Research On Human Behavior Perception And Recognition Technology

In recent years,with the rapid development of information science and technology,people's demand for intelligent life is also constantly improving.This demand promotes the continuous development and expansion of artificial intelligence technology.Human behavior recognition,as an important branch of artificial intelligence,is receiving extensive attention from international and domestic researchers.Two kinds of human behavior perception and recognition technologies are researched in our paper,respectively: human behavior perception and recognition technology based on node and human behavior perception and recognition technology based on CSI.Based on the research of these two human behavior perception and recognition technologies,our research mainly includes the following aspects:1.Make use of Open Pose,an open source platform,to perceive the human skeleton structure in video,and extract the coordinates of its nodes;2.Establish a CSI data collection platform to perceive human behavior,and explain the essence of CSI signals and the collection principle and process;3.The recognition effects of four machine learning algorithms on different human behavior data are analyzed,which are the KNN,QDA,SVM and RF.4.During the processing of node data,key video frames were extracted by clustering method,and several key nodes with high information entropy were selected from the nodes as the key points by calculating the information entropy of each key node;5.In the classification and recognition of the node data,we studied the recognition effects of two classification patterns on the node data respectively: template matching classify and Classification model classify;6.Processed the original CSI data by Outlier handling,low-pass filtering,PCA dimensionality reduction and other data preprocessing methods,and proposed a sliding window method to intercept the CSI data between the start and end points of human movement;7.Amplitude and phase characteristics of CSI behavior data on three groups of antenna pairs were extracted respectively,and the recognition effects of different machine learning classifiers on amplitude and phase characteristics of different antenna pairs were discussed.In the research on human behavior perception and recognition technology based on the node,the classification accuracy of the template matching classification for 7 human behaviors is about 78% on average,while the classification accuracy of the classification model is generally over 90%.In the study of human behavior perception and recognition technology based on CSI,the average classification effect of four machine learning classifiers on CSI data is up to about 85%.