Research And Implementation Of Human Behavior Recognition System Based On GCN

Human behavior recognition is one of the hot issues in the field of computer vision.The Internet has been accumulating data faster and faster;and the Big Data era has produced specialized behavioral identification data sets,such as Kinetics.In recent years,deep learning has promoted the development of computer vision.The enhancement of the performance of the graphics processing unit(GPU)by NVIDIA,AMD and other manufacturers not only accelerates the processing speed of the hardware,but also improves the computing power and accuracy of the deep neural network.The human behavior recognition algorithm has also greatly improved in speed processing and accuracy.This paper studies and implements the human behavior recognition system based on the human skeleton diagram.The human skeleton data is obtained by means of the pose estimation algorithm,which is used as the basic research and implementation of human behavior recognition method.This paper defines the sampling area and weight size on the bone map by redefining the sampling function and weight function used in the convolution operation,and obtains the convolution operation for the bone data of this paper.The Graph Convolutional Networks(GCN)model is designed to apply the trained model to the bone map for human behavior recognition.Based on the research of GCN human behavior recognition system,the human behavior recognition system of B/S mode is designed and implemented.The front end uses the React framework,and the back end is written in Node.js.The overall structure of the system consists of five parts,namely the behavior recognition model building module,the user registration and login module,the video acquisition module,the video upload module and the behavior recognition module.After preprocessing the data set,the behavior recognition model building module completes training and saving of the GCN model on the deep learning server.The user registration login module enables the user to register in the system.The video capture module can call the camera to capture video and save the video file.The video upload module transmits the video file to the back-end server;and the behavior recognition module completes the behavior analysis of the video,and generates and processes the video.After the video file.The behavior recognition algorithm of this paper is trained on the existing NVIDIA GTX 1080 GPU.The application can detect the changing human motion on the video clip and give the final behavior classification result,which achieves the expected effect.