Research On Wi-Fi Fingerprint Positioning Algorithm Based On EMD Filtering

With the popularity of mobile smart terminals,the demand for location services has proliferated,extending from the initial map route navigation to a wide range of areas such as smart homes,restaurant delivery,tourist guides and parking and car finding path planning.Unlike the outdoor environment where GNSS can meet most of the positioning needs,GNSS is unable to meet positioning needs indoors due to non-line of sight obstacles such as roofs and walls,as well as the complex and changing indoor environment,where GNSS has large positioning errors.As most mobile smart terminals have the capability to receive wireless signals and the widespread deployment of public Wi-Fi networks,this makes WiFi location fingerprinting-based positioning methods an important technical tool for indoor positioning.The Wi-Fi location fingerprint localization method can be divided into offline acquisition phase and online localization phase,where in the offline acquisition phase,the main work is to collect fingerprint data and carry out fingerprint preprocessing,and finally build a location fingerprint dataset;in the online localization phase,the fingerprint data at the Test Point(TP)is matched with the fingerprints in the fingerprint dataset for similarity according to the matching algorithm,and In the online location phase,the fingerprint data at the Test Point(TP)is matched with the fingerprints in the fingerprint dataset according to the matching algorithm,and the TP coordinates are predicted based on the Reference Point(RP)obtained from the matching.Since the current public fingerprint dataset has a single collection environment,simple area structure and insufficient data volume,this paper designs and constructs a Multi-dimension Underground Parking Fingerprint Dataset(MUF-Dataset);in the offline collection In the offline acquisition stage,the Received Signal Strength Indication(RSSI)is smoothed to address the fluctuation of the Wi-Fi signal due to signal interference,non-visual range propagation and multipath fading during the propagation process.The main work of this paper is as follows:Firstly,the fingerprint dataset MUF_Dataset was constructed in the main E underground car park of North China Electric Power University.23800 fingerprints were collected in 170 collection points over a period of 3 months,which also recorded the TP and RP coordinates,AP distribution,Wi-Fi signals in different frequency bands and the orientation of the fingerprint collectors.In addition,the number of stops made at the time of fingerprint collection,the temperature and humidity of the environment and the time of day were also recorded.Secondly,a fluctuation smoothing algorithm based on Empirical Mode Decomposition(EMD)filtering is proposed.Based on the fluctuation characteristics of the signal in the propagation process and the analysis of the time series signal,a time-based non-smooth non-linear RSSI sequence is constructed,the RSSI sequence is decomposed by EMD,and the Intrinsic Mode Function(IMF)selection criteria based on the energy analysis method and fluctuation coefficients are proposed.In the RSSI smoothing experiment and the localization experiment,the EMD filtering effectively smoothed the RSSI fluctuations and improved the localization accuracy compared with the traditional filtering method on the basis of preserving the differences in signal characteristics.Finally,on the basis of the traditional Weighted K-Nearest neighbor(WKNN)algorithm,adaptive dynamic K theory and fingerprint similarity metric that can describe the relationship between RSSI intensity difference and actual physical distance are introduced,and its improvement on localization accuracy is experimentally verified.Finally,based on the fluctuation smoothing method based on EMD,adaptive dynamic K and improved fingerprint similarity metric,the EMD-IWKNN algorithm is proposed and experimentally verified that the EMDIWKNN algorithm has good localization accuracy and stability.