Research On The Key Technology Of Pedestrian Autonomous Positioning

Pedestrian autonomous positioning technology based on inertial sensors does not need to rely on external infrastructure and historical training data,and only relies on sensors built-in smartphone to complete pedestrian position estimation.It has the advantages of strong autonomy,high update frequency,and full space coverage.At present,pedestrian autonomous positioning faces many challenges,such as the pedestrian walking modes and smartphone usage modes are complex and changeable;different users have obvious differences in height,weight,and walking habits;There are many sensor models,with different performances and lack of necessary calibration;sensor noise and insufficiently accurate positioning models have led to a rapid increase in the cumulative error inherent in autonomous positioning.This dissertation conducts an in-depth study on how to identify complex and changeable navigation activity modes,eliminate heading errors,eliminate step length estimation errors,and restrain position drift.The main research results and innovations of this dissertation are summarized as follows:(1)Aiming at the problem of complex and changeable pedestrian navigation activity mode,this dissertation utlizes the characteristics of magnetic signals to improve the accuracy and robustness of pedestrian step detection.on the basis of step detection,a smartphone usage mode recognition algorithm based on an ensemble classifier and a pedestrian walking mode recognition algorithm based on a multi-head convolutional attention mechanism are proposed.Ensemble classifier utlizes predictions from multiple nonlinear models to train the first-layer model and generate the second-layer training set and test set.Logistic classification is used in the secondary model to output the final smartphone usage mode.The pedestrian walking mode recognition algorithm effectively learns the timing features extracted by the convolutional neural network through the multi-headed attention mechanism,which helps to improve the recognition accuracy.(2)Aiming at the difficulty in obtaining the true value of heading and step length,this dissertation proposes a magnetic field-aided map matching algorithm.This method defines the displacement and heading of pedestrian trajectory,the variance and mean value of magnetic field sequence as observations,and defines the key point in map as hidden state of HMM,and use the Viterbi algorithm to solve the observation sequences,get the pedestrian true trajectory.Finally,the true trajectory is segmented by step events,and the heading and step length are marked.Magnetic observations bring more robust and accurate emission probability to HMM,thereby effectively reducing mismatches caused by similar indoor layouts.(3)Aiming at the problem of device heterogeneity and pedestrian heterogeneity,this dissertation proposes a personalized step-length estimation algorithm based on online active learning.The algorithm uses denoising autoencoder and LSTM to establishe a common step length model.On this basis of common model,dissertation proposes an unperceived model updating framework based on active online learning.Based on this framework,the actual walking trajectory is automatically obtained through the opportunistic map matching mechanism,and then the step-length prediction value of the common model is used as the weight to divide the pedestrian walking trajectory to obtain the displacement of each step.The displacement is used as a new label to update the common model,to obtain a more accurate personalized step length model that alleviates the influence of pedestrian heterogeneity,device heterogeneity,and different pedestrian motion modes on the accuracy of the step length estimation.(4)Aiming at the inherent cumulative error of inertial navigation,this dissertation proposes a hybrid fingerprint model with higher spatial resolution and more robustness as an indoor location signature.On this basis of the hybrid fingerprint model,this dissertation proposes single-step positioning and tracking algorithm based on the particle filter framework and the long-trajectory matching positioning algorithm to achieve high-precision pedestrian autonomous positioning without any dedicated infrastructure.(5)On the basis of pedestrian navigation activity recognition,pedestrian heading estimation,personalized step length estimation,and infrastructure-independent trajectory calibration,a cloud-based collaborative pedestrian autonomous positioning system under complex navigation activities is constructed.Experiments show that the system obtains a positioning accuracy of 1.8 meters,2.2 meters,and 3.3 meters with 75%confidence in offices,shopping malls and parking lots,respectively,laying a solid foundation for the practical application of intelligent location services.