detect the affect states. The experimental results show that the proposed one step method is able to detect the affect states reliably. The proposed method also performs inference on an edge device with limited compute resource. The proposed method achieved 0.996, 0.921, 0.96, and 0.777 values of overall precision, recall, mAP@0.5, and mAP@0.5-0.95 respectively.