Research On Human Behavior Recognition Technology Based On OneNET Platform

With the development of social information and intelligence,the future society will inevitably move toward the "Internet of Things" Internet of Things.The Internet of Things will change the way of research in the field of computer vision and artificial intelligence,and human behavior recognition as their hot research direction will naturally become a part of the Internet of Things.The human behavior recognition technology at the present stage is mainly the motion recognition of people in the video,but it is easily affected by such factors as the the shooting angle,the dynamic background,the execution subject and so on,And with large amounts of data,it takes time for transmission,Therefore,the data source and the algorithm have strict requirements.In response to these problems,this paper proposes the research of human behavior recognition technology based on the Internet of Things platform for the first time.It contains two parts:the Internet of Things platform and the behavior recognition algorithm.The Internet of Things platform uses China Mobile OneNET.Its greatest advantage is its support for cross-platform,cross-regional hardware device access and Io T applications.In this paper,the MQTT public protocol and RGMP private protocol are used as the device access protocol,The distance sensor and color camera based on the Kinect machine are used as the video acquisition module to provide the data source acquisition function;the development of the interface based on the RESTFUL API interface is implemented to interface with the platform,and the RTMP real-time information transmission protocol is used to implement the internal video push and pull of the platform;Using Kirin Development Board,TCP/IP network protocol and GPRS/WiFi network to establish a smart terminal based on OneNET platform as a data transmission channel.Next,A method of human behavior recognition based on gradient is proposed,which uses depth-corrected HOG and P-HOG methods to extract human form and time information in depth sequences.It uses depth-modified HOG and P-IIOG methods to extract human body shape and time information in depth sequences.Firstly,using the saliency detection method to denoise the original depth image and highlight human behavior in the scene,Secondly,use the depth-modified histogram of gradient(D-mHOG)to obtain the limbs and time information of the person in the depth sequence,and generate a DM array to further optimize the D-mHOG descriptor,thirdly,applying a pyramidal histogram of gradient(P-HOG)to the DM array to obtain further spatiotemporal structural information in the depth sequence,Finally,the random decision forest is used to classify and distinguish the extracted behavioral features.We also compare the accuracy of the proposed algorithm with the classic recognition model,and get the confusion matrix on the three data sets of MSR Actions 3D,MSR Action Pairs 3D,MSR Daily Activity 3D.The human behavior recognition technology based on the OneNET platform proposed in this paper not only performs well in human behavior recognition,but also makes full use of the advantages of the Internet of Things platform for easy access,data storage,video output,and message distribution.