Research On Wireless Indoor Fingerprinting-based Localization Method With Federated Learning

In recent years,the wireless indoor fingerprinting-based localization technology has been widely followed by researchers,due to its low cost and high versatility.However,since dynamic indoor environment and inherent defects in WiFi transceiver devices,wireless location fingerprints will change significantly over time,leading to a sharp drop of localization accuracy.In order to maintain the localization accuracy of fingerprinting-based localization model,researchers have explored and proposed a model update method based on crowdsourcing,and achieved good performance.However,these efforts lacked protection for location privacy during the model update process,resulting in the user's location information being easily leaked and exposing of the user's privacy.For the goal of maintaining the localization accuracy without the risk of privacy leaking,we proposed FLoc,a fingerprinting-based indoor localization system which updates the localization model via a federated learning framework.The FLoc localization system includes a model server and mobile devices carried by various users.In the FLoc system,each user maintains a localization model in own mobile device and completes indoor localization independently.In order to realize the function of automatically collecting and marking fingerprinting data on mobile devices,FLoc introduced Pedestrian Dead Reckoning(PDR).Each mobile device uses the locally collected fingerprinting data to update the localization model and at the same time periodically shares the encryption update parameters to the model server.On the model server,the encrypted update parameters of all local models are aggregated to update the server-side model,and then the model server can share the server-side localization model to each mobile device.In this way,the localization model of each mobile device not only contains its own update parameters,but also benefits from the update parameters of other mobile devices.Finally,FLoc conducts evaluation and verification in laboratory corridor and home room.Experimental results show that FLoc can automatically collect fingerprinting data through PDR,complete the update of the localization model,and has the ability to maintain the accuracy of the localization model,safely.