Indoor MEMS/UWB Integrated Localization Methods Based On Heavy-tailed Non-Gaussian Noise Model

The integrated positioning system based on micro electro mechanical systems(MEMS)/ultra wideband(UWB)has been widely used,such as indoor localization.In the line-of-sight(LOS)scenario,where the measurement noise obeys a Gaussian distribuiton,Kalman filter for the integrated MEMS/UWB indoor localization system can obtain the optimal localization accuracy in terms of minimum mean square error(MMSE).However,the measurement noise from the coupled MEMS/UWB indoor localization method in the non-line-of-sight(NLOS)scenario may has the heavy-tailed non-Gaussian characteristic,which will reduce the localizaiton accuracy of Kalman filter based indoor localization.To address the above problem this dissertation focuses on the indoor localization method with modeling the heavy-tailed nonGaussian distributed measurement noise,and improved loose and tightly coupled MEMS/UWB indoor localization methods are respectively proposed and the effectiveness and superiority of the proposed methods are demonstrated in the application of indoor localization.The main work of this dissertation is as follows:1.The coordinate frames definition,state and measurement error model in MEMS/UWB integrated localization method are given.The statistical characteristics of measurement noise in the loose coupled and tightly coupled MEMS/UWB localization methods under LOS/NLOS conditioan are analyzed by the actual data of MEMS/UWB integrated localization system,.The heavy-tailed no-Gaussian statistical characteristic of measurement noise in the loose coupled localization method and the skewed and skewed and non-Gaussian statistical characteristic of measurement noise in the tightly coupled loacalization method under the condition of NLOS are obtained2.The loose coupled MEMS/UWB localization method based on the Gaussian-Student’s t mixture(GSTM)distributed model and the tightly coupled MEMS/UWB localization method based on Student’s t-inverse Wishart(STIW)distribution model have,respectively,designed.The simulation results from MEMS/UWB loose integrated localization and tightly integrated localization show that state estimation methods based on the Student’s t distribution can obtain higher estimation accuracy then the trandional Kalman filter based on the Gaussian distribution under the NLOS condition due to the heavy-tailed characteristic of the Student’s t distribution.3.A new method based on the Normal-Bernoulli distribution model is proposed for the Gaussian/heavy-tailed distributed measurement noise caused by the LOS/NLOS errors of the MEME/UWB loose integrated indoor localization method,where the probability of Bernoulli random variable is modelled as a Beta distribution.The parameters of the proposed NormalBernoulli distribution are inferred by the variational Bayesian approach without fixed-point iterations.Simulation and experiment demonstrate that the proposed method has smaller localization error and faster operation speed than existing indoor localization methods based on the non-Gaussian state estimeation in the LOS/NLOS scenario.4.A new method based on the Normal-Categorical distribution model is proposed for the Gaussian/skewed and heavy-tailed distributed measurement noise caused by the LOS/NLOS errors of the MEMS/UWB tightly integrated indoor localization,in which the probability of Categorical random variable is modelled as a Dirichlet distribution.The measurement likelihood probability density can be adaptively adjusted in the Gaussian density preseted by the proposed method by the shape parameters of the Dirichlet distribution to better model the Gaussian/skewed and heavy-tailed distributed measurement noise.Simulation and experiment demonstrate that the proposed indoor localization method has smaller localization errors and faster operation speed than existing indoor localization methods based on the non-Gaussian state estimeation in the LOS/NLOS scenario.