Abstract：Teachers' classroom teaching action is an important part of classroom teaching activities， and the recognition of teachers' teaching action is of great significance for evaluating the quality of classroom teaching. A method for teachers' teaching action recognition based on spatial temporal graph convolutional networks (ST-GCN) is proposed in this paper. In this method， the human skeleton point information is first extracted by taking the single frame image in the teacher's teaching video as a unit， and then ST-GCN is used to aggregate multi-frame image information to identify teachers' teaching behaviors. To verify the effectiveness of the proposed method， two sets of video datasets containing 6 categories of teachers' daily teaching actions are constructed to conduct comparative experiments. Extensive experiments show that the teacher's teaching behavior recognition method based on ST-GCN can effectively eliminate the interference of irrelevant information in the classroom scene， and make full use of the spatiotemporal information generated between the skeleton points in the multi-frame images to accurately identify the typical teaching action of teachers. It has higher accuracy and stronger robustness. The related research in this paper can reflect the teaching status of teachers timely and effectively， help teachers optimize teaching behavior in time， and help smart teaching.
2023, Vol. 57 
