Research On Indoor Fusion Localization Method Based On Factor Graph Model

In recent years,with the rapid development of wireless sensor networks technology,locationbased service(LBS)has received more and more attention,and related localization requirements and technologies have emerged one after another.However,due to the obstruction of the satellite signal by indoor walls,global navigation satellite system cannot achieve precise positioning indoors.Therefore,domestic and foreign scholars have conducted a lot of research in the field of indoor positioning,which has given birth to many indoor localization technologies.Wireless local area networks(WLAN)are widely distributed indoors,and smart mobile terminals are also constantly popularized,providing good conditions for the development,application and promotion of integrated navigation and positioning technology.This article studies the indoor fusion localization method,and the main contents are as follows:First,to solve the problem of low accuracy and stability of WLAN positioning caused by received signal Strength Indication(RSSI)changing over time and decays too fast with distance,this paper studies fine-grained channel state information(CSI)and neural network probability weighted fingerprinting and proposes a CSI correction localization method using Dense Net,which is termed Cor Fi.This method first uses isolation forest to eliminate abnormal CSI,and then constructs a CSI amplitude fingerprint containing time,frequency,and spatial information;the improved densely connected network is trained to establish the relationship between the CSI fingerprint and the spatial position;the fingerprint library is expanded by the generalized extended interpolation,which not only reduces the manpower cost,but also provides a high-precision interpolated fingerprint library;a correction positioning method based on the combination of neural network probability weighted fingerprinting and K nearest neighbor(KNN)is also proposed,which makes full use of CSI signal information and improves the accuracy and robustness of positioning.Second,to alleviate the problems of insufficient accuracy and low stability of indoor positioning technology due to a single positioning source,this paper studies pedestrian dead reckoning(PDR),differential barometric altimetry,and the construction of factor graph in navigation and positioning,and also proposes an indoor fusion localization method based on factor graph.First,PDR is improved by using peak detection and step point filtering methods in step detection,and Gauss-Newton iteration method to calculate logarithmic model parameters in step length estimation,and jump elimination and filtering for heading angle in direction estimation;a mutation filtering method based on generalized extended extrapolation is proposed,which uses multiple extrapolation results to enhance the robustness of judgment,and is used to filter out and replaces abrupt values in the air pressure data;a factor graph model based on Cor Fi,PDR and differential barometric altimetry is constructed,and then an indoor fusion localization method based on factor graph is proposed.Finally,after experimental verification,the average positioning error of Cor Fi proposed in this paper is reduced by 29.3% compared with the traditional neural network probability weighted fingerprinting,and it is reduced by 51.5% compared with the KNN algorithm.In addition,compared with the extended Kalman filter and the federated filter,the proposed indoor fusion localization method based on factor graph reduces the average positioning error by 42.9% and40.3%,respectively.