Research On The Key Technology Of Indoor Localization Based On Single Station UWB And PDR

The rapid development of modern technology has greatly improved people’s lifestyles,and Location Based Service(LBS)based on wireless sensor technology has brought great convenience to people.According to statistics,people spend nearly 80%of their time indoors,which makes the significance and value of indoor positioning services continue to increase,and the requirements for indoor positioning are also increasing.Therefore,indoor positioning technologies based on active sensing methods such as Wi-Fi,BLE,Zigbee,RFID,UWB,Acoustic,and RFID and non-active sensing methods such as PDR and SINS have been continuously developed.Most of the active sensing based indoor positioning technologies are based on curve intersection or environmental fingerprints,and most of the non-active sensing based indoor positioning methods are based on inertial navigation.However,these positioning methods have their own limitations.Due to the complexity and instability of the indoor environment,the signal may encounter nonline-of-sight(NLOS)and multipath problems during signal propagation.This problem is often caused by the reflecting,scattering or refracting of the signal due to obstacles on the signal propagation route and may cause a larger error in signal flying time and received signal strength measurement,which may affect the indoor positioning accuracy based on the measurement of signal flying time or received signal strength.Indoor positioning technologies based on non-active sensing methods such as PDR and IMU are basically not affected by NLOS or multi-path problems,but are limited by their principles.This type of positioning method will accumulate errors over time,which will lead to a decrease in positioning accuracy.The positioning algorithms based on environmental fingerprints are generally not affected by non-line-of-sight or cumulative errors,but this method requires a stable and reliable description of the environment fingerprints and the environmental fingerprint library needs to be maintained and updated over time,which may increase the expenses.So it is difficult to completely solve the indoor positioning problem with a single positioning method。Thus,a variety of hybrid positioning algorithms have been proposed,trying to combine the advantages of various positioning algorithms.Aiming at the shortcomings of traditional positioning methods when deployed in medium and small indoor environment,the main research contents of this article are as follows:1.In view of the significant impact of NLOS phenomena on indoor positioning and the complexity of the indoor environment,which makes NLOS problems almost inevitable,this dissertation proposes a NLOS/LOS classification algorithm based on deep learning.In response to the problem that traditional NLOS/LOS algorithms either need to obtain historical data on ranging,or require manual construction of features which may result in inappropriate feature selection,or require a large number of redundant observations,the classification algorithm proposed in this dissertation combines convolutional neural networks with attention mechanisms to automatically extract appropriate features from the channel impulse response of UWB signals,and the introduction of a self-attention mechanism can effectively mine more information from data,thereby constructing a propagation environment that can effectively distinguish NLOS/LOS environments without much manual intervention,and can be further expanded to more accurately distinguish UWB signals.The designed algorithm has been tested using public datasets and datasets collected from two local locations.The experiment results show that the proposed classification algorithm can achieve better classification accuracy in distinguishing NLOS/LOS than other algorithms used for comparison,and can also achieve good recognition accuracy for different environments in multi-scene indoor environment classification based on UWB CIR information.2.Aiming at the problem of inaccurate range measurement for UWB ToA caused by environments,this paper proposes a depth learning model for ranging error estimation based on the previously proposed depth learning model for environment classification.This model can be used to estimate ranging errors under different environmental conditions,thereby improving the UWB ToA ranging accuracy.The ranging error estimation algorithm proposed in this article has been tested using locally collected UWB datasets.The experiment results show that the proposed error estimating model can accurately estimate the ranging error in different environments.After correcting the ranging error estimated by the proposed model,the average absolute error of range measurement under Los condition is 0.0339 meter,and the root mean square error is 0.0496 meter,and in the other three scenarios,the proposed error estimating model can also improve the accuracy of range measurement,while in the most difficult scenarios when the wall corners block the line of sight,the proposed model can also be applied to improve the ranging accuracy by estimating the ranging error.Experiments have proved the effectiveness of the proposed algorithm.3.Aiming at the problem of lacking enough UWB node in medium and small indoor environments for traditional indoor position algorithms and the decrease of positioning accuracy over time when using PDR based indoor positioning algorithm,a hybrid indoor positioning algorithm combining UWB ToA ranging from a single UWB node and PDR is proposed.The algorithm includes correcting UWB ToA ranging and combining corrected UWB ranging information with PDR for indoor positioning.First the indoor environment of the UWB ToA range measurement will be detected using the proposed indoor environment classification algorithm and then the UWB ToA ranging measurement will be corrected using the ranging error estimated by the proposed ranging error estimating model.After that,a extend Kalman filter is applied the corrected range measurement from the single UWB node is used as a constraint to fuse with the PDR for indoor positioning.Experiments shows that the proposed hybrid indoor positioning algorithm can effectively improve positioning accuracy even in severe NLOS situations,with the root mean square error of positioning increased from about 4 meter when only using PDR to about 1 meter.In the worst case,the root mean square error of PDR positioning reaches 18.530 meter while the proposed hybrid positioning algorithm can achieve a positioning accuracy of 3.397 meter.Experiments shows that the proposed hybrid indoor positioning algorithm based on the fusion of PDR and corrected range measurements from a single UWB node can provide more accurate positioning results compared to indoor location algorithms using a single positioning technology.