Research On Optimization Problem Of Smartphone Indoor Hybrid Positioning Using Wi-Fi RTT And PDR

Using wireless fidelity(Wi-Fi)round-trip time(RTT)and pedestrian dead reckoning(PDR)to realize sub-meter continuous hybrid positioning on smartphones is one of the most concerned research directions in the indoor positioning field.However,the complex and changeable indoor environment and flexible and diverse pedestrian activities affect the accuracy and stability of the Wi-Fi RTT/PDR indoor hybrid positioning.Therefore,it is necessary to optimize the Wi-Fi RTT/PDR indoor hybrid positioning algorithm,including non-line of sight(NLOS)and line of sight(LOS)distance measurement recognition,LOS range error compensation,improvement of RTT localization algorithm,optimization of complex pedestrian activity recognition algorithm,optimization of PDR localization algorithm and hybrid positioning algorithm,etc.,improving the scenario adaptability of hybrid localization algorithm.Funded by the National Key Research and Development Program "Indoor Hybrid Intelligent Positioning Technology"(2016YFB0502102),this thesis researched the optimization problem of smartphone Wi-Fi RTT/PDR hybrid positioning to improve the positioning accuracy and stability of Wi-Fi RTT,and enhance the robustness of PDR positioning,and improve the positioning accuracy of PDR,realizing the optimized adaptive RTT/PDR hybrid positioning.The main contributions are as follows:(1)To identify NLOS and LOS distances,the characteristics of RSS and distance measurements in NLOS and LOS environments are analyzed,and the NLOS and LOS recognition model based on the support vector machine is constructed.To obtain the available NLOS distance,the distribution characteristics of trusted distance and RSS in the NLOS environment are discussed,and a trusted NLOS distance recognition model is constructed.To improve the precision of LOS ranges,the error distribution of LOS distances is studied,and a LOS range compensation method based on least squares fitting is proposed.The experimental results show that the accuracy of LOS distance is improved by 40.96% after error compensation,and the average two-dimensional positioning error is reduced by 1.563 meters after the trusted NLOS distances participate in the positioning calculation.Based on the NLOS and LOS recognition,trusted NLOS distance identification,and LOS range compensation to obtain the available distance for positioning calculation,the RTT positioning effect is improved,and two-dimensional positioning accuracy and stability improved by 18.51% and29.27%,respectively,compared with the LS based on LOS.(2)Aiming at the difficulties in solving the three-dimensional position of smartphones,and the limited computing power of smartphones,a density-based spatial clustering of applications with noise(DBSCAN)assisted particle swarm optimization positioning algorithm is proposed,which realizes the solution of the three-dimensional position.A fast acquisition method of the three-dimensional search interval is proposed to quickly obtain the search area of the particle swarm optimization algorithm.A DBSCAN-assisted particle update strategy is proposed,which effectively reduces the computation amount of the particle swarm optimization algorithm.The experimental results show the computational efficiency improve by 84.43% after using the DBSCAN-assisted particle update strategy,and the mean two-dimensional positioning error of the proposed algorithm is 1.02 meters,the average height estimation error is0.317 meters,and the mean three-dimensional positioning error is 1.12 meters,which improves by 54.1%,compared to LS algorithm.(3)Aiming at the problems of RTT range measurements with large errors,unsatisfactory positioning accuracy,and difficult height estimation in complex scenarios,the error characteristics of range measurements in the complex scenario are analyzed,and the necessity of algorithm improvement is discussed,and the spatial relationship between range measurements is studied,and the concept of ranging difference fingerprint and the economical construction method of ranging difference fingerprint database are proposed.A high-precision two-dimensional RTT positioning model for complex scenes is constructed based on gaussian process regression,which obtains the two-dimensional positioning precision of 1.097 meters,which is improved by 68.51%,47.67% and 25.48%,respectively,compared with LS,NN and DBSCANassisted particle swarm optimization positioning algorithm.The height estimation is realized by combining DBSCAN-assisted particle swarm optimization localization algorithm,and the mean height estimation error reduces by 0.131 meters,compared with the DBSCAN-assisted particle swarm optimization positioning algorithm.(4)Aiming at imperfect pedestrian activity definition,poor activity recognition accuracy and general applicability of recognition method,the definition and classification of pedestrian activities are discussed and analyzed,and the optimal classification characteristics of smartphone modes and pedestrian movement patterns are studied,and the framework of pedestrian activity recognition method is constructed based on hierarchical classification idea.A pedestrian activity recognition algorithm based on the particle swarm optimized extreme learning machine is proposed.The smartphone mode recognition model and pedestrian movement pattern recognition model are established to realize the high-precision recognition of 24 pedestrian activities.The results show that the identification accuracy of the 24 activities is 99.91%,which is 5.04%,4.36% and 2.21% higher than those of the random forest,decision tree,and KNN,respectively.(5)Aiming at the problems of different acceleration noise,low gait recognition accuracy,and large step estimation error under various pedestrian activities,an adaptive acceleration filtering algorithm is proposed,and the gait recognition model based on the finite state machine is improved,and the construction characteristics of the step model are analyzed to improve the step-length estimation model,and the gait recognition and step estimation models for different pedestrian activities are optimized based on the genetic algorithm.Aiming at the problem of unsatisfactory heading estimation accuracy,a heading estimation algorithm optimized by the genetic algorithm is proposed.The gradient descent algorithm is used to assist the rapid convergence of the genetic algorithm,and the heading angle is corrected by combining with pedestrian activities.After verification,the gait recognition accuracy is 98.13%,the average error of the step-length estimation model is 0.014 meters,and the heading estimation accuracy is 1.577 degrees.(6)Aiming at the problems of difficult real-time switching of RTT positioning algorithm,discontinuous height estimation and the poor accuracy of hybrid positioning using RTT and PDR,a dynamic recognition method of complex scenario is proposed,which provides the foundation for the switching of RTT positioning algorithm.The barometric differential altimetry algorithm is studied,the stability of its height estimation is analyzed,and a hybrid height estimation method based on barometric differential altimetry is proposed to solve the problem of low continuity of height estimation,and a hybrid positioning framework based on EKF is built to realize the hybrid localization of RTT and PDR.An EKF optimization method based on the genetic algorithm is proposed,which realizes the optimization selection of noise and weight,and improves the stability and accuracy of the hybrid positioning algorithm.The threedimensional average positioning error of the hybrid positioning algorithm is 0.683 meters for activity 1,and the three-dimensional average positioning error is 0.767 meters for activity 7,and the three-dimensional mean positioning error is 0.706 meters for activity 13.