CN109739830A - A rapid construction method of location fingerprint database based on crowdsourced data - Google Patents

Abstract

The invention discloses a method for rapidly constructing a location fingerprint database based on crowdsourced data. S1, fission building based on PDR, particle filtering and map constraints; S2, correcting PDR accumulated errors in non-explicit landmark point areas based on MDS. The present invention constructs a location fingerprint database based on the fission method and the set path effective time threshold, and uses MDS combined with implicit landmark points to correct the PDR cumulative error, and at the same time, constructs a short-distance location fingerprint correlation model based on weighted multi-dimensional WiFi numerical features; compared with the prior art The location fingerprint database can be quickly constructed, the accumulative error of the sensor is reduced, the matching accuracy of the location and the fingerprint is improved, and the problems of frequently disturbing the user and low matching accuracy of the location and the fingerprint in the prior art are effectively solved.

Description

A rapid construction method of location fingerprint database based on crowdsourced data

technical field

The invention belongs to the technical field of communications, and in particular relates to a method for rapidly constructing a location fingerprint database based on crowdsourcing data.

Background technique

There are two existing methods for building a location fingerprint database based on crowdsourcing: one is the display crowdsourcing method, which is characterized in that it can obtain relatively accurate location fingerprint information, but it will frequently disturb users, resulting in a decrease in user experience, and user experience. Inaccurate operation will cause data pollution, hindering the actual promotion and application; the other is the implicit crowdsourcing method, which can avoid the phenomenon that users are frequently disturbed, but the matching problem of location and fingerprint needs to be considered. The implicit crowdsourced location fingerprint database construction method can only roughly match location and fingerprint.

The existing method of building a location fingerprint database based on PDR combined with implicit crowdsourcing can realize the matching problem of location and fingerprint, and can quickly build a relatively complete location fingerprint database. However, there is a large cumulative error in PDR, resulting in a low matching accuracy between position and fingerprint. The existing use of characteristic landmarks to correct the cumulative error of PDR can improve the matching accuracy of some areas, but the number of characteristic landmarks is limited, and the matching accuracy cannot be effectively improved. .

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for quickly constructing a location fingerprint database based on crowdsourced data to solve the problems of frequently disturbing users and low matching accuracy of location and fingerprint in the prior art.

In order to achieve the above object, the technical scheme that the present invention takes is:

A method for rapidly constructing a location fingerprint database based on crowdsourced data, comprising:

S1. Fission library building based on PDR, particle filter and map constraints, the steps include

S11. Train a short-distance location fingerprint association model;

S12. Build an indoor map;

S13, marking explicit landmark points and collecting fingerprint data of corresponding positions;

S14. Determine whether the crowdsourcing path data is effective;

S15, based on PDR, particle filter and map constraint multiple condition constraints to associate fingerprints and location points;

S16, setting the effective time threshold of the crowdsourcing path, within the effective time threshold, if the cumulative error of the sensor data falls within the tolerable range, then proceed to step S15, otherwise, the crowdsourcing path is invalid, and proceed to step S17;

S17. Build the database in sequence based on the fission method;

S2. Correct the cumulative error of the PDR in the non-explicit landmark region based on MDS, and the steps include:

S21. Calculate the area range of the implicit landmark point based on the fingerprint similarity;

S22. Count common paths between implicit landmark point regions;

S23. Search for a relatively effective crowdsourcing path;

S24. Redraw the new path between the implicit landmark point regions;

S25. Calculate the relative coordinates of the implicit landmark point area;

S26, accurately calculate the absolute coordinates of the implicit landmark point area based on MDS;

S27. Correct the accumulated PDR error of the non-explicit landmark region.

Preferably, the method for training the short-distance location fingerprint association model in step S11 includes:

A1. Based on multi-dimensional WiFi numerical feature fingerprint distance modeling;

A2. Training of short-distance location fingerprint association model.

Preferably, the method of step A1 based on multi-dimensional WiFi numerical feature fingerprint distance modeling is:

A11. Construct a relative fingerprint composed of relative subsequences;

A12. Calculate the similarity of relative subsequence pairs;

A13. Calculate the similarity of the relative subsequence pairs of fingerprints in the relative fingerprint pair, and obtain the similarity matrix of the relative fingerprint pair by traversal calculation;

A14. Use the dynamic programming algorithm to search the similarity matrix to find the best matching relative fingerprint pair.

Preferably, the training method of the short-distance location fingerprint association model in step A2 is:

A21. Obtain crowdsourced data around known landmarks;

A22. Extract the fingerprint data of the radiation area of the landmark point according to the fingerprint similarity;

A23. Calculate the fingerprint distance of the fingerprint data of the radiation area of the landmark point;

A24. Determine a short-distance location fingerprint association model based on the MDS.

Preferably, the method for judging whether the crowdsourcing path data is effective in step S14 is:

When there is only explicit landmark data in the database, it is set that the crowdsourcing route becomes valid only when the user walks near the explicit landmark, and the short-distance location fingerprint correlation model is used to calculate the starting point when the crowdsourcing route becomes valid. physical coordinates of the location point;

Determine whether the user has walked around the explicit landmark point according to the WiFi fingerprint similarity.

Preferably, in step S15, the method for constraining the associated fingerprint and location point based on multiple conditions of PDR, particle filtering and map constraint is:

Combining PDR, particle filtering and map constraint multiple techniques to build a location fingerprint database;

According to the built-in sensor data of the smartphone, the pedestrian steps, step length and heading are deduced, the user's walking path is obtained, the physical coordinates of the data collection points on the crowdsourcing path are obtained, and the matching of the position and the fingerprint is obtained;

Accurate crowdsourcing paths are obtained by using the dual constraints of particle filtering and map constraints.

Preferably, in step S17, the method of building the database in sequence based on the fission method is as follows:

When the fingerprint data of the non-explicit location points is established in the location fingerprint database, the effective starting location of the crowdsourced path data of other users is around any known location points, and the database is built based on the fission method;

Each fingerprint point on the location fingerprint database after the fission method is established corresponds to a rough physical coordinate.

Preferably, step S21 includes setting a fingerprint similarity threshold σ _sim , and the small area where the location point where the fingerprint similarity exceeds σ _sim is located is used as the implicit landmark point area;

Step S22 includes numbering the continuous paths to obtain the number of paths passing between the two landmark point regions simultaneously;

Step S23 includes sequentially selecting the path with the smallest fluctuation of the magnetometer data as the effective path between the two landmark point regions in the common path between the two implicit landmark point regions;

Step S24 includes redrawing a new path between the implicit landmark point regions according to the acceleration sensor and magnetometer sensor data of the effective path between the implicit landmark point regions;

Step S25 includes, starting from the explicit landmark nodes, sequentially calculating the coordinates of each implicit landmark point area relative to the explicit landmark points.

Preferably, the method for accurately calculating the absolute coordinates of the implicit landmark point region in step S26 based on the MDS is:

According to the relative coordinates of each implicit landmark point area and the explicit landmark point, a relative distance matrix D is constructed,

Among them, d _ij is the relative distance between the explicit landmark point, the implicit landmark point and the center position point, i=1,2,3,...,m; j=1,2,3...,m;

Set the center point as the origin, calculate the relative coordinates of the explicit and implicit landmark points and the origin according to the MDS, and update the original relative coordinates of the implicit landmark point;

The actual physical coordinates of each implicit landmark point are obtained according to the actual physical coordinates of the explicit landmark points.

Preferably, the method for correcting the PDR accumulation error of the non-explicit landmark region in step S27 is:

Each implicit landmark point is taken as the new starting point of crowdsourcing path around the point, and a new crowdsourcing path is drawn according to the subsequent part of the original path information passing through the point, the accumulated error of PDR is corrected, and the matching accuracy of position and fingerprint is improved.

The method for quickly constructing a location fingerprint database based on crowdsourcing data provided by the present invention has the following beneficial effects:

The present invention constructs a location fingerprint database based on the fission method and the set path effective time threshold, and uses MDS combined with implicit landmark points to correct the PDR cumulative error, and at the same time, constructs a short-distance location fingerprint correlation model based on weighted multi-dimensional WiFi numerical features; compared with the prior art The location fingerprint database can be quickly constructed, the accumulative error of the sensor is reduced, the matching accuracy of the location and the fingerprint is improved, and the problems of frequently disturbing the user and low matching accuracy of the location and the fingerprint in the prior art are effectively solved.

Description of drawings

Figure 1 is a flow chart of a method for rapidly constructing a location fingerprint database based on crowdsourced data.

Figure 2 is a flowchart of training a short-distance location fingerprint association model.

Fig. 3 is a schematic diagram of building a database based on fission.

FIG. 4 is a schematic diagram of calculating the relative coordinates of the implicit landmark point area.

Detailed ways

The specific embodiments of the present invention are described below to facilitate those skilled in the art to understand the present invention, but it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes Such changes are obvious within the spirit and scope of the present invention as defined and determined by the appended claims, and all inventions and creations utilizing the inventive concept are within the scope of protection.

According to an embodiment of the present application, referring to FIG. 1 , the method for rapidly constructing a location fingerprint database based on crowdsourced data in this solution includes:

S1. Fission building library based on PDR, particle filter and map constraints, the steps include:

S11. Train a short-distance location fingerprint association model;

Referring to FIG. 2, wherein step S11 specifically includes the following two steps

A1. Based on multi-dimensional WiFi numerical feature fingerprint distance modeling;

A2. Training of short-distance location fingerprint association model.

The specific method of step A1 based on multi-dimensional WiFi numerical feature fingerprint distance modeling includes:

A11. Construct a relative fingerprint composed of relative subsequences;

With two sets of fingerprints and Extend the rss value of each AP in the fingerprint to a set of rss vectors to form relative subsequences where i,j∈{1,…,N _n }. Constitute relative fingerprints from relative subsequences: and

A12. Calculate the similarity of relative subsequence pairs;

Two sets of different relative fingerprints and make up fingerprint pairs Respectively and fingerprint relative subsequence and Among them, j∈{1,…,N _n }, i∈{1,…,N _m }, namely pairs of relative subsequences that make up the fingerprint Relative fingerprint alignment based on weighted statistics of multi-dimensional WiFi numerical features similarity of can be expressed as:

Among them, _wi is the weighting coefficient (0≤wi ≤1, _i =1,2,3,4), is the rss level matching cost, Yes Spearman correlation coefficient between , N is the total number of all different APs in N _m and N _n , Yes The total number of APs in China, Yes The number of irrelevant APs in the middle.

A13. Calculate the similarity of the relative subsequence pairs of fingerprints in the relative fingerprint pair, and obtain the similarity matrix of the relative fingerprint pair by traversal calculation;

The similarity matrix of the relative fingerprint pair represents the relative fingerprint pair in, by formed matrix; compute first medium fingerprint relative subsequence pair similarity of Calculate by traversal middle similarity can be obtained The similarity matrix of

in,

A14. Use the dynamic programming algorithm to search the similarity matrix to find the best matching relative fingerprint pair;

get the similarity matrix After that, use the dynamic programming algorithm to search exist Find the best matching path C in ; according to the best matching path C of the two relative fingerprint pairs, the similarity (or fingerprint distance) of the fingerprint pair is accumulated as:

Using the above method of constructing relative fingerprint distance can reduce the influence of equipment heterogeneity on the positioning results.

The specific method of the training of the short-distance location fingerprint association model in step A2 includes:

A21. Obtain crowdsourced data around known landmarks;

Set some known landmark points on the indoor area to be located and collect the coordinates and rss information of the corresponding position. The user's handheld device collects rss data around the landmark points and uses PDR to record the walking path; in this process, in order to avoid the existence of equipment in the training process Heterogeneity problem, the mobile phone equipment that collects data at a known landmark point is the same as the mobile phone equipment that users use to collect crowdsourced data near this point. These data are not crowdsourced data in the real database building process, but are used for training short distances. Crowdsourced data collected specifically for estimation models.

A22. Extract the fingerprint data of the radiation area of the landmark point according to the fingerprint similarity;

According to the fingerprint similarity, the fingerprint data of the radiation area of the landmark point is extracted. Since the influence of equipment heterogeneity is avoided when the training data is collected, the fingerprint similarity can be calculated according to the rss Euclidean distance between the fingerprints.

A23. Calculate the fingerprint distance of the fingerprint data of the radiation area of the landmark point;

Divide the extracted fingerprint data around the landmark points according to the same crowdsourcing path to form training data; the training data contains paired data collection points, and the matched data collection points are located on the same crowdsourcing path, and their relative physical distances and fingerprint data are known. The relative fingerprint distance between each matching data collection point is calculated by using the fingerprint distance model to form a relative fingerprint distance matrix D _wifi .

A24. Determine short-distance location fingerprint association model based on MDS

The short-range location fingerprint association model is derived from fingerprint features and can be used to estimate actual physical distances. This model can infer the physical coordinate information of another point based on the fingerprint information between two points and the physical coordinate information of one point.

The short-distance location fingerprint association model is further obtained on the basis of the fingerprint distance model, and the model is obtained by training the optimal weights 0≤wi ≤1 ( _i =1,2,3,4); through MDS ( The Multidimensional ScalingAnalysis, MDS) algorithm generates the relative coordinates between the data collection points according to the relative fingerprint distance matrix D _wifi , obtains the relative distance L between the data collection points according to the relative coordinates, and obtains the physical relative distance between the data collection points from the PDR. Distance d, assuming that d and L are approximately proportional to d=kL, this formula includes 0≤wi≤1( _i =1,2,3,4) and k four unknown parameters, through the least squares method LS (Least Square) find the optimal unknown parameters to obtain the short-distance location fingerprint correlation model, in which the same crowdsourced path location data is used as the training data of the least squares method, and finally the accurate data of the known landmarks are used to verify the model's performance. Accuracy, find the error of the model.

S12. Build an indoor map;

The corresponding plane map is drawn according to a certain proportion of the area to be located, which is used for the constraints of particle filtering, and further improves the matching accuracy of the location and fingerprint in the process of building the database.

S13, marking explicit landmark points and collecting fingerprint data of corresponding positions;

Explicit landmark nodes are generally set at special physical locations, corridor intersections, stairway openings, and areas where pedestrians are easy to pass through. Before crowdsourcing the database, collect the physical coordinates and fingerprint data of explicit landmarks in advance and store them in the location fingerprint database.

S14. Determine whether the crowdsourcing path data is effective;

Due to the large cumulative error of PDR, if it is not corrected for a long time, the reliability of the collected crowdsourcing data is very low. In order to obtain relatively accurate crowdsourcing data, when there is only explicit landmark data in the database, it is set that the crowdsourcing path becomes effective when the user walks near the explicit landmark, and the short-distance location fingerprint correlation model is used to calculate The physical coordinates of the starting point when the crowdsourcing route becomes effective. According to the WiFi fingerprint similarity (shown in Equation 2), it is determined whether the user has walked around the explicit landmark.

S15, based on PDR, particle filter and map constraint multiple condition constraints to associate fingerprints and location points;

After the crowdsourcing route becomes effective, a location fingerprint database is constructed by combining PDR, particle filtering and map constraint multiple techniques. According to the data of the built-in sensors of the smartphone (gyroscope, magnetometer and accelerometer sensors), the pedestrian's steps, step length and heading are deduced, and the user's walking path is obtained;

Since the physical coordinates of the starting point when the crowdsourcing path is effective is known, the physical coordinates of the data collection points on the crowdsourcing path can be obtained, so as to obtain the matching of position and fingerprint; at the same time, the dual constraints of particle filtering and map constraints are used to obtain More accurate crowdsourcing routes.

S16, setting the effective time threshold of the crowdsourcing path, within the effective time threshold, if the cumulative error of the sensor data falls within the tolerable range, then proceed to step S15, otherwise, the crowdsourcing path is invalid, and proceed to step S17;

Since the sensor has accumulated error, in order to reduce the accumulated error, the accumulated error time threshold of the sensor is trained in advance, which is generally about 5-10 minutes. Within this time threshold, the accumulated error of the sensor data belongs to the tolerable range. When the time threshold is exceeded, the crowdsourcing route is invalid, and the collected data is not stored in the location fingerprint database.

S17. Build the database in sequence based on the fission method;

Referring to 3, the fission method refers to a series of valid path position points obtained from a known position point, which can also be used as an effective starting position point for crowdsourcing paths passing through these points. After the fingerprint data of the non-explicit location points is established in the location fingerprint database, the effective starting position of the crowdsourced path data of other users can be around any known location points (including explicit location points and valid crowdsourced path segments) . After the database is established by fission, each fingerprint point on the location fingerprint database corresponds to a rough physical coordinate.

Step S2, correcting the PDR cumulative error of the non-explicit landmark area based on MDS, the steps include:

S21. Calculate the area range of the implicit landmark point based on the fingerprint similarity;

Implicit landmark points are calculated by crowdsourcing paths, and the intersection of multiple crowdsourcing paths can be used as an implicit landmark point. Therefore, when more crowdsourcing paths are collected in the to-be-located area, more implicit landmark points can be calculated. Since the similarity of fingerprints around the same location point is very high, a fingerprint similarity threshold σ _sim can be set, and the small area where the fingerprint similarity exceeds the threshold is used as the implicit landmark point area, such as in Figure 3. area V. In order to reduce the influence of heterogeneity, the fingerprint similarity can be calculated after the relative fingerprint of the fingerprint is obtained.

S22. Count common paths between implicit landmark point regions;

The number of crowdsourcing paths passing through each landmark point (including implicit landmark points and explicit landmark points) is not necessarily the same. By numbering the continuous paths, the number of paths passing between two landmark point areas at the same time is obtained.

S23. Search for a relatively effective crowdsourcing path;

The cumulative error of each crowdsourcing path is different, and it is necessary to select the path with the smallest cumulative error among the common paths between the landmark points.

The magnetometer sensor is relatively stable for a long time, and is easily affected by the magnetic field of the mobile phone in a short time. The gyroscope sensor has a large accumulated error for a long time and is accurate in a short time. Therefore, the magnetic force can be selected in the common path between the two implicit landmark points in turn. The path with the least fluctuation in the count data is used as the effective path between the two punctuation areas.

S24. Redraw the new path between the implicit landmark point regions;

Based on the accelerometer and magnetometer sensor data of valid paths between implicit landmark point regions, new paths between implicit landmark point regions are redrawn.

S25. Calculate the relative coordinates of the implicit landmark point area;

The physical coordinates of the explicit landmark points are accurately known. Starting from the explicit landmark nodes, the coordinates of each implicit landmark point region relative to the explicit landmark points are calculated in turn. The process is as follows: If a crowdsourcing path passes through an explicit landmark point and an implicit landmark point area at the same time, the coordinates of the implicit landmark point area can be calculated, and then the crowdsourcing with the implicit landmark point area can be calculated. Additional implicit landmark point area coordinates connected by the path.

Referring to Figure 4, the explicit landmark points and the implicit landmark points A and E are connected by a path. According to the PDR algorithm and the coordinates of the explicit landmark point, the coordinates of the two points A and E can be obtained, while the implicit landmark point B and A have a path. If they are connected, the coordinates of point B can be obtained from the coordinates of point A, and so on to obtain the coordinates of other implicit landmark points connected by paths.

S26, accurately calculate the absolute coordinates of the implicit landmark point area based on MDS;

According to the relative coordinates of each implicit landmark point area and the explicit landmark point, the relative distance matrix D is constructed. Assuming that the number of explicit landmark points and implicit landmark points connected by the crowdsourcing path is m-1 in total, find this m- The center position point of 1 landmark point Find the relative distance d _ij between each location point (including explicit landmark point, implicit landmark point and central location point) according to the coordinates, where i=1,2,3,...,m, j=1,2, 3,...,m then the relative distance matrix is:

Set the center point as the origin, calculate the relative coordinates of the explicit and implicit landmark points and the origin according to the MDS, and update the original relative coordinates of the implicit landmark point. Further, the actual physical coordinates of each implicit landmark point are accurately obtained according to the actual physical coordinates of the explicit landmark points. Since the original relative coordinates of the implicit landmarks are calculated according to a single path, the error may be large. According to the MDS algorithm, the implicit landmarks can be mutually corrected to reduce the coordinate errors of the implicit landmarks.

S27. Correct the accumulated PDR error of the non-explicit landmark region.

Taking each implicit landmark point as the new starting point of the crowdsourcing path around the point, and drawing a new crowdsourcing path according to the subsequent part of the original path information passing through the point, the accumulated error of the PDR is further corrected, and the relationship between the position and the fingerprint is improved. Matching accuracy.

The present invention constructs a location fingerprint database based on the fission method and the set path effective time threshold, and uses MDS combined with implicit landmark points to correct the PDR cumulative error, and at the same time, constructs a short-distance location fingerprint correlation model based on weighted multi-dimensional WiFi numerical features; compared with the prior art The location fingerprint database can be quickly constructed, the accumulative error of the sensor is reduced, the matching accuracy of the location and the fingerprint is improved, and the problems of frequently disturbing the user and low matching accuracy of the location and the fingerprint in the prior art are effectively solved.

Although the specific embodiments of the invention have been described in detail with reference to the accompanying drawings, they should not be construed as limiting the protection scope of this patent. Within the scope described in the claims, various modifications and deformations that can be made by those skilled in the art without creative work still belong to the protection scope of this patent.

Claims (10)

1. A position fingerprint database rapid construction method based on crowdsourcing data is characterized by comprising the following steps:

s1, creating a library based on the fission of the PDR, the particle filter and the map constraint, wherein the steps comprise

S11, training a short-distance position fingerprint correlation model;

s12, constructing an indoor map;

s13, marking explicit landmark points and collecting fingerprint data of corresponding positions;

s14, judging whether the crowdsourcing path data is enabled;

s15, associating fingerprints and position points based on PDR, particle filtering and map constraint multiple conditions;

s16, setting an effective time threshold of a crowdsourcing path, and if the accumulated error of the sensor data is within a tolerable range within the effective time threshold, entering step S15, otherwise, failing the crowdsourcing path, and entering step S17;

s17, sequentially building a library based on the fission mode;

s2, correcting the PDR accumulated error of the non-explicit landmark region based on the MDS, wherein the steps comprise

S21, calculating an implicit landmark point area range based on the fingerprint similarity;

s22, counting common paths among the implicit landmark point areas;

s23, searching a relatively effective crowdsourcing path;

s24, redrawing a new path between the implicit landmark point areas;

s25, calculating the relative coordinates of the implicit landmark point area;

s26, accurately calculating the absolute coordinates of the implicit landmark point area based on MDS;

and S27, correcting the PDR accumulated error of the non-explicit landmark region.

2. The method for rapidly building a position fingerprint database based on crowdsourced data as claimed in claim 1, wherein the method for training the short-distance position fingerprint correlation model in step S11 comprises:

a1, modeling based on multi-dimensional WiFi numerical characteristic fingerprint distance;

and A2, training a short-distance position fingerprint correlation model.

3. The method for rapidly building a position fingerprint database based on crowdsourced data as claimed in claim 2, wherein the step a1 is based on multidimensional WiFi numerical characteristic fingerprint distance modeling method:

a11, constructing a relative fingerprint consisting of relative subsequences;

a12, calculating the similarity of the relative subsequence pairs;

a13, calculating the similarity of the fingerprint relative subsequence pairs in the relative fingerprint pairs, and obtaining a similarity matrix of the relative fingerprint pairs by adopting traversal calculation;

and A14, searching for the best matching relative fingerprint pair in the similarity matrix by adopting a dynamic programming algorithm.

4. The method for rapidly building a position fingerprint database based on crowdsourced data as claimed in claim 2, wherein the step a2 is a method for training a short-distance position fingerprint correlation model, comprising the steps of:

a21, acquiring crowd-sourced data around the known landmark points;

a22, extracting fingerprint data of the landmark point radiation area according to the fingerprint similarity;

a23, calculating the fingerprint distance of the fingerprint data of the landmark point radiation area;

and A24, determining a short-distance position fingerprint correlation model based on the MDS.

5. The method for quickly building a position fingerprint database based on crowdsourcing data according to claim 1, wherein the step S14 is to determine whether crowdsourcing path data is enabled by:

when only explicit landmark point data exists in the database, setting that when a user walks to the vicinity of the explicit landmark point, the crowdsourcing path starts to be effective, and calculating the physical coordinates of the initial position point when the crowdsourcing path starts to be effective by using a short-distance position fingerprint correlation model;

and judging whether the user has walked around the explicit landmark point according to the similarity of the WiFi fingerprint.

6. The method for rapidly building a location fingerprint database based on crowdsourced data as claimed in claim 1, wherein the step S15 is based on PDR, particle filtering and map constraint to associate fingerprints and location points under multiple conditions:

combining multiple technologies of PDR, particle filtering and map constraint to construct a position fingerprint database;

deducing the pedestrian step number, step length and course according to the built-in sensor data of the smart phone to obtain a user walking path, obtain the physical coordinates of data acquisition points on the crowdsourcing path, and obtain the matching of the position and the fingerprint;

and obtaining an accurate crowdsourcing path by adopting a particle filtering and map constraint dual constraint condition.

7. The method for rapidly building the position fingerprint database based on the crowdsourcing data according to claim 1, wherein the step S17 is implemented by sequentially building the database based on the fission mode segmentation:

when fingerprint data of non-explicit position points are established in the position fingerprint database, effective initial positions of crowdsourcing path data of other users are around any known position point, and the database is established based on a fission mode;

and each fingerprint point on the position fingerprint database after the fission mode library is built corresponds to a rough physical coordinate.

8. The method for rapidly building a position fingerprint database based on crowdsourced data as claimed in claim 1, wherein the step S21 includes setting a threshold σ of fingerprint similarity_simSimilarity of fingerprint exceeds sigma_simThe small area where the position point of (2) is located is taken as an implicit landmark point area;

step S22 includes numbering the continuous paths to obtain the number of paths passing through two landmark regions at the same time;

the step S23 includes sequentially selecting, from the common paths between the two implicit landmark regions, a path with the minimum fluctuation of magnetometer data as an effective path between the two landmark regions;

step S24 includes redrawing a new path between the implicit landmark regions according to the acceleration sensor and magnetometer sensor data of the effective path between the implicit landmark regions;

the step S25 includes sequentially calculating coordinates of each implicit landmark region relative to the explicit landmark points from the explicit landmark nodes.

9. The method for rapidly building a position fingerprint database based on crowdsourced data as claimed in claim 1, wherein the step S26 is based on the method of accurately calculating absolute coordinates of the implicit landmark region by MDS as follows:

constructing a relative distance matrix D according to the relative coordinates of each implicit landmark point area and each explicit landmark point,

wherein d is_ijRelative distances between explicit landmark points, implicit landmark points and center location points,

i＝1,2,3，…，m；j＝1,2,3…，m；

setting the central position point as an origin, calculating relative coordinates of the explicit and implicit landmark points and the origin according to the MDS, and updating the original relative coordinates of the implicit landmark points;

and solving the actual physical coordinates of each implicit landmark point according to the actual physical coordinates of the explicit landmark points.

10. The method for rapidly building a position fingerprint database based on crowdsourced data as claimed in claim 1, wherein the step S27 is to correct PDR accumulated errors of non-explicit landmark regions by:

and taking each implicit landmark point as a new starting point of a crowdsourcing path passing around the point, drawing a new crowdsourcing path according to the information of the subsequent part of original path passing through the point, correcting the accumulated error of the PDR, and improving the matching precision of the position and the fingerprint.