Research On Indoor Localization System Based On Multi-source Data Fusion And Improved Particle Filter

With the deepening development of the fourth scientific and technological revolution,the market of emerging technologies such as the Internet of Things will be widely exploited,indoor localization technology is one of them.Wi-Fi signal power can be used as a "fingerprint" to characterize geographic location information,fingerprint matching localization can be carried out based on it.However,the fluctuation of Wi-Fi signal is large due to the changeable indoor environment,so the localization accuracy is low.Geomagnetic intensity information is a stable information resource in indoor environment,and the intensity distribution of indoor geomagnetic signals is not uniform,which makes geomagnetic information can also be used as a favorable basis for distinguishing geographical location.However,geomagnetic intensity may be similar in different geographic locations.The indoor localization algorithms using geomagnetic information alone are complex and difficult to achieve.The Pedestrian Dead Reckoning localization based on the pedestrian motion characteristics is proposed to estimate the pedestrian position coordinates in a recursive form.This localization method has accumulated errors and can not achieve high-quality localization targets.Because of the obvious shortcomings of single Wi-Fi,geomagnetic and Pedestrian Dead Reckoning localization,in order to achieve complementary advantages and overcome each other's shortcomings,through the research and experimental analysis of the relevant literature,this paper regards the improved particle filter fusion localization system as the final localization scheme,and its research process is divided into the following three steps:(1)This paper extend fingerprint localization based on Wi-Fi and geomagnetism.In the offline stage,multiple Wi-Fi signals' power and geomagnetic intensity datas are collected at all reference points in the room for many times,the initial fingerprint database is generated by mean processing and Gauss fitting processing respectively,and then a rich extended fingerprint database is constructed by cubic spline interpolation processing.In the online stage,Wi-Fi and geomagnetic fingerprints of the locating points are collected,locations of locating points are estimated based on Self-adaptive Weighted K-nearest Neighbor(SA-WKNN)algorithm.Experiment shows that the extended fingerprint has better localization performance than the single fingerprint.(2)Aiming at Pedestrian Dead Reckoning localization,this paper collects acceleration,angular velocity and direction data during pedestrian walking through the inertial sensors of smart phone.According to the acceleration data,the step number and step length are estimated,the forward direction angle is estimated jointly based on the angular velocity and direction data,the current position coordinates can be estimated by combining the position coordinates of the previous step.It has been proved that the improved step length and direction estimation have achieved good results.(3)This paper uses Markov Chain Monte Carlo(MCMC)method to improve Particle Filter(PF).Based on the improved PF algorithm,the above two localization methods are fused.The extended fingerprint localization results are taken as the measurement values,the Pedestrian Dead Reckoning localization results are taken as the state values,thus the dynamic fusion localization is completed.Compared with the traditional fusion localization algorithm,the proposed algorithm can achieve higher accuracy and better stability.