Action Recognition Algorithms Based On Spatio-Temporal Difference Information

Action recognition is one of the most popular fundamental research of video pre-cessing problems,which is widely applied in smart city,human-computer interaction,online education and other video monitoring fields.Although progress has moved on in this field,there are still many problems remained,according to the difficulty of video action recognition and the redundancy of video content.In this paper,we focus on the the research on action recognition algorithms based on Spatio-Temporal Difference Information,while improving the accuracy and real-time performance in practical applications.In view of the shortcomings and disadvan-tages of current mainstream action recognition networks in performance and practice,this paper proposes the Spatio-Temporal Difference Information and its corresponding extraction module to improve the performance and computational efficiency of the algo-rithm.Real-time student classroom behavior monitoring system is realized combining with human detection algorithms.The main contributions of this work are summarized as follows:1.The video action semantic learning guided by the word vector method is pro-posed,based on the multimodal convolutional neural network action recognition method.In this paper,the differences of different network structures under d-ifferent modalities are realized and compared.The generalization ability of the algorithm is enhanced by the method of word vector method.The data is tested on the datasets such as Kinetics and Moments In Time,and high accuracy of the algorithm is verified.2.The Spatio-Temporal Difference feature and its extraction module are proposed.The Spatio-Temporal Difference feature extraction module explicitly extracts the spatial and temporal correlation of video actions,which effectively improves the recognition accuracy and speed of the action recognition network.The perfor-mance of the network is verified on the UCF101 data sets.3.A classroom behavior monitoring system integrating students' human detection algorithm and action recognition algorithm is built.The system uses YOLO detection network as the human body detection module,and carries out model compression and information distillation on the action recognition network,and finally realizes the real-time student classroom behavior monitoring system.