Human Action Recognition In The Range Of Wi-Fi With CNN And ELM

With the development of the Internet,human-computer interaction technology has become one of the core technologies in the field of artificial intelligence.More and more researchers are investing in this field,making the technology in this area developing rapidly.Some traditional human-computer interaction methods require the use of voice recognition,video image analysis,and some require the user to wear a sensor device,which will affect the user experience.Considering these problems,the human-computer interaction technology based on wireless signals has become a hot topic for technical personnel and has a broader application prospect.In this paper,a novel human motion recognition scheme based on Wi-Fi signal is proposed based on the theory of wireless signal processing and machine learning.In this paper,five aspects of system construction,data acquisition,data stream segmentation,feature extraction,and classification and identification are studied.The main work is as follows:This paper uses the software radio platform USRP N210 to build a communication system based on IEEE802.11a.At the same time,in the protocol,the module for channel estimation adds the function of preamble data collection.The preamble extracted during the communication process is a continuous data stream.To perform motion recognition,it is first necessary to extract independent motion information,that is,it is necessary to determine the start and end positions of the motion sample in the data stream.This paper proposes an endpoint detection algorithm based on data stream to extract the action information.In order to achieve effective human posture recognition,this paper studies the feature extraction.This paper proposes to use convolutional neural network(CNN)network to complete the feature extraction of motion samples.By contrasting with the statistical characteristics,the convolutional neural network is proved has obvious advantages in feature extraction of human motion samples.On the basis of the above,an extreme learning machine is used to achieve sample classification.Simulation and test results show that the proposed human motion recognition scheme can effectively realize the recognition of nine actions with a small number of samples.The recognition rate can reach 94.7%,which has certain research value and practicality,and provides a new solution to the problem of wireless signal recognition.