Research On Indoor Pedestrian Positioning Technology Based On Unrestricted Self-Contained Sensors

In recent years,as one of the most important technologies in navigation and positioning field,indoor pedestrian positioning has gained great concerns.In locationbased services(LBS),there are many indoor pedestrian positioning applications that almost are distributed in all our fields of life,such as emergency rescue,health care,commercial information push services and so on.With the rapid development of micro electromechanical system(MEMS)and widespread popularity of smartphones,it is becoming a near-ubiquitous self-contained indoor pedestrian positioning platform.However,it also brings some significant challenges including flexible placement of sensors and various usage modes of smartphones.Therefore,without the need for additional infrastructure assistance,unrestricted self-contained sensors based indoor pedestrian positioning has become a research focus in the relevant field.Pedestrian dead reckoning(PDR)is a relative positioning technique,which computes the relative location of a pedestrian by using step detection,step length estimation and heading determination.PDR can provide means of reducing the inertial error accumulation to the positioning solution by avoiding integral operation in traditional inertial navigation system(INS).Thus PDR algorithms make it possible to achieve reliable indoor pedestrian positioning based on self-contained sensors.This thesis focuses on indoor pedestrian positioning using PDR algorithms and improves all parts of PDR algorithms in terms of accuracy and robustness.The main works of the thesis are summarized as follows:Firstly,the dead reckoning based pedestrian positioning methods are studied.Based on the survey of previous works,the positioning principle,algorithm framework and traditional method are discussed.Moreover,the advantages of existing algorithms and their drawbacks in complex situations of practical applications are analyzed.Selfcontained sensors based PDR positioning algorithms are implemented.Extensive field tests have been conducted to verify the performance of the PDR algorithm.The test results demonstrate that it is inadequate to achieve high accuracy in pedestrian positioning only based on PDR algorithm and more improvements are indispensable.Secondly,context awareness and motion recognition assisted PDR algorithms are studied.The characteristics of sensors signal in different contexts are analyzed and automatic identification of the contexts of carrying a smartphone is implemented.In consideration of daily motion modes of a pedestrian in indoor environment,several relatively common motions are discussed and recognized using inertial signal features in time and frequency domains.For indoor complex situations,a conditional random fields(CRFs)based continuous motion recognition algorithm for natural PDR is proposed.Compared with existing algorithms,the proposed algorithm can recognize continuous motions more accurately.Consequently,the motion recognition results are utilized to assist step length model training of PDR algorithm.Moreover,in view of the context awareness results,a principal component analysis(PCA)based heading determination algorithm is proposed.Above optimization algorithms provide opportunities for natural PDR system to improve the positioning performance.Thirdly,the optical flow based pedestrian step length modeling algorithm is investigated.The optical flow between sequential frames of images is obtained by adopting the dense optical flow algorithm.The translations are converted into displacements with sensor attitude acquisition that is resistant to hands shaking while walking.To estimate the displacements between frames precisely,a Gaussian kernel function based adaptive weight is applied to suppress the outliers of these displacements.Moreover,to exclude undesirable training samples and improve the performance further,a linear regression mechanism with self-pruning function is proposed in the model training process.Fourthly,the pedestrian heading determination algorithm free from the distortion of magnetic anomalies is studied.Magnetometer errors distributed in indoor environment including soft iron errors and hard iron errors are analyzed.To distinguish the trustable magnetic readings from the abnormal ones,a two-phase decision filter scheme is implemented.Furthermore,a fused heading determination algorithm with magnetic anomalies identification function is proposed.When the magnetic measurements are unreliable,the pedestrian heading is only gained from inertial sensors.Instead,the pedestrian heading is corrected periodically with reliable magnetic measurements.Therefore,the performance of pedestrian heading determination is improved by correcting the errors from inertial sensors and avoiding negative effects of magnetic anomalies.Fifthly,the vector graph assisted indoor pedestrian positioning algorithm is studied.Particle filter is adopted to implement indoor pedestrian positioning.The status updates from PDR and constraints from an indoor vector graph are fused to acquire the posterior distribution of a pedestrian's position.In our proposed particle filter algorithm,objects in vector graph including walls and furniture are chosen as constraints to limit a pedestrian's trajectory.In addition,multidimensional particles are used that contain estimated parameters of step length and heading besides position coordinates.Therefore,the PDR propagation model can be learned and corrected during the operation of proposed particle filter algorithm.To save the computational cost in intersection detection and thus improve the system tractability,map construction and optimization of the vector graph are addressed and binary particle weight is designed.In conclusion,the indoor pedestrian positioning technologies in unrestricted context and complex environment are developed in this thesis.PDR algorithm is improved and hybrid indoor positioning solution is proposed,which provide theory evidence and experimental reference for the wide application of self-contained sensors based indoor pedestrian positioning.