Research On RFID Motion Recognition Technology In Complex Environment

With the development of technologies such as the Internet of Things,5G and the Industrial Internet,it is becoming more and more important to accurately perceive human motions.RFID technology is a powerful choice for motion recognition due to its low cost,small size and battery-free advantages.However,most existing RFID-based motion recognition schemes are designed under ideal experimental environments,and they are not suitable for use in complex realworld environments.Because,in the real scene,RFID is usually arranged in the indoor environment,and the indoor environment has complex multipath effects.The multipath of the indoor environment will affect the accuracy of motion recognition from two aspects:First,the multipath signals will be mixed with each other at the tag end and interfere with each other,resulting in the masking of motion-related features;Second,the multipath is heavily dependent on the environment.When the environment changes,the multipath will also change,resulting in a system with higher recognition accuracy in a certain environment.When the environment changes,the recognition accuracy will drop significantly.Therefore,in order to realize the implementation of RFID intelligent motion recognition applications,this paper studies the complex multipath in the indoor environment and proposes two solutions to the multipath problem:(1)Aiming at the problem of multipath mixed masking motion characteristics,we propose an indoor motion recognition scheme based on multipath perception:multipath signals actually contain rich information for motion recognition.By decoupling multipath signals,analysis AoA of each path can extract the corresponding features of motion recognition from it,and realize high-precision motion recognition in indoor environment.Specifically,we extract the signal at the tag end and eliminate hardware noise through data preprocessing,use the MUSIC algorithm to decouple the multipath signal to obtain the signal spatial spectrum,and finally use the CNN-LSTM based deep learning framework to extract motion features from the spatial spectrum To identify the corresponding motion mode.(2)Aiming at the problem that the recognition accuracy of the system will decrease due to the multipath will change with the environment,we propose a self-weight exercise recognition scheme based on multipath suppression:multipath signals can be further classified into target path signals reflected by the human body,and other multipath signals reflected by other things.The latter signal is less relevant to human motions and is more susceptible to environmental changes,so we suppresse other multipath signals and enhances target path signals.Specifically,we use the phase correlation of adjacent channels of adjacent tags to align the target path signal,thereby enhancing the target path signal while averaging other multipath signals to a small value.Then,we use the periodicity of self-weight exercise motions to propose a two-stage recognition scheme:single motion recognition based on DTW,and a set of motion recognition based on the voting method.In order to verify the recognition performance of the motion recognition system,this paper has conducted extensive experiments on the above scheme under different scenarios and conditions.The results show that the above scheme can achieve the design goal of achieving high recognition accuracy in a high multipath environment,and the latter scheme can ensure that the recognition accuracy is hardly affected when the environment changes.The implementation of the scheme proposed in this paper is of great significance to the application of RFID-based intelligent sensing,which can increase the scope of application of RFID technology and expand the functions of RFID systems,and has high practical application value.