RSS Missing Value Estimation With Adaptive Context Generative Adversarial Networks Model

In recent years,with the rapid development of mobile communication technology and the popularity of computer technology,the demand for service to obtain location information in outdoor environments(e.g.,Baidu Map,Google Map,Amaps)is increasing day by day in recent years due to the availability of GPS in most smart phones.Unfortunately,GPS does not perform well in indoors,construction and basement,and places close to the wall as the signal from the GPS is too weak to come across most construction,thus making GPS hard for indoor localization.Since GPS positioning technology is limited to providing accurate positioning information outdoors,indoor positioning technology has attracted widespread attention.At present,Wi-Fi positioning is currently the mainstream method in indoor positioning.With the growing number of Wi-Fi devices(e.g.,smart phones,smart TVs,Internet of Things),the viability of using existing Wi-Fi infrastructure as a platform for Location Based Services is becoming increasingly attractive.In such cases,Indoor Positioning Systems will provide precise,scalable,and real-time positioning information about the mobile devices.However,RSS value in the fingerprint database will change with the variability of the indoor environment,we usually need to constantly re-measure the value in the fingerprint database,which leads to high cost and long time,especially in the dynamic environment with large positioning area.So the construction of fingerprint database is the key to Wi-Fi positioning system.To address this problem,We propose a method of constructing fingerprint database based on Adaptive COntext Generative Adversarial Networks Model,and apply it to simulation environment and field scene,which mainly includes the following aspects:(1)We review the methods of constructing fingerprint database in the field of indoor positioning and analyze the advantages and problems of these methods.Then we introduced the mainstream indoor positioning technology at the same time,and most importantly,study and elaborate the Generative Adversarial Networks Model and Autoencoder in detail.(2)We normalize,cut and transform the simulation data generated by the ray tracing technology and the data collected in the field scene,and finally convert the processed data into the input data of the model by using the binary mask technology.(3)Adaptive COntext Generative Adversarial Networks Model is proposed in this paper.The model only needs to measure part of feedback,and then learn the feedback distribution to finally predict the missing fingerprint at a specific location,thus reducing the time and cost of collecting data.The author of the paper conducted a simulation experiment.The experimental results show that the accuracy of indoor positioning is significantly improved,and the labor cost is greatly reduced.