Fingerprint Localization Through CSI Based Region Partition

With the popularity of smart terminals and mobile Internet,the world has entered the era of smart manufacturing and digital consumption.Location-based services have penetrated into all aspects of people's lives,and have broad application prospects in industries such as catering,logistics,retail,manufacturing,medical care,and security.As the space of modern buildings is getting bigger and bigger and the structure is getting more and more complicated,people spend more and more time indoors.The growing demand for indoor positioning has made indoor positioning a hot topic in positioning technology research.The CSI-based partition fingerprint localization process includes two stages,offline and online,like the existing fingerprint localization.The first task in the offline phase collects the CSI fingerprint data of the reference point and establishes a complete fingerprint database.Due to the interference of external factors,the directly collected CSI data often has abnormal points,packet loss and unstable waveforms.Therefore,it is necessary to preprocess it when constructing the fingerprint library to improve the stability of the fingerprint.The Hampel function is used to detect anomalies.The detected anomalies will be deleted and appropriate values will be inserted after the deletion.When the packet is lost,the mean method will be used to recover.The pre-processed CSI data builds an overall fingerprint library.Then train a classification model to distinguish the undetermined points in the LOS area and the NLOS area due to fixed obstacles.For each CSI receiver,CSI data is collected and marked in its LOS area and NLOS area respectively,and the classification model is trained using the marked CSI data.In the online phase,the system uses the trained classification model to determine the area range of the transmitter relative to each receiver first,and then obtains the reference point fingerprint set within the area based on the area range.Take the intersection of the fingerprint sets of the transmitting end and the reference points of each receiving end to obtain a new reference point fingerprint set.This new reference point set constitutes the current temporary fingerprint library of the transmitting end,and the KWNN algorithm is used to calculate the transmitting end's Position coordinates.Experiments show that the classification algorithm based on neural network and SVM is better than the KNN algorithm in distinguishing the range of undetermined points.The fingerprint matching algorithm of partition + KWNN is smaller than the KWNN algorithm and the K-Means + WKNN algorithm in the average positioning error and the median positioning error.