A Student Action Recognition Algorithm Based on Adjusted Variational Auto Encoder
Simin Li, Yaping Dai, Ye Ji, Kaoru Hirota, Wei Dai · 2021
Student action recognition plays an important role in detecting students learning behaivor in online courses. In order to improve the accuracy of student action recognition in classroom, an algorithm based on Adjusted Variational Auto Encoder (AVAE) is proposed. By adjusting the traditional Variational Auto Encoder (VAE) method, the proposed algorithm can effectively extract the characteristic parameters of students from classroom video images, and then recognize the students' action. Experiments show that the proposed algorithm for student action recognition preforms better than traditional VAE algorithms with higher accuracy and convergence speed, and improves the recognition accuracy by 5.13% compared with traditional Convolutional Neural Network (CNN) method.