Student Action Recognition Based on Fuzzy Broad Learning System

Abstract

Automatic recognition of student action is an important means to evaluate students' learning status in the class. It also provides a technique for measuring the effectiveness of teaching. However, the complexity of student action poses a challenge to automatic recognition. In this paper, a student action recognition method based on the fuzzy broad learning system (fuzzy BLS) is proposed. Fuzzy BLS is designed by merging the Takagi-Sugeno (TS) fuzzy system into BLS. As a neuro-fuzzy model, fuzzy BLS overcomes some problems, such as suffering from a time-consuming training stage and a large number of fuzzy rules. To get more abundant local features from student action images, we use the Scale-Invariant Feature Transform (SIFT) descriptor combined with the Local LogEuclidean Multivariate Gaussian $(\mathrm{L}^{2}\mathrm{E}\mathrm{M}\mathrm{G})$ descriptor to extract image features. Then, the extracted features are fed into fuzzy BLS after dimension reduction. The experimental results on the self-built dataset have shown that the proposed student action recognition method achieves better performance than other benchmarking methods.