Online Auxiliary Evaluation of Physical Education Teaching Based on Facial Expression Recognition

Internet plus technology and artificial intelligence technology are widely used in online sports teaching and curriculum evaluation tasks. However, existing deep network-based online facial expression recognition is susceptible to complex scenarios such as lighting, and occlusion, which directly affect the accuracy of course evaluation. To this end, this paper designs an emotion recognition network based on spatiotemporal hypergraph convolution for robust online emotion analysis. Specifically, we collect facial video sequences from different clients and generate corresponding facial landmark sequences. On the server side, an effective spatiotemporal hypergraph convolutional network is deployed, in which the hypergraph convolution module can exploit high-order relationships between facial landmarks. To verify the effectiveness of our model, we conducted extensive comparative experiments on two public expression datasets and our self-built dataset. The experimental results show that the proposed model obtains higher accuracy and effectively improves the quality of physical education teaching evaluation.