ABSTRACT
The arrival of COVID-19 has changed the way traditional classes are conducted, and online teaching has never been more popular. While there are many advantages to online teaching, there are also extremely obvious disadvantages, one of which is the tendency to lack concentration. For this reason, this study uses video images from the DAiSee dataset, a new sampling script, deep learning neural networks, and a new PAD emotion model to systematically assess student concentration. Our test set uses 21 short videos from the DAISee dataset, sampling a total of 1,866 frames. The final results showed that the accuracy of the neural network was approximately 80%. The results of the test set on the PAD model showed that the percentage of attentive listeners was 65.9%, while the percentage of highly inattentive listeners was 6.2%. This study constructed a complete concentration monitoring system for online classrooms centred on smart education which can provide the information of students' concentration in real time. © 2022 ACM.