Research On Human Motion Recognition Based On WiF

Thanks to the development of Io T and sensor technology,Human Activity Recognition(HAR)has achieved a wide range of applications in smart home,healthcare and fall detection.Human activity recognition is generally divided into two types: contact and non-contact.The contact method puts extra burden on the body.WiFi-based human activity recognition methods have attracted the attention of researchers because of their advantages such as easy deployment,no visual restrictions and no invasion of people’s privacy.Meanwhile,with the increasing speed of computer computing,feature extraction of WiFi signals using deep neural networks to achieve higher accuracy classification has accelerated the development of human activity recognition.In this paper,we address the problems of low recognition accuracy,poor adaptation to environment,lack of training dataset,and inability to visualize in some current WiFi-based human activity recognition techniques,and study a deep learning using Multi-layer Bi-directional Long Short-Term Memory Network(MBLSTM),A deep learning method using Multi-layer Bi-directional Long Short-Term Memory Network(MBLSTM)is investigated to improve the recognition accuracy.We also use Meta-Learning to train WiFi data to improve the ability of human activity recognition system to adapt to the environment,and develop Digital Twins(DT)system for human activity recognition.(1)To address the problem of low accuracy of human activity recognition based on WiFi,this paper proposes a deep learning method of MBLSTM for non-contact activity detection of humans.The method can learn forward and backward activity features from the original WiFi time series,and introduce an attention mechanism to assign different weights to the learned features,and finally perform activity recognition.The experimental results show that the proposed method achieves an accuracy of more than 96% for the recognition of six activities in multiple rounds of testing,which exceeds other benchmark methods used for comparison.(2)To address the problem that human activity recognition cannot be visualized,this paper innovatively combines human activity recognition and digital twin.This paper uses metalearning to train the human activity recognition network and increase the network’s ability to adapt to the environment,and then sends the recognition results to the digital twin system.In the digital twin system,real-world physical activities are mapped onto a digital spatial human model,and the activities are evaluated for safety,and warnings can be issued quickly when dangerous activities occur.Meanwhile,the training of the network requires a large amount of data,and this paper uses a public WiFi collection tool to collect the data,which is able to record the WiFi signals transmitted from the transmitter to the receiver,and make the collected data into a public dataset.