A high-precision fusion positioning method for smartphone Wi-Fi RTT / RSS / map

Abstract

The present invention discloses a high-precision fusion positioning method for Wi-Fi RTT / RSS / map of a smart phone, comprising the following steps: S1, using BP neural network to construct an RTT ranging error compensation model; designing the optimal Wi-Fi AP layout strategy, maximizing the positioning performance of a limited number of APs, and using weighted adaptive least squares method to obtain RTT ranging positioning results; S2, through the Wi-Fi signal path loss model, constructing an artificial Wi-Fi RSS / RTT fingerprint database based on indoor grid map data and Wi-Fi base station location, introducing K-Means clustering analysis algorithm, and using weighted K nearest neighbors to carry out fingerprint positioning; S3, using particle filtering to fuse the two positioning results; introducing map data to carry out "point constraints" and "line constraints" to obtain fusion positioning results. The present invention can improve the stability and accuracy of Wi-Fi FTM ranging positioning.

Description

Technical Field

[0001] The present invention relates to the field of indoor positioning technology, and in particular to a smart phone Wi-Fi RTT / RSS / map high-precision fusion positioning method. Background Art

[0002] With the continuous development of wireless communications and location-based services (LBS), the demand for high-precision indoor positioning in the fields of smart cities, Internet of Things technology, emergency rescue, security monitoring, etc. is increasing day by day. It has gradually become a necessary requirement for people to locate and navigate inside buildings such as shopping malls, airports, and hospitals. However, due to the complex indoor environment, the Global Positioning System (GPS) cannot provide high-precision positioning services indoors, which in turn promotes the development of indoor positioning technology. After years of research, indoor positioning methods based on infrared signals, radio frequency identification (RFID), ultra-wideband (UWB), Bluetooth, and geomagnetism have become relatively mature. These positioning technologies have their own advantages, and the positioning accuracy is also different. In addition, positioning technologies such as UWB and RFID have poor mobile phone compatibility, and it is difficult to balance cost and accuracy.

[0003] Wi-Fi positioning has the advantages of low cost, easy implementation, and high mobile phone compatibility. The location of the mobile device can be estimated by measuring the Wi-Fi signal received strength (RSS) and signal flight time. Under the latest Wi-Fi Fine Time Measurement (FTM) protocol, users can directly measure the round-trip time (RTT) of the Wi-Fi signal to estimate the distance between the mobile phone and the Wi-Fi base station. Currently, there are many smartphones and Wi-Fi base stations on the market that support the FTM protocol, which can be used for Wi-Fi FTM positioning. Google has also opened the development API of Wi-Fi FTM, and related positioning programs can be developed based on smartphones. Although the smartphone Wi-Fi Fine Time Measurement (FTM) technology has brought new solutions to indoor location estimation, its positioning accuracy and stability are greatly reduced by the multipath effect, non-line-of-sight interference, etc., which brings great challenges to the smartphone Wi-Fi FTM positioning application. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a high-precision fusion positioning method for Wi-Fi RTT / RSS / map of a smart phone, which realizes the organic integration of Wi-Fi RTT ranging positioning, RTT / RSS hybrid fingerprint positioning and map information, and ensures the continuity and accuracy of Wi-Fi positioning.

[0005] Technical solution: A high-precision fusion positioning method for Wi-Fi RTT / RSS / map of a smart phone, comprising the following steps:

[0006] S1, uses BP neural network to build RTT ranging error compensation model; designs the optimal Wi-Fi AP layout strategy to maximize the positioning performance of a limited number of APs, and uses weighted adaptive least squares method to obtain RTT ranging positioning results;

[0007] S2, through the Wi-Fi signal path loss model, based on indoor grid map data and Wi-Fi base station locations, build an artificial Wi-Fi RSS / RTT fingerprint database, introduce the K-Means clustering analysis algorithm to cluster the fingerprint database, and use weighted K nearest neighbors to perform fingerprint positioning;

[0008] S3, based on the positioning results of steps S1 and S2, particle filtering is used to fuse the two positioning results; map data is introduced to carry out "point constraints" and "line constraints" to obtain a fused positioning result.

[0009] Further, the detailed implementation steps of step S1 are as follows:

[0010] S11, assuming that at a certain point S, the RTT ranging data of the point relative to q Wi-Fi APs is measured The theoretical distance between this point and the Wi-Fi AP is d = {d1, ..., d i , …, d q}, i = 1, ..., q, d i represents the theoretical plane distance of the point relative to the i-th AP, then the ranging error e rtt The expression is:

[0011]

[0012] Among them, e q represents the ranging error of the point relative to the qth AP,

[0013] If a point S measures N pieces of Wi-Fi RTT data relative to q Wi-Fi networks, then N pieces of ranging error data can be obtained:

[0014]

[0015] Take the average of the error data in the above formula and get the average ranging error at point S The error model is constructed using "RTT ranging data → point number → error vector". The point number is used as the index to establish the connection between the measurement data and the point position using the RTT data measured at the point position, and the error compensation value is indexed by the point position information. If a point S obtains N pieces of Wi-Fi RTT data relative to q Wi-Fi RTT data, the mapping relationship of "RTT ranging data → point number → error vector" is expressed as:

[0016]

[0017] in, Indicates the distance between the midpoint S of the Nth data and the qth Wi-Fi AP; d Ni Indicates the distance between the midpoint S of the Nth data and the i-th Wi-Fi AP;

[0018] For all points in the area, the relationship between RTT measurement data and error data is constructed according to the above formula to obtain the data set used for BP neural network training;

[0019] S12, inputting the data set obtained in step S11 into the BP neural network, returning the error index number, retrieving the error compensation value corresponding to the index number, and calibrating the real-time measured RTT data;

[0020] S13, calculating the variance based on the N pieces of Wi-Fi ranging data of a certain point S obtained in step S11:

[0021]

[0022]

[0023] in, rss ji represents the jth Wi-Fi distance measurement and RSS value, Indicates the ranging and RSS average of N Wi-Fi data;

[0024] Collect RTT / RSS data at different distances, count the distance and variance data, and use the linear fitting method to simulate the relationship between distance and RTT / RSS data variance:

[0025]

[0026] Among them, k rtt , ε represents the fitting coefficient of RTT ranging variance simulation, k rss and δ represent the fitting coefficients of the RSS measurement variance simulation;

[0027] Based on the variance data, the Cramer-Rao bounds of each point in the positioning area are calculated, and the mean function of the Cramer-Rao bounds of each point in the area is used as the fitness function of the genetic algorithm. The optimal AP layout is obtained based on the genetic algorithm.

[0028] Design Wi-Fi data quality weight ω i The expression is:

[0029]

[0030] in, represents the distance relative to the i-th AP, and q represents the number of Wi-Fi APs;

[0031] During the positioning process, the weight of Wi-Fi measurement data is adaptively adjusted, and the least square positioning method is applied to finally obtain the Wi-Fi RTT ranging positioning result.

[0032] Furthermore, the Cramer-Rao bounds of each point in the positioning area are expressed as follows:

[0033]

[0034] Where I represents the Fisher information matrix, and tr() represents obtaining the trace of the matrix;

[0035]

[0036]

[0037]

[0038] Where q represents the number of APs; (x, y), (x i ,y i ) are the coordinates of the reference point and the i-th AP respectively; are the measurement variances of RTT and RSS relative to the i-th AP, β i is the path loss coefficient of the ith AP.

[0039] Further, the detailed implementation steps of step S2 are as follows:

[0040] Step S21, establishing a Wi-Fi signal path loss model:

[0041] R i =R0+10β i log 10 (D i )+τ

[0042] Among them, R i , R0 are the distances D iand Wi-Fi RSS value at D0, β i is the path loss coefficient, τ represents the random measurement error;

[0043] Step S22, based on the indoor grid map data and the Wi-Fi base station locations, calculate the plane distance of the indoor reference point relative to each base station:

[0044]

[0045] Among them, (X, Y), (x i ,y i ) are the reference point coordinates and base station coordinates respectively. Substitute the calculated plane distance into the path loss model to obtain the RSS value at the reference point. Calculate the plane distance and RSS data of the reference point relative to q APs respectively. Combined with the reference point coordinates, the RTT / RSS fingerprint at this reference point is obtained as follows:

[0046]

[0047] in, Respectively represent the distance and RSS data of the reference point relative to the qth AP;

[0048] Calculate all reference points in the area according to the above formula to obtain the RTT / RSS artificial fingerprint library of the positioning area;

[0049] Step S23, using K-Means clustering algorithm to cluster fingerprint data with similar spatial distribution into one category, and obtain several cluster centers; in the online matching and positioning stage, first obtain the cluster center that is most similar to the spatial characteristics of the measured data, and perform fingerprint positioning with the local fingerprint library corresponding to the cluster center; use the weighted K nearest neighbor algorithm to perform fingerprint matching, dynamically adjust the K value during positioning, and use the weighted sum of similar fingerprint position coordinates as the matching result:

[0050]

[0051] in, is the estimated fingerprint position, P i is the position of similar fingerprint i, ξ i is the weight value,

[0052] Further, the detailed implementation steps of step S3 are as follows:

[0053] S31, when positioning for the first time, the RTT / RSS fingerprint matching result is used; after the system starts running, the RTT ranging positioning result output by step S1 is used as the system state of the particle filter model, and the particle weight is updated by the RSS / RTT fingerprint matching of step S2; at the same time, a sliding window of length M is initialized to store the estimated system position, using the "first in, first out" principle;

[0054] Assume that the number of particles is N, and the initial weight of particles is 1 / N, then a particle α i The coded form is as follows:

[0055]

[0056] in, and Respectively represent the distance and Wi-Fi RSS value used to encode particles, and Will be used to update particle weights, η i represents the particle weight;

[0057] S32, after loading the map data, according to the relationship between the particle swarm after weight update, the system state position and the indoor building structure, the particles that "penetrate the wall" or "are in abnormal position" in the particle swarm are dynamically eliminated, and the particle filter estimation fusion positioning is continued;

[0058] S33, after the system starts executing for a certain period of time, if the position sequence is filled in the sliding window, a dynamic time warping algorithm is used to evaluate the similarity between the estimated position sequence and the map path, dynamically adjust the real-time position trajectory, and output the front end position of the adjusted trajectory as the fusion positioning result.

[0059] Compared with the prior art, the present invention has the following significant effects:

[0060] 1. The present invention constructs a high-precision fusion positioning model of Wi-Fi RTT / RSS / map for smartphones based on particle filtering, realizes the organic integration of Wi-Fi RTT ranging positioning, RTT / RSS hybrid fingerprint positioning and map information, and can provide a feasible solution for improving the stability and accuracy of Wi-Fi FTM ranging positioning and developing a fusion positioning model for smartphones;

[0061] 2. The Wi-Fi and map data used in the present invention are more adaptable to mobile phones and can be used to develop mobile phone Wi-Fi indoor positioning systems. Unlike other high-precision indoor positioning methods that require additional deployment of positioning base stations, the present invention can perform high-precision positioning using only Wi-Fi devices in the positioning environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1It is a flow chart of the present invention. DETAILED DESCRIPTION

[0063] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0064] like Figure 1 As shown, the smart phone Wi-Fi RTT / RSS / map high-precision fusion positioning method of the present invention includes the following specific implementation steps:

[0065] Step 1: Learning model-driven Wi-Fi RTT ranging positioning;

[0066] The BP neural network is used to construct the RTT ranging error compensation model to improve the RTT ranging accuracy. The optimal Wi-Fi AP (AP, Access Point) layout strategy is designed to maximize the positioning performance of a limited number of APs, and the weighted adaptive least squares method is used to carry out RTT ranging positioning.

[0067] Step 11, constructing a training data set for the RTT error compensation model;

[0068] Assume that the number of Wi-Fi APs is q, and the RTT distance measurement data of a point S relative to the Wi-Fi AP is measured. The theoretical distance between this point and the Wi-Fi AP is d = {d1, ..., d i , …, d q}, i = 1, ..., q, d i represents the plane distance of the point relative to the i-th AP, then the ranging error e rtt Calculate as follows:

[0069]

[0070] Among them, e q represents the ranging error of the point relative to the qth AP,

[0071] If a point S measures N RTT data relative to q Wi-Fi APs, N ranging error data can be obtained:

[0072]

[0073] Taking the average of the error data in the above formula, we can get the average ranging error at point S: To facilitate model training, the present invention uses the following logical relationship "RTT ranging data → point number → error vector" to construct an RTT error compensation model; using the point number as an index, the RTT data measured at the point is used to establish a connection between the measurement data and the point, and the error compensation value is indexed by the point information. If a point S obtains N RTT data of the point relative to q Wi-Fi APs, the mapping relationship between the RTT ranging data, point number and error vector can be expressed as:

[0074]

[0075] in, Indicates the distance between the midpoint S of the Nth data and the qth Wi-Fi AP; d Ni Indicates the distance between the midpoint S of the Nth data and the i-th Wi-Fi AP.

[0076] For all points in the area, the relationship between RTT measurement data and error data is constructed according to the above formula to obtain the data set used for BP neural network training.

[0077] When constructing the RTT ranging data training set, the amount of data for each point should be no less than 1000. If the number of Wi-Fi APs is q, the data dimension for training at each point is q×1000. Therefore, the input layer dimension of the BP neural network should also be consistent with the number of Wi-Fi APs.

[0078] Step 12: RTT ranging error model training;

[0079] Using the data set constructed in step 11, the BP neural network is used to train the RTT ranging error model, and an optimized gradient descent method is designed to dynamically adjust the learning rate to improve the performance of the BP neural network. During online ranging compensation, the real-time measured RTT data is first input into the BP neural network, the error index number is returned, the error compensation value corresponding to the index number is retrieved, and the real-time measured RTT data is calibrated;

[0080] Step 13, optimal node layout design;

[0081] Based on the Wi-Fi ranging data of a certain point obtained in step 11, the variance can be calculated:

[0082]

[0083]

[0084] in, and rss ji represents the distance measurement and RSS value of the jth Wi-Fi data, j = 1, 2, ... N, and Indicates the ranging and RSS average of N Wi-Fi data.

[0085] Collect RTT / RSS data at different distances, count the distance and variance data, and use the linear fitting method to simulate the relationship between distance and RTT / RSS data variance:

[0086]

[0087] Among them, k rtt , ε represents the fitting coefficient of RTT ranging variance simulation, k rss and δ represent the fitting coefficients of the RSS measurement variance simulation.

[0088] Based on the variance data, the Cramer-Rao bound (CRB) of each point in the positioning area is calculated, and the mean function of the Cramer-Rao bound of each point in the area is used as the fitness function of the genetic algorithm. The optimal AP layout is obtained based on the genetic algorithm.

[0089] The Cramer-Rao bound calculates the theoretical positioning accuracy of a point based on the variance characteristics of the RTT / RSS measurement data in the positioning area. Therefore, under the optimal AP layout, the sum of the Cramer-Rao bounds of all test points will take the minimum value, which is also the basis for the convergence of the genetic algorithm. The Cramer-Rao bound calculation method for Wi-Fi RTT / RSS hybrid positioning is:

[0090]

[0091] Where I represents the Fisher information matrix, tr() represents the trace of the matrix, and I xx , I yy , I xy , I yx The calculation is as follows:

[0092]

[0093]

[0094]

[0095] Where q represents the number of APs; (x, y), (x i ,y i ) are the coordinates of the reference point and the i-th AP respectively; are the measurement variances of RTT and RSS relative to the i-th AP, which can be calculated by the variance simulation formula, β i is the path loss coefficient of the ith AP.

[0096] Based on the ranging error compensation in step 11 and the optimal AP layout in step 12, the Wi-Fi data quality weight ω is designed. i , the expression is:

[0097]

[0098] in, Represents the distance relative to the i-th AP.

[0099] According to formula (6), the measurement data weight is adaptively adjusted during the positioning process, and the least square positioning method is applied to finally obtain the RTT ranging positioning result.

[0100] Step 2: Wi-Fi RSS / RTT fingerprint positioning based on cluster analysis;

[0101] Considering that RTT and RSS data can be measured synchronously and taking into account the human and material investment in offline fingerprint database construction, the Wi-Fi signal path loss model and mathematical calculation method (Formula (8) calculates the plane distance and Formula (7) inverts Wi-Fi RSS data) are used to construct an artificial Wi-Fi RSS / RTT fingerprint database. The K-Means clustering analysis algorithm is introduced to cluster the fingerprint database to reduce the occurrence of false matching during fingerprint positioning. The weighted K nearest neighbor (WKNN) is used to carry out fingerprint positioning.

[0102] Step 21: Analyze the source of RSS data error and compensation method, incorporate factors such as dielectric material, dynamic and static motion into the path loss model parameter optimization, design a path loss model parameter optimization strategy, and consider the number of Wi-Fi APs and hardware differences in actual scenarios to establish a Wi-Fi signal path loss model:

[0103] R i =R0+10β i log 10 (D i )+τ (7)

[0104] Among them, R i , R0 is the distance D i and Wi-Fi RSS value at D0, β i is the path loss coefficient, and τ represents the random measurement error.

[0105] In this embodiment, it is necessary to fit the Wi-Fi signal path loss model of all APs according to the RSS data to reduce the RSS inversion inaccuracy problem caused by device heterogeneity.

[0106] Step 22: Calculate the plane distance of the indoor reference point relative to each base station based on the indoor grid map data and the Wi-Fi base station location.

[0107]

[0108] Among them, (X, Y), (x i ,y i ) are the reference point coordinates and base station coordinates respectively. Substitute the calculated plane distance into the path loss model to obtain the RSS value at the reference point. Calculate the plane distance and RSS data of the reference point relative to q APs respectively. Combined with the reference point coordinates, the RTT / RSS fingerprint η at the reference point can be obtained as:

[0109]

[0110] in, They represent the distance and RSS data of the reference point relative to the qth AP respectively.

[0111] Calculate all reference points in the area according to formulas (8) and (9) to obtain the RTT / RSS artificial fingerprint library of the positioning area.

[0112] Step 23, fingerprint library cluster analysis and matching;

[0113] The K-Means clustering algorithm is used to cluster fingerprint data with similar spatial distribution into one category to obtain several cluster centers. In the online matching and positioning stage, the cluster center that is most similar to the spatial characteristics of the measured data is first obtained, and the fingerprint positioning is carried out using the local fingerprint library corresponding to the cluster center. The weighted K nearest neighbor algorithm (WKNN) is used for fingerprint matching. The K value is dynamically adjusted during positioning to reduce the error value introduced by abnormal fingerprint matching results to the position estimation, and the weighted sum of similar fingerprint position coordinates is used as the matching result:

[0114]

[0115] in, is the estimated fingerprint position, P i is the position of similar fingerprint i, ξ i is the weight value,

[0116] Step 3: Map / RTT / RSS fusion positioning based on particle filtering;

[0117] Based on the positioning results of step 1 and step 2, particle filtering is used to fuse the two positioning results, and map data is introduced to carry out "point constraints" and "line constraints". The fused positioning results are optimized from two aspects: single positioning result constraints and continuous path optimization.

[0118] Step 31, particle filter fusion;

[0119] When the system is positioned for the first time, the RTT / RSS fingerprint matching result is used. After the system starts running, the RTT ranging positioning result output in step 1 is used as the system state of the particle filter model, and the particle weight is updated with the RSS / RTT fingerprint matching in step 2. At the same time, a sliding window of length M is initialized to store the estimated system position, using the "first in, first out" principle.

[0120] The particle initialization mode needs to be consistent with the fingerprint composition, and the position is randomly generated in the positioning area. Assuming that the number of particle sets is N and the particle initialization weight is 1 / N, then a particle α i The encoding form can be expressed as follows:

[0121]

[0122] in, and Respectively represent the distance and Wi-Fi RSS value used to encode particles, and Will be used to update particle weights, η i Represents the particle weight.

[0123] Step 32, importing map data and executing point constraints;

[0124] After loading the map data, based on the relationship between the particle swarm, system position and indoor building structure after weight update, the particles that "penetrate walls" or have "abnormal positions" in the particle swarm are dynamically removed, and the particle filter estimation fusion positioning is continued.

[0125] Step 33, combining the map path data and executing the line constraint;

[0126] After the system starts executing for a few moments, if the position sequence is filled in the sliding window, the dynamic time warping algorithm (DTW) is used to evaluate the similarity between the estimated position sequence and the map path, dynamically adjust the real-time position trajectory, and output the front end position of the adjusted trajectory as the fusion positioning result.

[0127] It should be noted that in this embodiment, the order of step 1 and step 2 has no priority. The work of the two steps includes ranging and RSS data collection, artificial fingerprint library construction, RTT ranging error model training, and RSS path loss model training. The order of these four tasks has no priority. It can be understood that after the data is collected in the offline stage, the model training can be carried out synchronously, but all offline stage tasks must be completed before the online positioning starts.

Claims

1. A high-precision fusion positioning method for Wi-Fi RTT / RSS / map of smart phones, characterized in that: The steps include: S1, uses BP neural network to build RTT ranging error compensation model; designs the optimal Wi-Fi AP layout strategy to maximize the positioning performance of a limited number of APs, and uses weighted adaptive least squares method to obtain RTT ranging positioning results; S2, through the Wi-Fi signal path loss model, based on indoor grid map data and Wi-Fi base station locations, build an artificial Wi-Fi RSS / RTT fingerprint database, introduce the K-Means clustering analysis algorithm to cluster the fingerprint database, and use weighted K nearest neighbors to perform fingerprint positioning; S3, based on the positioning results of steps S1 and S2, the two positioning results are fused by using particle filtering; map data is introduced to carry out "point constraint" and "line constraint" to obtain a fused positioning result; The detailed implementation steps of step S1 are as follows: S11, assuming that at a certain point S, the RTT ranging data of the point relative to q Wi-Fi APs is measured The theoretical distance between this point and the Wi-Fi AP is d = {d1, ..., d i , …, d q }, i = 1, ..., q, d i represents the theoretical plane distance of the point relative to the i-th AP, then the ranging error e rtt The expression is: Among them, e q represents the ranging error of the point relative to the qth AP, If a point S measures N pieces of Wi-Fi RTT data relative to q Wi-Fi networks, then N pieces of ranging error data can be obtained: Take the average of the error data in the above formula and get the average ranging error at point S The error model is constructed using "RTT ranging data → point number → error vector". The point number is used as the index to establish the connection between the measurement data and the point position using the RTT data measured at the point position, and the error compensation value is indexed by the point position information. If a point S obtains N pieces of Wi-Fi RTT data relative to q Wi-Fi RTT data, the mapping relationship of "RTT ranging data → point number → error vector" is expressed as: in, Indicates the distance between the midpoint S of the Nth data and the qth Wi-Fi AP; d Ni Indicates the distance between the midpoint S of the Nth data and the i-th Wi-Fi AP; For all points in the area, the relationship between RTT measurement data and error data is constructed according to the above formula to obtain the data set used for BP neural network training; S12, inputting the data set obtained in step S11 into the BP neural network, returning the error index number, retrieving the error compensation value corresponding to the index number, and calibrating the real-time measured RTT data; S13, calculating the variance based on the N pieces of Wi-Fi ranging data of a certain point S obtained in step S11: in, rss ji Indicates j Wi-Fi distance measurement and RSS value, Indicates the ranging and RSS average of N Wi-Fi data; Collect RTT / RSS data at different distances, count the distance and variance data, and use the linear fitting method to simulate the relationship between distance and RTT / RSS data variance: Among them, k rtt , ε represents the fitting coefficient of RTT ranging variance simulation, k rss and δ represent the fitting coefficients of the RSS measurement variance simulation; Based on the variance data, the Cramer-Rao bounds of each point in the positioning area are calculated, and the mean function of the Cramer-Rao bounds of each point in the area is used as the fitness function of the genetic algorithm. The optimal AP layout is obtained based on the genetic algorithm. Design Wi-Fi data quality weight ω i The expression is: in, represents the distance relative to the i-th AP, and q represents the number of Wi-Fi APs; During the positioning process, the Wi-Fi measurement data weight is adaptively adjusted, and the least squares positioning method is applied to finally obtain the Wi-Fi RTT ranging positioning result.

2. According to the smart phone Wi-Fi RTT / RSS / map high-precision fusion positioning method of claim 1, it is characterized in that: The Cramer-Rao bounds for each point in the location region are expressed as follows: Where I represents the Fisher information matrix, and tr() represents obtaining the trace of the matrix; Where q represents the number of APs; (x, y), (x i ,y i ) are the coordinates of the reference point and the i-th AP respectively; are the measurement variances of RTT and RSS relative to the i-th AP, β i is the path loss coefficient of the ith AP.

3. According to claim 1, the smart phone Wi-Fi RTT / RSS / map high-precision fusion positioning method is characterized in that: The detailed implementation steps of step S2 are as follows: Step S21, establishing a Wi-Fi signal path loss model: R i =R0+10β i log 10 (D i )+τ Among them, R i , R0 are the distances D i and Wi-Fi RSS value at D0, β i is the path loss coefficient, τ represents the random measurement error; Step S22, based on the indoor grid map data and the Wi-Fi base station locations, calculate the plane distance of the indoor reference point relative to each base station: Among them, (X, Y), (x i ,y i ) are the reference point coordinates and base station coordinates respectively. Substitute the calculated plane distance into the path loss model to obtain the RSS value at the reference point. Calculate the plane distance and RSS data of the reference point relative to q APs respectively. Combined with the reference point coordinates, the RTT / RSS fingerprint at this reference point is obtained as follows: in, Respectively represent the distance and RSS data of the reference point relative to the qth AP; Calculate all reference points in the area according to the above formula to obtain the RTT / RSS artificial fingerprint library of the positioning area; Step S23, using K-Means clustering algorithm to cluster fingerprint data with similar spatial distribution into one category, and obtain several cluster centers; in the online matching and positioning stage, first obtain the cluster center that is most similar to the spatial characteristics of the measured data, and perform fingerprint positioning with the local fingerprint library corresponding to the cluster center; use the weighted K nearest neighbor algorithm to perform fingerprint matching, dynamically adjust the K value during positioning, and use the weighted sum of similar fingerprint position coordinates as the matching result: in, is the estimated fingerprint position, P i is the position of similar fingerprint i, ξ i is the weight value, 4. According to claim 1, the smart phone Wi-Fi RTT / RSS / map high-precision fusion positioning method is characterized in that: The detailed implementation steps of step S3 are as follows: S31, when positioning for the first time, the RTT / RSS fingerprint matching result is used; after the system starts running, the RTT ranging positioning result output by step S1 is used as the system state of the particle filter model, and the particle weight is updated by the RSS / RTT fingerprint matching of step S2; at the same time, a sliding window of length M is initialized to store the estimated system position, using the "first in, first out" principle; Assume that the number of particles in the set is N, and the particle initialization weight is 1 / N, then a particle α i The coded form is as follows: in, and Respectively represent the distance and Wi-Fi RSS value used to encode particles, and Will be used to update particle weights, η i represents the particle weight; S32, after loading the map data, according to the relationship between the particle swarm after weight update, the system state position and the indoor building structure, the particles with "penetrating the wall" or "abnormal position" in the particle swarm are dynamically eliminated, and the particle filter estimation fusion positioning is continued; S33, after the system starts executing for a certain period of time, if the position sequence is filled in the sliding window, a dynamic time warping algorithm is used to evaluate the similarity between the estimated position sequence and the map path, dynamically adjust the real-time position trajectory, and output the front end position of the adjusted trajectory as the fusion positioning result.

Images