Research On Aiding Information Based Indoor Inertial Pedestrian Positioning

Among many existing indoor pedestrian positioning methods,the self-contained in-ertial pedestrian positioning is a positioning means with importance in GPS-denied areas.It only relies on the readings from inertial measurement unit(IMU)mounted on the pedes-trian to reckon a pedestrian's positions,thus independent from pre-installed infrastructures and having the self-contained feature.However,due to the inherent inertial drifts,the po-sitioning accuracy and robustness deteriorate with time.In practical applications,other aiding information other than inertial measurements are incorporated for hybrid position-ing.The processing flow for such hybrid positioning can be partitioned to trajectory gen-eration phase and trajectory calibration phase.The former denotes generating a trajectory through inertial calculation and the latter denotes increasing the positioning accuracy by incorporating other aiding information.As the the trajectory generation phase is quite mature,many existing methods currently focus on the study of the trajectory calibration phase.Although these trajectory calibration methods can fuse different types of aiding in-formation to achieve relatively accurate results,they lack inner connection and overall framework.Inspired by these methods,this thesis proposes two unified frameworks for trajectory calibration.We also elaborate the two frameworks by including the five rep-resentative aiding information:ranging from UWB,signal strength from the iBeacons,position estimations from GPS,Wifi fingerprints and the perpendicular building structure information.Experiments are carried out which validates the frameworks adopting the mentioned information.The main contents and achievements are as follows:1.A particle filter based traj ectory calibration framework is proposed.The framework can fuse many types of aiding information and thus highly adaptive.The founda-tion of the framework is a particle filter.In the framework,the pose updates in the trajectory generation phase are adopted for prediction of the particles at the next epoch.Weight update strategies for the particles are designed respectively for dif-ferent aiding information.After the classical particle resampling step,the calibrated trajectory can be acquired.As the framework has the "predict-update" processing flow,real time implementation is possible.2.By transforming the fusion problem to an optimization problem,the optimization based framework is proposed.The framework try to form an error energy function adopting the information from the aiding information,including the pose updates information through inertial calculation.By minimizing the error energy function,the optimal positions of the pedestrians can be estimated under the sense of least squares.In this framework,as all measurements are needed to form the error energy function,the position are estimated in a batch and real time processing is not seem-ingly available.Normally,the results from the optimization framework is more accurate and robust than the particle filter framework.3.Other than the two frameworks,some specific methods are also proposed during the elaboration with respect to the mentioned five aiding information.For the UWB based ranging information,the noise due to NLoS are taken into consideration.For the iBeacon based information,in the particle filter based framework,the likelihood of the signal strength(calculated from a Gaussian regression process)is adopted to update the weights of particles during particle filtering.In optimization framework,the range(between the iBeacon and the user)and the variance of the range derived from the Gaussian regression process are adopted to form the error energy function.For the aiding information of Wifi fingerprints,in the particle filter based frame-work,the weights are updated according to the vicinity of the fingerprint space,which can avoid the inconsistency in traditional Gaussian process based methods and can lower the computational cost.For the optimization based framework,the method can calibrate crowd-sourced trajectories and can generate a Wifi based fin-gerprint map for commercial positioning.Compared to some dedicated and time consuming way for building the fingerprint map by surveying,this method can sig-nificantly boost the efficiency.A virtual anchor method based on building structure is proposed.This method can fully maintain the self-contained feature of pedes-trian inertial positioning.The revisits of sites and perpendicular corridor structure derived from trajectories can be adopted for compensating the inertial drifts.Fol-lowing the thought,this method can fit both the particle filter framework and the optimization framework.A pre-processing method for GPS position estimations is proposed.After pre-processing,the GPS position estimations can be adopted to calibrate trajectories under both frameworks.The pre-processing method can significantly increase the positioning accuracy.