Research On Key Technology In The Location Sensing Of Indoor Mobile Targets

Indoor location based seirvice consists of teclhnologies that are aiming to localize targets where satellite signals are missing,including indoors and urban canyons.Indoor positioning is complicated,because of multi-path effects,low sensor accuracies,and diverse user habits.Therefore,current location-based services still suffer from poor extraction of magnetic filed fingerprints,low Wi-Fi positioning accuracy and high complexity of tracking algorithms.In order to solve these problems,the paper studies the location distinguishability and feature extraction methods of magnetic field signals,attitude-free and high precision positioning systems based on Wi-Fi and magnetic signals,and tracking positioning systems based on magnetic field and inertial measurements for poor Wi-Fi signal environments.The work is concluded as follows.(1)The distinguishability of magnetic field and the extraction of location fingerprints.Indoor magnetic fields are time stable,space distinguishable and infrastructure-free,therefore are proper location sources for high precision positioning with low-cost carriers.Aiming at the problem of insufficient position differentiation of consumer magnetometer in signal sampling,this paper analyzes the influence of electromagnetic interference,soft and hard iron effects,magnetometer measurements gain difference and axis alignment error in smartphones.It is proved that the calibration of coarse-grained soft hard iron and sensor noise are the main reasons leading to the heterogeneity and user difference of magnetic field signal sampling equipments.On this basis,in order to evaluate the location differentiation of magnetic field fingerprints,this paper presents a method to measure the distinguishability of magnetic field fingerprints,which is independent of positioning algorithms.In this paper,the distinguishaility of various magnetic field fingerprint feature-extraction methods are compared on consumer smartphones.Experiments reveal that the position fingerprint of magnetic field with high distinguishabilities has stronger position differentiation,and then improves the positioning performance.(2)Attitude-free high-precision positioning based on the fusion of Wi-Fi and magnetic field signals.Wi-Fi access points are widely deployed in modern buildings,and Wi-Fi signals generated by access points are uniquely labeled and have a small coverage.Considering the high local distinguishable of the indoor magnetic field,the fusion positioning of Wi-Fi and magnetic field and Wi-Fi is an important way to realize the continuous and high precision positioning.Aiming at the problems of volatility of Wi-Fi signals,heterogeneity of Wi-Fi and magnetic field signal,and strict limit of terminal attitudes of traditional particle filtering algorithm,this paper presents a location algorithm based on convolution neural network with positioning fingerprint-image learning.The algorithm designs a mixed position image composed of Wi-Fi and magnetic field,and realizes the classification and recognition algorithm of indoor position points.Aiming at the problem that using small dataset training model is easy to lead to over-fitting,this paper proposes a multi-phase training method,which improves the learning ability of branch positioning neural network.Experimental results show that the positioning method can achieve high precision positioning under a variety of mobile phone orientations,user behaviors and usage modes.(3)Low-cost tracking based on the fusion of inertial navigation and magnetic fields.The fusion of inertial navigation and magnetic field positioning is the key technology to realize seamless indoor positioning.Aiming at the problem of high overhead of traditional magnetic field fingerprint matching algorithm,a low-overhead fingerprint matching method based on the centroid characteristics of indoor magnetic field is proposed in this paper.In this method,the low-cost particle-weight update algorithm is realized by using the spatial low frequency characteristic of indoor magnetic field.In order to improve the positioning performance of pedestrian track inference system,this paper also improves the attitude estimation,user pace detection,step calculation and motion direction estimation algorithm of mobile phone.The improvements include a step detection algorithm based on convolution neural network to improve the accuracy of step recognition;an adaptive matching fingerprint based on Bayesian inference of attitude estimation;a step length estimation algorithm based on genetic algorithm and a weight update method based on directional filtering in order to accelerate particle convergence.Experiments reveal that the system can realize a positioning accuracy of one meter at 80%with low computational over~head.As a summary,the paper studies the distingusiability and fingerprint extraction of indoor magnetic fields,the high precision fusion positoing of Wi-Fi and magnetic fields and the low-cost positioning method of magnetic fields.The paper providesr a set of algorithms aming at positioning feature extraction,positioning system initialization and continuous positioning.The results improve the content and methods of indoor positioning research,and the theories of continuous position estimation for whole space-time carriers are enriched.