Research On WiFi Indoor Positioning Algorithm Based On Multi-feature Fusion And Continuous Feature Scaling

With the advancement of social modernization and the popularity of smartphones in people's lives,Location-Based Services(LBS)are increasingly affecting our lives,people's demand for Location Based Services is increasing.Indoor positioning,as an important technology in location-based services,has received increasing attention in recent years.The complexity of the indoor environment has led to the existing positioning methods such as the Global Positioning System(GPS),which cannot achieve good positioning results indoors.Since WiFi largely covered on various occasions,such as hospitals,supermarkets,airports,etc.,employing WiFi signals for indoor positioning has become an option for people to achieve efficient,economical and simple indoor positioning.However,how to use WiFi to achieve more precise positioning and how to deal with different needs in various scenarios is an urgent problem to be solved.This article mainly works from the following two aspects:For coarse positioning(classification)problems in indoor positioning field,using the WiFi fingerprint strength information obtained by WiFi communication and other information features obtained together to carry out multi-feature fusion,instead of the traditional method using only WiFi fingerprint as features,two positioning systems based on ensemble model(XGBoost and LightGBM)and multi-feature fusion are proposed.We conducted a number of comparative experiments using real data.Experiments show that the multi-feature fusion method can improve the positioning accuracy compared with the traditional method using only WiFi fingerprint as features.Compared with the traditional random forest model,the ensemble model(XGBoost and LightGBM)can achieve better positioning effect,and we also discuss the scenarios in which they apply properly.For refine positioning(coordinate positioning)problems in indoor positioning field,in the light of the feature-scaling-based k-nearest neighbor(FS-kNN)algorithm ‘fuzzy boundary' problem,a new continuous feature-scaling model is proposed,which uses continuous weights instead of the discrete weights used in the FS-kNN,and does not need divide the entire RSSI space into the intervals.This grid less scheme avoids the difficulty of the weight selection at the common boundary of the adjacent intervals that could meet in the grid-based method of FSkNN.The proposed method performs best among the counterparts in the experiments.The robustness of the algorithm are verified.