A Method for Analyzing Teacher Behavior in Classroom Based on the Long- and Short-Term Features of Pose Sequences

Abstract

The analysis of teachers’ behaviors in classroom videos is helpful to objectively evaluate teaching, promote teaching reform, and improve teaching effectiveness. With the continuous introduction of deep learning models, there have been many analysis methods, but the recognition accuracy is not high enough. In this paper, we construct a new model combining convolutional neural network and recurrent neural network to learn and fuse short-term features with long-term features, and use the spatio-temporal information of videos to get the behavior categories. In our experiments, we collected teacher classroom videos in teaching scenarios to construct the dataset. The 2D pose heatmap sequences are extracted from the videos as model inputs, which are used to exclude the interference of environmental noise in the teaching scene. The experimental results show that this method can achieve higher recognition accuracy compared with the PoseR(2+1)D method on the teacher behavior dataset for teacher behavior analysis in teaching scenes.