Research On Indoor Positioning Based On WIFI/PDR And Its Application In Seamless Navigation Technology

Nowadays,the Global Navigation Satellite System(GNSS)has basically realized the real-time and high-precision positioning in the outdoor environment.However,with the acceleration of urbanization,there has been great demand for pedestrian positioning services in terms of dense building sites disturbed by GNSS signals,and it has also driven great progress in indoor positioning technology in recent years.In this paper,the fusion research of the WiFi and PDR positioning was conducted by focusing on the basis of WiFi and PDR positioning experiments,and the indoor and outdoor scene recognition was performed combined with the common indoor and outdoor ubiquitous signals,and the feasibility of the algorithm was finally verified based on a large number of experimental data.Contents were shown as follows:I.In view of the insufficient samples and utilization of signal features in CSI fingerprint database of WiFi positioning,a positioning algorithm of using the mixed CSI amplitude and phase information based on the Convolutional Neural Network(CNN)was proposed,and the Improved-Aquila Optimizer(IAO)was adopted to optimize the CNN network structure to further improve the accuracy of WiFi positioning.In the offline stage,the S-G filter was employed for the extracted amplitude and phase signal after linear transformation to eliminate the noise data generated by the multipath effect during the acquisition;then the IAO-CNN model was built to extract the CSI signal characteristics in the database along with the optimization of CNN hyperparameter combination.In the online phase,the test point was inputed into the trained network structure for regression prediction to obtain the estimated results of the current position.According to the actual scene test,it was shown that in the positioning area of about 5×10 m~2,the average positioning accuracy of the algorithm in the paper reached 1.410 m,which increases by 54.68%and 54.04%respectively compared with AO-CNN and CNN models,and the positioning error of53%of the data points was less than 1.220 m,which effectively improved the positioning accuracy of WiFi fingerprint.II.For the existing drift error of PDR positioning technology,the improved PDR algorithm for signals from the accelerometer,gyroscope and magnetometer modules collected by smartphones was proposed in the paper.Specifically,zero-crossing detection constrained by the double threshold was adopted to detect gait and the nonlinear Weinberg model to estimate the step length,and drift error was corrected according to the heading output by both the magnetometer and the gyroscope.Finally,two different test routes were designed in two indoor environments.The experimental results showed that by using the improved PDR algorithm,the average error in the“L”experimental route with a total length of 83.4 m reached 1.129 m,and the cumulative average error reached 2.275 m after detour of 12 laps on a rectangular route of 32 m long and 31.2 m wide,which verifies the positioning effect of pedestrians walking in line and turning,and based on which,in order to further improve the positioning accuracy of single WiFi and PDR,fusion positioning model of BP neural network optimized by the Improved Whale Optimization Algorithm(IWOA)was proposed.Through experimental verification,it shows that the average positioning accuracy of using the fusion positioning model was improved by 52.88%and 3.91%respectively compared with that of the single WiFi and PDR,which better meets the accuracy requirements of indoor positioning.III.For the accuracy of indoor and outdoor scene recognition and switching remaining to be improved,an improved BP-Adaboost classification algorithm was put forward in this paper through the analysis of common indoor and outdoor ubiquitous characteristics.Specifically,combined with the IWOA-BP neural network as a weak classifier,the classification results of the strong classifier are output by the Adaboost algorithm.Finally,the analysis of accuracy of indoor and outdoor identification was performed collected by continuous collection of 10,000 of data–number of GNSS satellites,light intensity,magnetic field intensity and sound intensity at day and night.The results showed that the accuracy of scene recognition using the improved classification algorithm reached 97.8%,which improved by 2.2%and 3.2%compared with the traditional random forest algorithm and SVM algorithm respectively,meeting the demands of seamless indoor and outdoor navigation switching.Figure[51]Table[11]Reference[95]...