Indoor Location Technology Based On The Fusion Of WiFi Fingerprint And PDR

With the extensive coverage of WiFi wireless access points and popularity of smart phones,indoor positioning technology based on mobile devices has become a hot topic in the current indoor positioning research.Indoor location based on WiFi,which has high-speed communication,low cost,high coverage rates and other advantages,has become widely used.The representative positioning algorithm of WiFi is fingerprint positioning.The inertial measurement location is the location estimation algorithm based on the built-in sensor of the mobile terminal.The representative algorithm is the Pedestrian Dead Reckoning algorithm.However,the stability of WiFi fingerprint positioning is poor.PDR algorithm needs the initial position of target and is easy to produce comulative error.A fusion algorithm based on WiFi fingerprint and PDR is proposed.On the one hand,PDR algorithm only relies on inertial sensor and can provide a more stable positioning results;On the other hand,fingerprint positioning can not only provide initial position coordinates for PDR algorithm,but also can correct the PDR positioning parameters.This paper first investigates the learning location algorithm and the matching location algorithm which all belong to the fingerprint location algorithm.Based on the research status of the learning location algorithm,the existing problems and drawbacks of WiFi fingerprint location algorithm based on machine learning are pointed out.These problems and drawbacks exist in the main modules of positioning system,including location clustering,kernel function selection,parameter setting,and algorithm complexity.The goal of this dissertation is to overcome these problems and improve the algorithms.Most of the learning location algorithm has a longer training time when the database is large,and can't realize the advantages of small sample nonlinear regression,which seriously affects the real-time and positioning accuracy of the positioning system.At the same time,the selection of the kernel function of the machine learning algorithm is limited and the parameter setting is complex,which affects the training effect of the location model.For example,Support Vector Machine(SVM)regression method as the learning location algorithm is widely used.The selection of the kernel function of SVM is limited by Mercer theorem and the penalty factor needs to be set,which make the positioning accuracy can't be improved better.In view of the above problems,a double stage location algorithm based on Region Locking and Relevance Vector Regression was put forward.The algorithm not only solves the problem of large database,but also achieves better training results.The experiment shows that the positioning accuracy is higher than the support vector regression method.Aiming at the problem of the traditional step model in the PDR algorithm can not adapt to the different walking state.And in indoor environment,the direction sensor and gyroscope can be disturbed by the external factors,which lead to relatively large errors of heading angle estimation.Therefore,this paper proposes micro-heading angle estimation methods based on micro-scene.It is shown that the improved step model has higher accuracy under different walking speed,and the heading angle estimation algorithm can improve the accuracy of direction angle measurement.Finally,this paper makes some experiments to compare the fusion location algorithm with other algorithms.Experiments show that the fusion localization algorithm can not only improve the positioning accuracy and the robustness of the algorithm,but also improve the real-time performance of the positioning system.