Classroom Teaching Evaluation Based on Facial Expression Recognition
Xiaoyu Tang, Wang-yue Peng, Sirui Liu, Jianwen Xiong
Abstract
Due to the lack of attention to students' emotional state in the traditional teaching mode, the teaching effect and detailed evaluation are hard to assess. This paper proposes a smart teaching evaluation method based on facial expression recognition in classroom, which is more real-time, more objective and more fine-grained. This method considers the students' emotional states and combines the emotional state model with the traditional teaching evaluation method. The classic convolutional neural network AlexNet is used to complete the pre-training of facial expression recognition and achieved an average accuracy of 92.68% in JAFFE and 99.10% in Ck+. Image samples that capture and track students' facial expression at several key time points will be sent into the well-trained network and then the emotional states of each student will come out. The engagement weights of different expressions were given according to the distribution of six basic expressions in the PAD emotional state model. The comparison between experimental conclusion and teacher's evaluation shows that the method is correct and effective, and plays an intelligent and efficient role in teaching evaluation.