JoWUA

Abstract

Facial expressions are physical changes that reflect an individual's feelings, emotions, intentions, or social interactions. In the field of computer vision, facial expression analysis needs a higher level of knowledge. To increase student involvement in class, in recent years increased in employing technology to track and evaluate students' facial expressions. Facial expressions, as a nonverbal form of communication, can provide valuable insights into students' emotional states. This research article aims to explore the potential of class engagement monitoring using facial expressions. Hence, we proposed a Bi-upsampling Feature Pyramid Network (BiusFPN) with multiple attention map integration. We used ResNet18 as a backbone and we also introduced an inter-coordinate attention model which improves the feature extraction from local spatial extended mode with different coordinate representations. This engagement analysis model precisely detects the facial changes and yields an accurate outcome. We integrate both Channel and spatial attention mechanisms to fuse different attention maps for the final representation. The result of the classification layer will be Disengaged or partially engaged or engaged. This approach achieves 68.16% accuracy for the DAiSEE dataset and 83% for the WACV dataset. In this way, the proposed