CN105704676A - Method for improving fingerprint indoor positioning precision through employing signal time correlation - Google Patents

Abstract

The invention provides a method for improving the fingerprint indoor positioning precision through employing signal time correlation, and the method comprises the steps: carrying out repeated signal intensity measurement of each calibration point in an indoor region, and obtaining a signal intensity fingerprint map, wherein each signal is corresponding to one time sequence vector; (user) uploading a signal intensity time sequence of a plurality of signals measured at a certain position; enabling a user to carry out the matching of user data and the fingerprint map, obtaining the most possible corresponding position, and transmitting a positioning result to a user client; and enabling the user to obtain his or her positioning result. Through the mining of the time correlation information of the signal intensity time sequence, the method achieves novel positioning modeling and discrimination, and can improve the indoor positioning accuracy and reliability.

Description

Utilize the method that signal time dependency improves fingerprint indoor position accuracy

Technical field

The present invention relates to communication, technical field of navigation and positioning, in particular it relates to a kind of method utilizing temporal correlation to improve Wi-Fi fingerprint location precision。

Background technology

Positioning service is the technology having wide application prospect for many years in various fields, and gps satellite location and all kinds of digital map navigation facilitate huge numbers of families。In the epoch that this mobile interchange is fast-developing, indoor positioning is the chance of last meter of Information Mobile Service, public safety, parking lot, market locating guide, the social numerous areas such as interaction, commodity market popularization of making friends, and is required for good indoor positioning technologies and provides support。Indoor positioning technologies is one of technology of mobile interchange and the popular research and development of Internet of Things epoch。

Nowadays, indoor positioning mainly adopts the position that the technology such as wireless telecommunications, architecture, inertial navigation location realize indoor occupant and object to determine。Except the cellular localization technology of communication network, common indoor wireless location technology also has Wi-Fi, bluetooth, RFID, ZigBee, ultrasound wave etc.。Wherein belonging to the application of Wi-Fi WLAN and cover the most extensive, current family, market, dining room hotel environment have increasing Wi-Fi Hotspot to provide service。Based on Wi-Fi hardware facility widely basis and premise, the Wi-Fi technical research realizing location is utilized constantly to improve, currently mainly there are two kinds of solutions: a kind of probabilistic model being experience test and setting up fingerprint map, another kind is to utilize signal propagation model to carry out location estimation calculating。Due to problems such as time synchronized, interchannel noise, sensor accuracies, the positioning precision of TDOA technology is difficult to be increased dramatically。Meanwhile, big data cloud computing platform gives Wi-Fi the more thinking of fingerprint positioning method and opportunity。

But the precision based on the fingerprint positioning method of Wi-Fi is difficult to meet higher demand all the time, main reason is that the time variation of signal, complex environment multipath effect, hardware sensitivity and mobile object influence of noise etc.。Many research work also develop many methods to improve the precision of Wi-Fi fingerprint positioning method, as utilized the sensor that embeds of mobile equipment to find out the position of mobile object, the method a large amount of training data of energy quick obtaining that gunz crowd raises, also has the method for machine learning to reduce the time delay etc. of positioning stage。

Through the retrieval of prior art literature is found, signal strength values is modeled analyzing by overwhelming majority scientific research as independent variable, the temporal correlation of signal is regarded as the factor affecting positioning precision simultaneously。However we have found that just because of complex environment factor, causing that unlike signal has different temporal correlations at diverse location, this information can extract as extra information to improve the precision of location。

Summary of the invention

For defect of the prior art, it is an object of the invention to provide a kind of method utilizing signal time dependency to improve fingerprint indoor position accuracy。

According to a kind of method utilizing signal time dependency to improve fingerprint indoor position accuracy provided by the invention, comprise the steps:

Step 1: training data under gathering line；

Step 2: each Wi-Fi signal projector for each scaling point place sets up gaussian probability distribution, and obtains the parameter of gaussian probability distribution with training data under the line gathered；

Step 3: the parameter being distributed based on described gaussian probability, is trained hyperbolic surface boundary, obtains the interval E that signal intensity RSS data positions in probability space, completes signal intensity fingerprint map；

Step 4: receiving the data that user uploads, the data that described user uploads include signal intensity matrix x and Location Request；Wherein, the acquisition methods of signal intensity matrix x is: used mobile terminal to take multiple measurements wireless signal strength by user, it is thus achieved that the signal intensity matrix x of different routers；

Step 5: according to Location Request, data user uploaded are compared with signal intensity fingerprint map, it is considered to D ties up seasonal effect in time series dependency, is mapped as the vector of probability space, by the coupling of hyperbolic surface boundary condition and correlation coefficient ρ, find the position of maximum of probability as positioning result；

Step 6: positioning result is returned to user。

Step 7: positioning result is fed back by user, if user feedback positioning result is correct, then puts into the signal intensity matrix x of measurement and trains described gaussian probability to be distributed in data base；If user feedback positioning result is wrong, then the signal intensity matrix x of measurement is not put into and data base trains described gaussian probability be distributed。

Preferably, described step 1 comprises the steps:

Step 1.1: indoor positioning region is divided into interval, multiple location and sets k scaling point；

Step 1.2: at each scaling point place, open the scanning equipment of Wi-Fi signal strength, demarcate the positional information of place scaling point, the signal that n Wi-Fi signal projector within the scope of the signal search of scanning equipment is sent, set trace interval τ and scanning times m；Make scanning equipment start scanning survey, and measurement data be stored as signal intensity matrix x:

Wherein, x_i, 1≤i≤n, represent the time series vector of the signal sent at a scaling point place i-th Wi-Fi signal projector；X_lj, 1≤l≤n, 1≤j≤m, represent that the signal that the l Wi-Fi signal projector sends within the scope of a scaling point place is to signal search carries out the measurement data that the secondary signal strength measurement of jth obtains；

Step 1.3: using signal intensity matrix x as training data under line。

Preferably, described step 2 comprises the steps:

Step 2.1: the time series vector x to each scaling point r place_iConsider the signal time dependency measured D time, build D and tie up gaussian probability distribution f_r(x_i) it is:

$f_r\left(x_i\right)=\frac{1}{{\left(2\mathrm{\π}\right)}^{\frac{D}{2}}\det {\left(\mathrm{\Σ}\right)}^{\frac{1}{2}}}{e}^{-\frac{1}{2}{\mathrm{\Δ}}^2}$

Wherein, determinant, Δ are asked in det (*) expression²For mahalanobis distance, e represents the nature truth of a matter；

Δ²=[(x-μ)^T∑^-1(x-μ)]；

Wherein, ∑ represents correlation matrix；

Step 2.2: D is tieed up gaussian probability distribution f_r(x_i) each dimension carry out coordinate system rotation, rotate 45 °, make mahalanobis distance Δ²It is for conversion into the orthogonal representation of only quadratic term；

Step 2.3: utilize maximum Likelihood, calculates the parameter μ of gaussian probability distribution, ∑, draws the fingerprint map of Wi-Fi signal strength；

First, building and utilize measurement data to constitute correlation calculations matrix A, every D adjacent data constitutes a vector:

Wherein, x_ijRepresent that signal i-th Wi-Fi signal projector sent at a scaling point place carries out the measurement data that jth time signal strength measurement obtains, 1≤j≤m；

Then A is utilized to calculate the parameter μ of gaussian probability distribution, ∑:

$\mathrm{\μ}=\left[\begin{array}{c}{\mathrm{\μ}}_1\\ {\mathrm{\μ}}_2\\ .\\ .\\ .\\ {\mathrm{\μ}}_D\end{array}\right]$

Wherein, μ is the mean vector of signal intensity；μ_i, i=1,2 ..., D, represent A^TI-th vector average；

∑=[Cov [A_k,A_j]], k=1,2 ..., D；J=1,2 ..., D

Wherein, A_kFor A^TKth vector, A_jRepresent A^TJth vector；

For the correlation coefficient D dimensional vector of signal, owing to signal communication process is stationary random process, ρ_iRepresent that i ties up the correlation coefficient under interval；

Σ ρ_iIt is expressed as:

Especially, when D is equal to 2, correlation matrix ∑ is expressed as:

$\mathrm{\Σ}=\left[\begin{array}{cc}{\mathrm{\σ}}^2& {\mathrm{\ρ\σ}}^2\\ {\mathrm{\ρ\σ}}^2& {\mathrm{\σ}}^2\end{array}\right]$

Wherein, σ²Represent the variance of a certain row element in signal intensity matrix x。

Preferably, described step 2.2 comprises the steps:

Find one group of orthonormal basis u of correlation matrix ∑_i, i=1,2 ..., m, then transformation matrix U is U=[u₁,u₂,...,u_m]^T, wherein, u_iRepresent i-th orthonormal basis；

If distribution variable y=U (x-μ), then f_r(x_i) transformed after obtain D under new coordinate system and tie up gaussian probability distribution f_r(y | μ, ∑):

$f_r\left(y\right|\mathrm{\μ},\mathrm{\Σ})=\frac{1}{{\left(2\mathrm{\π}\right)}^{\frac{D}{2}}{\left({\mathrm{\Π}}_{i=1}^m{\mathrm{\λ}}_i\right)}^{\frac{1}{2}}}{e}^{-\frac{1}{2}\left({\mathrm{\Σ}}_{i=1}^m\frac{1}{{\mathrm{\λ}}_i}{y}_i{y_i}^T\right)}$

Wherein, λ_iRepresent ith feature value；Y_iRepresent i-th dimension variable。

Preferably, described step 3 comprises the steps:

Step 3.1: be distributed based on described gaussian probability, carries out border training to sample in probability space；Scaling point r correspondence sample space interval in physical space is E, is defined as:

E={x | f_r(y|μ(r),Σ(r))≥f_r±δ(y|μ(r±δ),Σ(r±δ))}

Wherein, x represents signal intensity matrix, y represents distribution variable, δ represents error distance, μ (r) represents the mean vector of scaling point r place signal intensity, Σ (r) represents the mean vector of scaling point r ± δ place signal intensity that the correlation matrix at scaling point r place, μ (r ± δ) represent under error distance δ, and Σ (r ± δ) represents the correlation matrix at the scaling point r ± δ place under error distance δ；Distance corresponding in symbol ± expression physical space, wherein, in symbol+expression physical space, correspondence is remote, and in symbol-expression physical space, correspondence is near

Step 3.2: in the sample space E of location, classification boundaries is hyperbolic curved surface, and the mathematical notation of hyperbolic curved surface is:

$\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{m}{\Σ}}\frac{{y_{i,j}}^2}{{\mathrm{\λ}}_{i,j}}-\underset{i=1}{\overset{n}{\Σ}}\[\frac{{(y_{i,j}\±\sqrt{2}\mathrm{\δ}\▿{\mathrm{\μ}}_i^{\′})}^2}{{\mathrm{\λ}}_{i,1}^{\±}}+\underset{j=1}{\overset{m}{\Σ}}\frac{{y_{i,j}}^2}{{\mathrm{\λ}}_{i,j}^{\±}}\]\≤\underset{i=1}{\overset{n}{\Σ}}\ln \frac{\det \left({\Σ}_i\right)}{\det \left({\Σ}_i^{\±}\right)}$

Wherein, y_i,jRepresent in i-th signal, the distribution variable of jth dimension in D dimension；λ_i,jRepresent the jth eigenvalue of i-th signal, λ_i,1Represent first eigenvalue of i-th signal, μ '_iRepresent the strength mean value of i-th signal, ∑_iRepresent the correlation matrix of i-th signal, symbolRepresenting gradient function, distance corresponding in subscript ± expression physical space, wherein, in subscript+expression physical space, correspondence is remote, and in subscript-expression physical space, correspondence is near；

Step 3.3: utilize the parameter of the parameter training hyperbolic surface boundary that described gaussian probability is distributed, as the benchmark of location in probability space。

Preferably, described step 5 comprises the steps:

Step 5.1: obtaining Location Request, data user uploaded are compared with signal intensity fingerprint map；

Wherein, signal intensity fingerprint map includes the n dimension mean vector mean that the average of n Wi-Fi signal is constituted:

Mean=[μ₁,μ₂,…,μ_n]

And the D of each Wi-Fi signal ties up (n*D) that correlation coefficient constitutes and ties up correlation vector corr:

Wherein, σ_iRepresent the standard deviation of i-th Wi-Fi signal, wherein, i=1,2 ..., n；ρ_i,jRepresent the jth dimension correlation coefficient of i-th Wi-Fi signal, wherein, i=1,2 ..., n, j=2,3 ..., D；

When comparing, first compare the mean vector mean in fingerprint, find a fingerprint positions of front w relatively more similar in the threshold range set；Again in this w similar fingerprint positions, compare the Euclidean distance of correlation vector corr, find the minimum correlation vector mated most of Euclidean distance in order to determine the final position of user；W<n；

Step 5.2: using every for the data of Location Request adjacent D data as one group, is placed in D with the form of D dimensional vector and ties up in probability space；Under line, the hyperbolic surface boundary of training is as location condition, by the coupling of the correlation coefficient D dimensional vector ρ of signal, finds physical spatial location corresponding for signal intensity matrix x；

Step 5.3: consider the physical spatial location that the comparison of fingerprint is corresponding with signal intensity matrix x, determine final positioning result；The method specifically considering decision is as follows:

First calculate the dependability parameter η of hyperbolic surface boundary, f can be made_r(y | μ (r), Σ (r))=f_r±δThe function maxima that (y | μ (r ± δ), Σ (r ± δ)) sets up；

Then according to η and one threshold value η₀Comparison, η₀Span take (0,0.3)；If η≤η₀, then representing hyperbolic surface boundary reliable, the physical spatial location corresponding for signal intensity matrix x step 5.2 obtained is as positioning result；Otherwise, if i.e. η > η₀, then the positioning result reliability representing hyperbolic surface boundary is not good, then the minimum correlation vector mated most of described Euclidean distance obtained according to step 5.1 determines that the final position of user is as positioning result。

Compared with prior art, the present invention has following beneficial effect:

1, the present invention can pass through to excavate signal intensity seasonal effect in time series temporal correlation information, it is proposed to the modeling of novel location and method of discrimination, it is possible to heighten accuracy and the reliability of indoor positioning；

2, the present invention based on existing Indoor Locating Model, when situation about can not judge occur, it is provided that extra auxiliary positioning information, can improve the realization of indoor locating system further。

3, the present invention utilizes the method that signal time dependency improves Wi-Fi fingerprint indoor position accuracy, by seizing the temporal correlation information of adjacent signal strength further, strengthens the reliability to location and accuracy。

Accompanying drawing explanation

By reading detailed description non-limiting example made with reference to the following drawings, the other features, objects and advantages of the present invention will become more apparent upon:

Fig. 1 is the flow chart of steps of the present invention。

Fig. 2 is dimensional Gaussian example profile。

Detailed description of the invention

Below in conjunction with specific embodiment, the present invention is described in detail。Following example will assist in those skilled in the art and are further appreciated by the present invention, but do not limit the present invention in any form。It should be pointed out that, to those skilled in the art, without departing from the inventive concept of the premise, it is also possible to make some changes and improvements。These broadly fall into protection scope of the present invention。

The invention provides a kind of temporal correlation utilizing wireless signal and improve the method based on Wi-Fi fingerprint indoor position accuracy, comprise the following steps: utilize mobile phone or other there is the equipment of Wi-Fi strength measurement module, each scaling point place at room area, signal to each signal projector, all the interval identical time carries out repeated signal strength measurement, the all corresponding time series vector of each signal, is stored as data base；Data are processed and analyze by server, set up the higher-dimension gaussian probability distributed model of its intensity for each signal of each position, and utilize the signal intensity time series vector of identical interval, calculate probability distribution parameters by maximal possibility estimation；Based on this again through methods such as machine learning, the hyperbola border of training boundary, thus obtaining the fingerprint map of signal intensity；Signal is carried out temporal correlation analysis simultaneously, calculates temporal correlation data base, to utilize signal time dependency and physical location information to carry out auxiliary positioning；Server waits that user transmits signal intensity and Location Request；User uploads the signal intensity time series of its multiple signals arrived in certain position measurement, i.e. multiple vectors, and upload onto the server end；User data and fingerprint map are mated by server, obtain most probable correspondence position, and positioning result is beamed back subscription client；User obtains the positioning result of oneself。

According to a kind of method utilizing signal time dependency to improve fingerprint indoor position accuracy provided by the invention, comprise the steps:

Step 1: training data under gathering line；

Step 2: each Wi-Fi signal projector for each scaling point place sets up gaussian probability distribution, and obtains the parameter of gaussian probability distribution with training data under the line gathered；

Step 3: the parameter being distributed based on described gaussian probability, is trained hyperbolic surface boundary, obtains the interval E that signal intensity RSS data positions in probability space, completes signal intensity fingerprint map；

Step 4: receiving the data that user uploads, the data that described user uploads include signal intensity matrix x and Location Request；Wherein, the acquisition methods of signal intensity matrix x is: used mobile terminal to take multiple measurements wireless signal strength by user, it is thus achieved that the signal intensity matrix x of different routers；

Step 5: according to Location Request, data user uploaded are compared with signal intensity fingerprint map, it is considered to D ties up seasonal effect in time series dependency, is mapped as the vector of probability space, by the coupling of hyperbolic surface boundary condition and correlation coefficient ρ, find the position of maximum of probability as positioning result；

Step 6: positioning result is returned to user；

Step 7: positioning result is fed back by user, if user feedback positioning result is correct, then puts into the signal intensity matrix x of measurement and trains described gaussian probability to be distributed in data base；If user feedback positioning result is wrong, then the signal intensity matrix x of measurement is not put into and data base trains described gaussian probability be distributed。

Described step 1 comprises the steps:

Step 1.1: indoor positioning region is divided into interval, multiple location and sets k scaling point；

Step 1.2: at each scaling point place, open the scanning equipment of Wi-Fi signal strength, demarcate the positional information of place scaling point, the signal that n Wi-Fi signal projector within the scope of the signal search of scanning equipment is sent, set trace interval τ and scanning times m；Make scanning equipment start scanning survey, and measurement data be stored as signal intensity matrix x:

Wherein, x_i, 1≤i≤n, represent the time series vector of the signal sent at a scaling point place i-th Wi-Fi signal projector；X_ij, 1≤l≤n, 1≤j≤m, represent that the signal that the l Wi-Fi signal projector sends within the scope of a scaling point place is to signal search carries out the measurement data that the secondary signal strength measurement of jth obtains；

Step 1.3: using signal intensity matrix x as training data under line。

Described step 2 comprises the steps:

Step 2.1: the time series vector x to each scaling point r place_iConsider the signal time dependency measured D time, build D and tie up gaussian probability distribution f_r(x_i) it is:

$f_r\left(x_i\right)=\frac{1}{{\left(2\mathrm{\&pi;}\right)}^{\frac{D}{2}}\det {\left(\mathrm{\&Sigma;}\right)}^{\frac{1}{2}}}{e}^{-\frac{1}{2}{\mathrm{\&Delta;}}^2}$

Wherein, determinant, Δ are asked in det (*) expression²For mahalanobis distance, e represents the nature truth of a matter；

Δ²=[(x-μ)^T∑^-1(x-μ)]；

Wherein, ∑ represents correlation matrix；

Especially, as D=2, Gauss distribution and Ma Shi distance particularly as follows:

$f(x_1,x_2)=\frac{1}{2{\mathrm{\&pi;\&sigma;}}^2\sqrt{1-{\mathrm{\&rho;}}^2}}{e}^{-\frac{1}{2{\mathrm{\&sigma;}}^2\left(1-{\mathrm{\&rho;}}^2\right)}\&lsqb;{(x_1-\mathrm{\&mu;})}^2+{(x_2-\mathrm{\&mu;})}^2-2\mathrm{\&rho;}(x_1-\mathrm{\&mu;})(x_2-\mathrm{\&mu;})\&rsqb;}$

${\mathrm{\&Delta;}}^2={(x-\mathrm{\&mu;})}^T{\mathrm{\&Sigma;}}^{-1}(x-\mathrm{\&mu;})=\frac{1}{{\mathrm{\&sigma;}}^2(1-{\mathrm{\&rho;}}^2)}\&lsqb;{(x_1-\mathrm{\&mu;})}^2+{(x_2-\mathrm{\&mu;})}^2-2\mathrm{\&rho;}(x_1-\mathrm{\&mu;})(x_2-\mathrm{\&mu;})\&rsqb;$

Step 2.2: D is tieed up gaussian probability distribution f_r(x_i) each dimension carry out coordinate system rotation, rotate 45 °, make mahalanobis distance Δ²It is for conversion into the orthogonal representation of only quadratic term；

Step 2.3: utilize maximum Likelihood, calculates the parameter μ of gaussian probability distribution, ∑, draws the fingerprint map of Wi-Fi signal strength；

First, building and utilize measurement data to constitute correlation calculations matrix A, every D adjacent data constitutes a vector:

Wherein, x_ijRepresent that signal i-th Wi-Fi signal projector sent at a scaling point place carries out the measurement data that jth time signal strength measurement obtains, 1≤j≤m；

Then A is utilized to calculate the parameter μ of gaussian probability distribution, ∑:

$\mathrm{\&mu;}=\left[\begin{array}{c}{\mathrm{\&mu;}}_1\\ {\mathrm{\&mu;}}_2\\ .\\ .\\ .\\ {\mathrm{\&mu;}}_D\end{array}\right]$

Wherein, μ is the mean vector of signal intensity；μ_i, i=1,2 ..., D, represent A^TI-th vector average；

Σ=[Cov [A_k,A_j]], k=1,2 ..., D；J=1,2 ..., D

Wherein, A_kFor A^TKth vector, A_jRepresent A^TJth vector；

ρ is the correlation coefficient D dimensional vector of signal, owing to signal communication process is stationary random process, and ρ_iRepresent that i ties up the correlation coefficient under interval；

Σ ρ_iIt is expressed as:

∑ contains training data under line and obtains the reference value of positional parameter。

When D is equal to 2, correlation matrix ∑ is expressed as:

$\mathrm{\&Sigma;}=\left(\begin{array}{cc}{\mathrm{\&sigma;}}^2& {\mathrm{\&rho;\&sigma;}}^2\\ {\mathrm{\&rho;\&sigma;}}^2& {\mathrm{\&sigma;}}^2\end{array}\right)=\frac{\sqrt{2}{\mathrm{\&sigma;}}^2(1+\mathrm{\&rho;})}{2}\left(\begin{array}{cc}1& 1\\ 1& 1\end{array}\right)+\frac{\sqrt{2}{\mathrm{\&sigma;}}^2(1-\mathrm{\&rho;})}{2}\left(\begin{array}{cc}1& -1\\ -1& 1\end{array}\right)$

Wherein, σ²Represent the variance of a certain row element in signal intensity matrix x, i.e. the variance of the signal sequence that certain Wi-Fi signal projector of scaling point place sends。

Described step 2.2 comprises the steps:

GSO (GramSchmidtOrthogonallization) orthogonalization method can be utilized, find one group of orthonormal basis u of correlation matrix ∑_i, i=1,2 ..., m, then transformation matrix U is U=[u₁,u₂,...,u_m]^T, wherein, u_iRepresent i-th orthonormal basis；

If distribution variable y=U (x-μ), then f_r(x_i) transformed after obtain D under new coordinate system and tie up gaussian probability distribution f_r(y | μ, ∑):

$f_r\left(y\right|\mathrm{\&mu;},\mathrm{\&Sigma;})=\frac{1}{{\left(2\mathrm{\&pi;}\right)}^{\frac{D}{2}}{\left({\&Pi;}_{i=1}^m{\mathrm{\&lambda;}}_i\right)}^{\frac{1}{2}}}{e}^{-\frac{1}{2}\left({\mathrm{\&Sigma;}}_{i=1}^m\frac{1}{{\mathrm{\&lambda;}}_i}{y}_i{y_i}^T\right)}$

Wherein, λ_iRepresent ith feature value, equation can be madeThe λ set up_i；Y_iRepresent i-th dimension variable。

As D=2, orthogonal basis and eigenvalue are respectively as follows:

$\begin{array}{cc}{u}_1=\left[\begin{array}{c}\frac{\sqrt{2}}{2}\\ \frac{\sqrt{2}}{2}\end{array}\right]& u_2=\left[\begin{array}{c}-\frac{\sqrt{2}}{2}\\ \frac{\sqrt{2}}{2}\end{array}\right]\end{array}$

$\begin{array}{cc}{\mathrm{\&lambda;}}_1=\frac{\sqrt{2}{\mathrm{\&sigma;}}^2(1+\mathrm{\&rho;})}{2}& {\mathrm{\&lambda;}}_2=\frac{\sqrt{2}{\mathrm{\&sigma;}}^2(1-\mathrm{\&rho;})}{2}\end{array}$

Described step 3 comprises the steps:

Step 3.1: be distributed based on described gaussian probability, carries out border training to sample in probability space；Scaling point r correspondence sample space interval in physical space is E, is defined as:

E={x | f_r(y|μ(r),Σ(r))≥f_r±δ(y|μ(r±δ),Σ(r±δ))}

Wherein, x represents signal intensity matrix, y represents distribution variable, δ represents error distance, μ (r) represents the mean vector of scaling point r place signal intensity, Σ (r) represents the mean vector of scaling point r ± δ place signal intensity that the correlation matrix at scaling point r place, μ (r ± δ) represent under error distance δ, and Σ (r ± δ) represents the correlation matrix at the scaling point r ± δ place under error distance δ；Distance corresponding in symbol ± expression physical space, wherein, in symbol+expression physical space, correspondence is remote, and in symbol-expression physical space, correspondence is near

Step 3.2: in the sample space E of location, classification boundaries is hyperbolic curved surface, and the mathematical notation of hyperbolic curved surface is:

$\underset{i=1}{\overset{n}{\&Sigma;}}\underset{j=1}{\overset{m}{\&Sigma;}}\frac{{y_{i,j}}^2}{{\mathrm{\&lambda;}}_{i,j}}-\underset{i=1}{\overset{n}{\&Sigma;}}\&lsqb;\frac{{(y_{i,j}\&PlusMinus;\sqrt{2}\mathrm{\&delta;}\&dtri;{\mathrm{\&mu;}}_i^{\&prime;})}^2}{{\mathrm{\&lambda;}}_{i,1}^{\&PlusMinus;}}+\underset{j=2}{\overset{m}{\&Sigma;}}\frac{{y_{i,j}}^2}{{\mathrm{\&lambda;}}_{i,j}^{\&PlusMinus;}}\&rsqb;\&le;\underset{i=1}{\overset{n}{\&Sigma;}}\ln \frac{\det \left({\&Sigma;}_i\right)}{\det \left({\&Sigma;}_i^{\&PlusMinus;}\right)}$

Wherein, y_i,jRepresent in i-th signal, the distribution variable of jth dimension in D dimension；λ_i,jRepresent the jth eigenvalue of i-th signal, λ_i,1Represent first eigenvalue of i-th signal, μ '_iRepresent the strength mean value of i-th signal, ∑_iRepresent the correlation matrix of i-th signal, symbolRepresenting gradient function, distance corresponding in subscript ± expression physical space, wherein, in subscript+expression physical space, correspondence is remote, and in subscript-expression physical space, correspondence is near；

As D=2, this border is hyperboloid, and its expression formula is:

$\left\{\begin{array}{c}(\frac{{y_1}^2}{{\mathrm{\&lambda;}}_1}+\frac{{y_2}^2}{{\mathrm{\&lambda;}}_2})-(\frac{{(y_1+\sqrt{2}\mathrm{\&delta;}\&dtri;\mathrm{\&mu;})}^2}{{{\mathrm{\&lambda;}}_1}^{\&prime;}}+\frac{{y_2}^2}{{{\mathrm{\&lambda;}}_2}^{\&prime;}})\&le;log\frac{{\mathrm{\&lambda;}}_1{\mathrm{\&lambda;}}_2}{{{\mathrm{\&lambda;}}_1}^{\&prime;}{{\mathrm{\&lambda;}}_2}^{\&prime;}}\\ (\frac{{y_1}^2}{{\mathrm{\&lambda;}}_1}+\frac{{y_2}^2}{{\mathrm{\&lambda;}}_2})-(\frac{{(y_1-\sqrt{2}\mathrm{\&delta;}\&dtri;\mathrm{\&mu;})}^2}{{{\mathrm{\&lambda;}}_1}^{\&prime;\&prime;}}+\frac{{y_2}^2}{{{\mathrm{\&lambda;}}_2}^{\&prime;\&prime;}})\&le;log\frac{{\mathrm{\&lambda;}}_1{\mathrm{\&lambda;}}_2}{{{\mathrm{\&lambda;}}_1}^{\&prime;\&prime;}{{\mathrm{\&lambda;}}_2}^{\&prime;\&prime;}}\end{array}\right.$

Step 3.3: utilize the parameter of the parameter training hyperbolic surface boundary that described gaussian probability is distributed, as the benchmark of location in probability space。

Described step 5 comprises the steps:

Step 5.1: obtaining Location Request, data user uploaded are compared with signal intensity fingerprint map；

Wherein, signal intensity fingerprint map includes the n dimension mean vector mean that the average of n Wi-Fi signal is constituted:

Mean=[μ₁,μ₂,…,μ_n]

And the D of each Wi-Fi signal ties up (n*D) that correlation coefficient constitutes and ties up correlation vector corr:

Wherein, σ_iRepresent the standard deviation of i-th Wi-Fi signal, wherein, i=1,2 ..., n；ρ_i,jRepresent the jth dimension correlation coefficient of i-th Wi-Fi signal, wherein, i=1,2 ..., n, j=2,3 ..., D；

When comparing, first compare the mean vector mean in fingerprint, find a fingerprint positions of front w relatively more similar in the threshold range set；Again in this w similar fingerprint positions, compare the Euclidean distance of correlation vector corr, find the minimum correlation vector mated most of Euclidean distance in order to determine the final position of user；W<n；

Step 5.2: using every for the data of Location Request adjacent D data as one group, is placed in D with the form of D dimensional vector and ties up in probability space；Under line, the hyperbolic surface boundary of training is as location condition, by the coupling of the correlation coefficient D dimensional vector ρ of signal, finds physical spatial location corresponding for signal intensity matrix x；

Step 5.3: consider the physical spatial location that the comparison of fingerprint is corresponding with signal intensity matrix x, determine final positioning result；The method specifically considering decision is as follows:

First calculate the dependability parameter η of hyperbolic surface boundary, f can be made_r(y | μ (r), Σ (r))=f_r±δThe function maxima that (y | μ (r ± δ), Σ (r ± δ)) sets up；

Then according to η and one threshold value η₀Comparison, η₀Span take (0,0.3)；If η≤η₀, then representing hyperbolic surface boundary reliable, the physical spatial location corresponding for signal intensity matrix x step 5.2 obtained is as positioning result；Otherwise, if i.e. η > η₀, then the positioning result reliability representing hyperbolic surface boundary is not good, then the minimum correlation vector mated most of described Euclidean distance obtained according to step 5.1 determines that the final position of user is as positioning result。

In one embodiment, ambient parameter is:

Mobile terminal device: six Android intelligent, is all Nexus4, and the operating system that every smart mobile phone is equipped with six smart mobile phones of 1.5GHzSnapdragonAPQ8064CPU and 2GRAM is all AndroidJellyBean (4.2)。These six smart mobile phones carry out indoor positioning as testing mobile phone side by side。

Server: grand base 4930G notebook computer, Duo dual core processor, the internal memory of 2G, the dominant frequency of 2G。

Above specific embodiments of the invention are described。It is to be appreciated that the invention is not limited in above-mentioned particular implementation, those skilled in the art can make a variety of changes within the scope of the claims or revise, and this has no effect on the flesh and blood of the present invention。When not conflicting, embodiments herein and the feature in embodiment can arbitrarily be mutually combined。

Claims (7)

1. one kind utilizes the method that signal time dependency improves fingerprint indoor position accuracy, it is characterised in that comprise the steps:

Step 1: training data under gathering line；

Step 2: each Wi-Fi signal projector for each scaling point place sets up gaussian probability distribution, and obtains the parameter of gaussian probability distribution with training data under the line gathered；

Step 3: the parameter being distributed based on described gaussian probability, is trained hyperbolic surface boundary, obtains the interval E that signal intensity RSS data positions in probability space, completes signal intensity fingerprint map；

Step 4: receiving the data that user uploads, the data that described user uploads include signal intensity matrix x and Location Request；Wherein, the acquisition methods of signal intensity matrix x is: used mobile terminal to take multiple measurements wireless signal strength by user, it is thus achieved that the signal intensity matrix x of different routers；

Step 5: according to Location Request, data user uploaded are compared with signal intensity fingerprint map, it is considered to D ties up seasonal effect in time series dependency, is mapped as the vector of probability space, by the coupling of hyperbolic surface boundary condition and correlation coefficient ρ, find the position of maximum of probability as positioning result；

Step 6: positioning result is returned to user。

2. the method utilizing signal time dependency to improve fingerprint indoor position accuracy according to claim 1, it is characterised in that described step 1 comprises the steps:

Step 1.1: indoor positioning region is divided into interval, multiple location and sets k scaling point；

Step 1.2: at each scaling point place, open the scanning equipment of Wi-Fi signal strength, demarcate the positional information of place scaling point, the signal that n Wi-Fi signal projector within the scope of the signal search of scanning equipment is sent, set trace interval τ and scanning times m；Make scanning equipment start scanning survey, and measurement data be stored as signal intensity matrix x:

Wherein, x_i, 1≤i≤n, represent the time series vector of the signal sent at a scaling point place i-th Wi-Fi signal projector；X_lj, 1≤l≤n, 1≤j≤m, represent that the signal that the l Wi-Fi signal projector sends within the scope of a scaling point place is to signal search carries out the measurement data that the secondary signal strength measurement of jth obtains；

Step 1.3: using signal intensity matrix x as training data under line。

3. the method utilizing signal time dependency to improve fingerprint indoor position accuracy according to claim 2, it is characterised in that described step 2 comprises the steps:

Step 2.1: the time series vector x to each scaling point r place_iConsider the signal time dependency measured D time, build D and tie up gaussian probability distribution f_r(x_i) it is:

$f_r\left(x_i\right)=\frac{1}{{\left(2\mathrm{\&pi;}\right)}^{\frac{D}{2}}\det {\left(\mathrm{\&Sigma;}\right)}^{\frac{1}{2}}}{e}^{-\frac{1}{2}{\mathrm{\&Delta;}}^2}$

Wherein, determinant, Δ are asked in det (*) expression²For mahalanobis distance, e represents the nature truth of a matter；

Δ²=[(x-μ)^T∑^-1(x-μ)]；

Wherein, ∑ represents correlation matrix；

Step 2.2: D is tieed up gaussian probability distribution f_r(x_i) each dimension carry out coordinate system rotation, rotate 45 °, make mahalanobis distance Δ²It is for conversion into the orthogonal representation of only quadratic term；

Step 2.3: utilize maximum Likelihood, calculates the parameter μ of gaussian probability distribution, ∑, draws the fingerprint map of Wi-Fi signal strength；

First, structure utilizes measurement data to constitute correlation calculations matrix A:

Wherein, x_ijRepresent that signal i-th Wi-Fi signal projector sent at a scaling point place carries out the measurement data that jth time signal strength measurement obtains, 1≤j≤m；

Then A is utilized to calculate the parameter μ of gaussian probability distribution, ∑:

$\mathrm{\&mu;}=\left[\begin{array}{c}{\mathrm{\&mu;}}_1\\ {\mathrm{\&mu;}}_2\\ .\\ .\\ .\\ {\mathrm{\&mu;}}_D\end{array}\right]$

Wherein, μ is the mean vector of signal intensity；μ_i, i=1,2 ..., D, represent A^TI-th vector average；

Σ=[Cov [A_k,A_j]], k=1,2 ..., D；J=1,2 ..., D

Wherein, A_kFor A^TKth vector, A_jRepresent A^TJth vector；

For the correlation coefficient D dimensional vector of signal, owing to signal communication process is stationary random process, ρ_iRepresent that i ties up the correlation coefficient under interval；σ²Represent the variance of a certain row element in signal intensity matrix x；

Σ σ, ρ_iIt is expressed as:

4. the method utilizing signal time dependency to improve fingerprint indoor position accuracy according to claim 3, it is characterised in that described step 2.2 comprises the steps:

Find one group of orthonormal basis u of correlation matrix ∑_i, i=1,2 ..., m, then transformation matrix U is U=[u₁,u₂,...,u_m]^T, wherein, u_iRepresent i-th orthonormal basis；

If distribution variable y=U (x-μ), then f_r(x_i) transformed after obtain D under new coordinate system and tie up gaussian probability distribution f_r(y | μ, ∑):

$f_r\left(y\right|\mathrm{\&mu;},\mathrm{\&Sigma;})=\frac{1}{{\left(2\mathrm{\&pi;}\right)}^{\frac{D}{2}}{\left({\&Pi;}_{i=1}^m{\mathrm{\&lambda;}}_i\right)}^{\frac{1}{2}}}{e}^{-\frac{1}{2}\left({\mathrm{\&Sigma;}}_{i=1}^m\frac{1}{{\mathrm{\&lambda;}}_i}{y}_i{y_i}^T\right)}$

Wherein, λ_iRepresent ith feature value；Y_iRepresent i-th dimension variable。

5. the method utilizing signal time dependency to improve fingerprint indoor position accuracy according to claim 4, it is characterised in that described step 3 comprises the steps:

Step 3.1: be distributed based on described gaussian probability, carries out border training to sample in probability space；Scaling point r correspondence sample space interval in physical space is E, is defined as:

E={x | f_r(y | μ (r), Σ (r)) >=f_r±δ(y | μ (r ± δ), Σ (r ± δ)) }

Wherein, x represents signal intensity matrix, y represents distribution variable, δ represents error distance, μ (r) represents the mean vector of scaling point r place signal intensity, Σ (r) represents the mean vector of scaling point r ± δ place signal intensity that the correlation matrix at scaling point r place, μ (r ± δ) represent under error distance δ, and Σ (r ± δ) represents the correlation matrix at the scaling point r ± δ place under error distance δ；Distance corresponding in symbol ± expression physical space, wherein, in symbol+expression physical space, correspondence is remote, and in symbol-expression physical space, correspondence is near；

Step 3.2: in the sample space E of location, classification boundaries is hyperbolic curved surface, and the mathematical notation of hyperbolic curved surface is:

$\underset{i=1}{\overset{n}{\&Sigma;}}\underset{j=1}{\overset{m}{\&Sigma;}}\frac{{y_{i,j}}^2}{{\mathrm{\&lambda;}}_{i,j}}-\underset{i=1}{\overset{n}{\&Sigma;}}\&lsqb;\frac{{\left(y_{i,j}\&PlusMinus;\sqrt{2}\mathrm{\&delta;}\&dtri;{\mathrm{\&mu;}}_i^{\&prime;}\right)}^2}{{\mathrm{\&lambda;}}_{i,1}^{\&PlusMinus;}}+\underset{j=2}{\overset{m}{\&Sigma;}}\frac{{y_{i,j}}^2}{{\mathrm{\&lambda;}}_{i,j}^{\&PlusMinus;}}\&rsqb;\&le;\underset{i=1}{\overset{n}{\&Sigma;}}\ln \frac{\det \left({\mathrm{\&Sigma;}}_i\right)}{\det \left({\mathrm{\&Sigma;}}_i^{\&PlusMinus;}\right)}$

Wherein, y_i,jRepresent in i-th signal, the distribution variable of jth dimension in D dimension；λ_i,jRepresent the jth eigenvalue of i-th signal, λ_i,1Represent first eigenvalue of i-th signal, μ '_iRepresent the strength mean value of i-th signal, ∑_iRepresent the correlation matrix of i-th signal, symbolRepresenting gradient function, distance corresponding in subscript ± expression physical space, wherein, in subscript+expression physical space, correspondence is remote, and in subscript-expression physical space, correspondence is near；

Step 3.3: utilize the parameter of the parameter training hyperbolic surface boundary that described gaussian probability is distributed, as the benchmark of location in probability space。

6. the method utilizing signal time dependency to improve fingerprint indoor position accuracy according to claim 1, it is characterised in that described step 5 comprises the steps:

Step 5.1: obtaining Location Request, data user uploaded are compared with signal intensity fingerprint map；

Wherein, signal intensity fingerprint map includes the n dimension mean vector mean that the average of n Wi-Fi signal is constituted:

Mean=[μ₁,μ₂,…,μ_n]

And the D of each Wi-Fi signal ties up (n*D) that correlation coefficient constitutes and ties up correlation vector corr:

Wherein, σ_iRepresent the standard deviation of i-th Wi-Fi signal, wherein, i=1,2 ..., n；ρ_i,jRepresent the jth dimension correlation coefficient of i-th Wi-Fi signal, wherein, i=1,2 ..., n, j=2,3 ..., D；

When comparing, first compare the mean vector mean in fingerprint, find a fingerprint positions of front w relatively more similar in the threshold range set；Again in this w similar fingerprint positions, compare the Euclidean distance of correlation vector corr, find the minimum correlation vector mated most of Euclidean distance in order to determine the final position of user；W<n；

Step 5.2: using every for the data of Location Request adjacent D data as one group, is placed in D with the form of D dimensional vector and ties up in probability space；Under line, the hyperbolic surface boundary of training is as location condition, by the coupling of the correlation coefficient D dimensional vector ρ of signal, finds physical spatial location corresponding for signal intensity matrix x；

Step 5.3: consider the physical spatial location that the comparison of fingerprint is corresponding with signal intensity matrix x, determine final positioning result；The method specifically considering decision is as follows:

First calculate the dependability parameter η of hyperbolic surface boundary, f can be made_r(y | μ (r), Σ (r))=f_r±δThe function maxima that (y | μ (r ± δ), Σ (r ± δ)) sets up；

Then according to η and one threshold value η₀Comparison, η₀Span take (0,0.3)；If η≤η₀, then representing hyperbolic surface boundary reliable, the physical spatial location corresponding for signal intensity matrix x step 5.2 obtained is as positioning result；Otherwise, if i.e. η > η₀, then the positioning result reliability representing hyperbolic surface boundary is not good, then the minimum correlation vector mated most of described Euclidean distance obtained according to step 5.1 determines that the final position of user is as positioning result。

7. the method utilizing signal time dependency to improve fingerprint indoor position accuracy according to claim 1, it is characterised in that also comprise the steps:

Step 7: positioning result is fed back by user, if user feedback positioning result is correct, then puts into the signal intensity matrix x of measurement and trains described gaussian probability to be distributed in data base；If user feedback positioning result is wrong, then the signal intensity matrix x of measurement is not put into and data base trains described gaussian probability be distributed。