Research On Smartphone IMU/Geomagnetism/WiFi FTM Fusion Localization Problems

Smartphone-based high-accuracy indoor positioning technology is a hot topic in the navigation community,while there is still a great challenge to achieve stable,reliable,and high-precision indoor localization due to the factors like the complexity of the indoor environments,the randomness of pedestrian walking,etc.After year’s development,geomagnetic positioning(GP)and pedestrian dead reckoning(PDR)based on smartphone IMU sensors reveal great advantages in positioning investments and application scenarios but have shortcomings such as low accuracy and error accumulation.With smartphones gradually supporting the WiFi fine time measurement(FTM),WiFi FTM ranging-based localization provides a new proposal for the indoor location-based service.However,the indoor complex topologies lead to the instability and low-accuracy problem of WiFi ranging.In theory,using the indoor geomagnetic field can make up for the positioning of the areas with poor WiFi ranging,WiFi positioning can also reduce the searching space of geomagnetic positioning,and the IMU sensors fusion further improves the localization continuity.Therefore,it is of great value to study the fusion localization problems of smartphone IMU/geomagnetism/WiFi FTM.This dissertation is funded by the National Key Research Program “Indoor Hybrid Intelligent Positioning Technology” project(2016YFB0502102)and the China Scholarship Council(CSC202006420023).Theoretical and experimental studies were conducted on the problems including the geomagnetic field stability,calibration and compensation,optimized geomagnetic positioning algorithms,magnetic-assisted WiFi ranging and positioning performance improvement,smartphone IMU-based pedestrian navigation pattern recognition and the improved PDR position estimation methods,and geomagnetic/WiFi FTM/PDR multi-source fusion positioning.The main contributions are as follows:(1)The overall variation characteristic of the indoor geomagnetic field is pointed out,and the effects of different indoor dielectric materials on the stability of the geomagnetic field are analyzed.The measurement and compensation model of the discrete geomagnetic characteristics is studied and a data fluctuation compensation model for the sequential geomagnetic characteristics is proposed.Experiments show that there is a correlation between the position distinguishability of the discrete and sequential geomagnetic features and the number of the discrete geomagnetic features,the size of the fingerprint database,and the length of the sequential geomagnetic features.The accuracy of the K-nearest neighbor(KNN)-based geomagnetic matching can reach 3 meters when using five geomagnetic features with a “12 m ×12 m”geomagnetic database;when the geomagnetic sequence collected for 4 seconds is used for matching,the positioning accuracy is better than 2 meters based on the fast-dynamic time warping(FAST-DTW)algorithm.(2)The single-point magnetic positioning(SPMP)based on the enhanced mind evolutionary algorithm(EMEA)and the sequence-based magnetic positioning(SBMP)using an enhanced genetic algorithm-based extreme learning machine(EGA-ELM)are constructed,respectively.Experiments show that the EMEA-based SPMP delivers accuracy better than 2 meters,which is improved by 36.24%,24.90%,and 32.86%compared to the KNN,mean square difference(MSD),and multi-magneticfingerprint-fusion(MMFF)methods,respectively.The EGA-ELM model obtains an accuracy of 0.97 meters when the number of the hidden layer node is 200,and the model construction time is lower than 90 seconds,which is 3 times,4 times,6 times,and 12 times faster than that of the learning vector quantization(LVQ),back propagation(BP),recurrent neural network(RNN)and convolutional neural network(CNN),respectively.(3)The error distribution of the smartphone-based WiFi FTM ranging in non-line-of-sight(NLoS)conditions is analyzed.The ranging compensation method and the ranging error model based on the semi-parametric regression are constructed.The geomagnetic-aided WiFi ranging positioning optimization strategies are proposed,including the geomagnetic-assisted fine-grained WiFi position estimation method and the simultaneous WiFi ranging compensation and localization based on the geomagnetism and enhanced genetic algorithm(EGA).The experimental results in NLoS environments show that the average ranging accuracy of the semi-parametric error model-based ranging compensation method is 1.33 meters,and the ranging accuracy is improved by 30.73%,which is better than that of the parameterization methods like least squares(LS)and non-least squares(NLS).The mean localization accuracy of the geomagnetic-assisted WiFi ranging positioning is better than 1.8meters,which is improved by more than 51.7% compared to the classic LS algorithm.(4)Aiming at the problems of blurred time-domain feature boundaries of IMU sensors,algorithm complexity,and poor universality of recognition models under different pedestrian navigation modes,12 common pedestrian navigation modes are defined and the combination of the motion axis sensors data to realize pedestrian navigation pattern recognition is used.Considering the algorithm complexity of the popular machine learning models,the extreme learning machine(ELM)optimized by an enhanced genetic algorithm(EGA)is adopted and a pedestrian navigation mode recognition model based on EGA-ELM is constructed.Experiments show that when a single experimenter participates in the experiment,the navigation patterns recognition accuracy of the EGA-ELM model is 94.5%;when multiple experimenters participate in the experiment,the average recognition accuracy of the EGA-ELM model is 95%,which means that the model recognition performance is stable and accurate.(5)Aiming at the problems of step recognition which is easily affected by pedestrian activities,poor adaptation of the step length model,and inaccurate heading estimation,an adaptive step recognition algorithm considering the navigation mode,an Auto-Weinberg step length estimation model and the heading calibration method with map information and geometric constraints are proposed,respectively.The experimental results show that the over(+)/miss(-)detection rates of the adaptive gait recognition algorithm considering the navigation mode under the five walking states are 0.0%,-0.7%,+ 1.3%,+1.0%,-2.0%,the performance is better than the finite state machine(FSM)algorithm and the peak detection method;the cumulative error rate of the step length estimation under slow,normal and fast walking states of the Auto-Weinberg model is 1.1%,0.86%,4.1%,and the average cumulative error rate is2.1%,which is better than that of the Weinberg model,Shin model,and constant model;the average heading estimation accuracy of the proposed heading calibration method is better than 5 degrees,which is better than that of the heading estimation methods of Mahony complementary filtering,electronic compass.(6)Concentrating on the problems of gross location error in the geomagnetic positioning and WiFi FTM positioning,error accumulation in the PDR method,and the particle degradation in the particle filtering,an integrated geomagnetic/PDR positioning model based on EKF,a tightly coupled WiFi FTM/PDR fusion model based on EKF,and a genetic particle filter(GPF)are proposed,respectively.A GPF-based geomagnetic/PDR/WiFi FTM multi-source fusion localization model is constructed.Experiments in the NLoS environment show that the EKF-based integrated positioning models can achieve meter-level positioning accuracy.Moreover,the localization accuracy of the geomagnetic/PDR/WiFi FTM fusion model based on GPF is 0.81 meters,the root mean square error is 0.99 meters,and the positioning accuracy with a confidence level of 50% is better than 0.69 meters.This dissertation has 103 figures,39 tables,and 205 references.