Based on the Neural Network Classification of Human Behavior Research

Character action behavior classification is an important task of video analysis, it has a wide range of applications in security, sports, entertainment and other fields. However, due to the complexity and diversity of the scene environment, as well as the variability and difficulty in defining the character action behavior, video-based character action behavior classification is a challenging task. Human behavior recognition is becoming a research direction that attracts attention nowadays, and has been widely used in the fields of fall person monitoring, dangerous event warning, human-computer interaction, virtual reality and so on. Traditional algorithms can only extract shallow features and cannot model the spatio-temporal relationship effectively. In recent years, two-stream convolutional neural networks based on deep learning have become increasingly popular due to their dual-stream architecture, which can effectively model spatio-temporal relations with high accuracy. However, due to the complexity and variability of video scenes, human behavior recognition technology still faces many difficulties. For example, the feature extraction method of the two-flow convolutional neural network is simple, and the spatio-temporal information cannot be sufficiently extracted to model the long video sequence. In this paper, a character action behavior classification method based on dual-flow hybrid neural network is proposed. The spatial and temporal features of video are extracted by using dual-flow convolutional neural network (CNN), and then the two features are fused and sequence modeled by using long short-term memory network (LSTM), and the accurate recognition of character action behavior is finally realized. The reliability of the algorithm is verified by experiments.