| In the fields of navigation services,mobile social,public safety,and smart city construction,the determination of location information has become an important element for people to obtain ubiquitous services.Thanks to the global deployment of satellite systems such as Beidou,outdoor target positioning has become relatively mature location information source with a wide range of applications.However,the occlusion sensitivity of satellite signals makes it unable to provide a strong guarantee for indoor positioning.Therefore,wireless indoor positioning technology has become one of the current research hotspots in the field of network communication.The method based on Wi-Fi signal has accumulated many years of research in the field of indoor positioning.However,in modern multi-floor buildings such as large-scale shopping malls,hospitals,and teaching buildings,it is difficult to meet the accuracy requirements solely relying on Wi-Fi positioning methods.Especially in multi-floor buildings designed with a hollow cylindrical structure system,wireless electromagnetic signals represented by Wi-Fi establish more transmission paths between layers,which leads to positioning failures.Although the introduction of barometers can solve the problem of determining the position between floors of multi-floor buildings to a certain extent,the effect is not good in hollow cylindrical structures.In the intra-story positioning of buildings,due to the large area,the currently widely used method based on the Received Signal Strength(RSS)fingerprint database has a huge overhead in the construction of the sample database,and is also affected by the intensity of the flow of people in the application,which makes it difficult to deploy in practical applications.This paper is based on the practical application requirements of multi-floor indoor positioning with modern hollow cylindrical structures,and aims at the problems of large positioning errors caused by the influence of inter-layer signal paths in the multi-floor buildings based on the fingerprint database and the excessively high cost of deployment and limited adaptability in multifloor buildings with hollow cylindrical structure.The Multi-floor Indoor Localization based on Machine Learning(MIL-ML)is proposed.In the inter-layer positioning stage,a magnetic fingerprint database based on Support Vector Machine(SVM)is constructed,and a method for recognizing inter-layer activity patterns is proposed to realize the interlayer positioning of buildings;In the intra-layer positioning stage,a positioning method based on Cooperative Random Forest(Co-Forest)is proposed,which uses a large number of samples without location labels and a small number of samples with location labels to establish and repeatedly optimize random forest ensemble classifiers to achieve low-cost intra-layer positioning.The research results show that the proposed MIL-ML multi-floor positioning method can take into account low acquisition cost,low maintenance cost,high robustness and high positioning accuracy. |
