Research And Implementation Of Environmental Adaptation Of Wireless Indoor Localization

Indoor location service plays an important role in people's daily life.The signal strength of satellite positioning system widely used outdoor will be greatly reduced due to the blockage of buildings,thus limiting the use of the system indoor.At present,WiFi devices are cheap,practical and widely deployed.WiFi-based wireless positioning has received widespread attention.And WiFi-based indoor device-free positioning can locate users without carrying any electronic devices.Compared to active positioning,which requires users to carry electronic equipment,device-free positioning has greater advantages than that in terms of cost and convenience.Then we focus on indoor devicefree positioning based on channel state information(CSI).Many studies assume that the distribution of CSI fingerprints remains stable over time,but this assumption is not actually true,because WiFi signals are easily affected by environmental factors such as furniture changes and device position changes,etc.These factors disable the original positioning system,then people have to respend a lot of time and manpower to retrain the positioning system.Rebuilding is a very inefficient way of working hindering the wide application of WiFi.We attempt to make indoor device-free location system to cope with the change of the indoor environment at a minimum cost.This article first improves the performance of device-free positioning system.We first select robust positioning features from RSSI,CSI amplitude,and phase features,etc.and denoise the characteristics of the original CSI using isolation forest algorithm.And we propose to use one-dimensional convolutional neural network(1D-CNN)for devicefree location,which achieves better positioning results than other neural networks.This article focuses on the analysis of environmental change issues.From the perspective of fingerprint distribution in single-position on the micro-level and fingerprint distribution in overall room on the macro-level,we first analyze the impact of environmental changes on CSI fingerprint characteristics and device-free positioning model.And device-free positioning with domain adaption based on this analysis are verified.Then we design a positioning model based on deep neural network with domain adaptation.The model unifies the positioning training and the domain adaptation.It not only performs positioning training based on a large amount of CSI data in the source domain,but also uses semantic alignment(SA)label-based to perform feature domain adaptation,and use only a small amount of target domain data to complete the domain adaptation.Then the target loss functions of the two processes are unified to train the positioning model with environment adaptation.In order to implement a complete adaptive device-free positioning system,we also propose to use a fully connected deep neural network(DNN)classification algorithm for scene recognition,and simultaneously propose a two-level discrimination scheme for model update judgment.Finally,we compared our scheme not only with the industry's environmental adaptive positioning method,but also with the method of applying transfer learning theory to the environmental adaptive positioning in multiple real experimental rooms.The results show that the environmental adaptive positioning scheme can significantly reduce recalibration.In the case of workload,the average improvement effect of positioning is more than 90%,which is close to the positioning benchmark accuracy of the target domain.At the same time,the scene discrimination scheme can achieve a discriminative level close to 100%.All reuslts prove that the proposed scheme is effective and advanced in responding to environmental changes.