Design And Implementation Of Human Action Recognition System Based On Graph Convolutional Neural Network

With the dissemination of a great deal of data and information on the Internet,network health has become a problem that can not be ignored by the society.Harmful network information poses serious harm to the health of the country,society and people.Using relevant technology to analyze the information in pictures,videos and other documents can prevent the dissemination of information in a timely manner.Action recognition is a very challenging subject,which plays an important role in video and picture comprehension.Action recognition is also widely used in medical test,intelligent monitoring and other fields.The topic is based on the graph convolutional neural network to design the action recognition algorithm,the realization of human action recognition system,the relevant technology research and application,for the sake of improving the recognition accuracy of the human motion recognition algorithm,to achieve a complete function of the action recognition system,and ultimately attain the goal to meet the needs of users.This subject has mainly completed these works:1.Firstly,it is clear that the human pose estimation algorithm is used to preprocess the video and pictures of action recognition,and it is clear that the dynamic changes of bones can provide a lot of action information,and the video is modeled as a spatio-temporal graph model.The model feature extraction scheme was explored,and several classic feature extraction schemes were compared in spatial dimension and time dimension respectively.The method of selecting graph convolutional neural network for spatial feature extraction and temporal convolutional network for temporal feature extraction were defined.2.Secondly,put forward the multi-filter dynamic convolution neural network algorithm,the algorithm is based on the figure convolution neural network,combines the multi-filter graph convolution structure,dynamic skeleton diagram structure and multi-scale time domain convolution structure,optimize the space and time feature extraction scheme,implements the new action recognition model,improves the accuracy on the data set.3.Finally,human action recognition system was designed and implemented,based on the multi-filter dynamic convolution neural network model,set up system related to the environment,deployment of the model in the system application.Through the requirements analysis of system,the function of data collection,service of action recognition and data display of the movement recognition system is designed,and the human movement recognition system is realized.Finally,the function of motion recognition system is tested,also the performance of the system.A human action recognition system which can meet the needs of users is obtained.