A New Feature Fusion Network for Student Behavior Recognition in Education
A. Jisi, Shoulin Yin
Abstract
ABSTRACT Behavior recognition is a research hotspot in the field of computer vision and it also is a challenging task. In particular, student behavior analysis has an impact on the efficiency of classroom education. Aiming at the complex student behavior recognition problem in the video, we propose a new feature fusion network for student behavior recognition in education in this paper. The new feature fusion network contains two main stages: feature extraction and classification. First, we combine spatial affine transformation network with convolutional neural network to extract more detailed features. Then, the weighted sum method is adopted to fuse the spatial-temporal features, and the softmax classifier is improved for classification recognition to improve the final recognition result. Experiments are carried out on standard human behavior data HMDB51, UCF101 and real student behavior data. The results show that the proposed algorithm can achieve better recognition effect than other state-of-the-art recognition algorithms.