CN106162871A - A kind of indoor fingerprint positioning method based on interpolation - Google Patents

Abstract

A kind of indoor fingerprint positioning method based on interpolation, including: step 1, use Kalman filtering to process pending RSSI (received signal strength) value of known location；Step 2, uses Cokriging interpolation algorithm that the RSSI value of the unknown position point in region, location is carried out optimum, unbiased, Linear Estimation by treated known location RSSI sample；All beaconing nodes are repeated step 1 and step 2 by step 3, all one-dimensional fingerprint vector obtained carry out union operation and can successfully build the multidimensional fingerprint vector in region, location；Step 4, uses vector similarity matching algorithm to obtain the grid that similarity front number vector from high to low is corresponding, is i.e. the scope of node locating；Step 5, uses K central point clustering algorithm to classify front nummber the grid obtained, and extracts the cluster head (bunch geometric center) of most bunches of the number containing grid and exports as the actual location result of node.

Description

A kind of indoor fingerprint positioning method based on interpolation

Technical field

The invention mainly relates to wireless sensor network indoor positioning field, relate to indoor fingerprint location side based on interpolation Method.

Background technology

Universal along with sensor network, user for sensor device actual position information attention degree increasingly High.Presently, there are more wireless location technology, as indoor in the outdoor positioning technology such as GPS, AGPS and outer, WIFI, bluetooth etc. fixed Position technology.Wherein GPS relative maturity, its employing positions based on signal transmission time difference TDOA, and this location technology is in outdoor essence Spend higher.In indoor positioning field, due to complexity and the uncontrollability of indoor environment, therefore a lot of indoor positioning technologies are all There is specific application scenarios.

Wireless sensor network indoor positioning technologies achieves significant achievement, fingerprint location technology and base the most both at home and abroad Location technology in RSSI enjoys extensive concern.Certain progress that fingerprint location technology achieves in indoor positioning field, as Novel fingerprint mechanism (NFM) indoor locating system of Ma YW et al. research and development, this system uses receptor and emitter to obtain location Data, and use six location mechanisms to improve positioning precision.Modamed et al. uses gyroscope to tie mutually with fingerprint matching algorithm Positioning result is improved by the indoor orientation method closed, but the orientation method of single gyro instrument has bigger limitation.Li Yan Monarch et al. proposes the localization method utilizing mass-rent to update fingerprint base, but needs periodically expired fingerprint to be carried out sieve and subtract.Xiao Yalong et al. Corresponding physics space length is characterized, it is proposed that a kind of based on multidimensional scaling and region based on signal intensity difference between diverse location The wireless indoor location method of refinement, the method effectively reduces the expense of training stage fingerprint collecting and improves location Precision.The development of the location technology being simultaneously based on RSSI is the rapidest, as Sorour S et al. utilizes adjacent position node The intrinsic spatial coherence of RSSI proposes a kind of localization method based on manifold alignment, and the method uses obtain to have RSSI sky Between the data set of dependency and a part of reference information, obtain position to be measured by semi-supervised learning algorithm；But the method needs Exceptional value is filtered and computation complexity is higher.Gao Renqiang et al. proposes one and combines fuzzy mathematics logic The WiFi location algorithm that (Fuzzy Logic) is theoretical, but handheld device receives signal fluctuation considerable influence positioning precision.Shi Ke Et al. for the problem such as circumstance complication, channel congestion in indoor positioning, it is proposed that a kind of based on support vector regression 802.11 Wireless indoor location method, the method is effectively improved final positioning precision.Xu Kun et al. launches power meeting for the unknown The problem reducing network positions performance proposes a kind of indoor positioning optimized algorithm launching power for the unknown, and this algorithm is fine Improve positioning performance.Promote positioning precision based on propagation loss model and also have certain development, as Wang Yue et al. for Indoor positioning receives power disturbance and causes estimating the problem that positional precision is on the low side, it is proposed that employing horse can husband Monte Carlo (MCMC) sampling improves the indoor positioning algorithms of positioning precision.Chen Wenjian et al. combines geographical coding principle and planar bar code technology Propose the indoor orientation method that QR (Quick Response) mark guides, improve the new platform of innovation and application.

The positioning precision of existing more indoor orientation method is obviously improved, but the precision of fingerprint location method is the highest In propagation model method；Owing to tradition is relatively big based on RSSI fingerprinting localization algorithm workload in setting up fingerprinting process, need survey Amount point carries out RSSI value measurement one by one, and this is a process loaded down with trivial details and that workload is the biggest.

Summary of the invention

The invention solves the problems that the disadvantages mentioned above of prior art, propose a kind of indoor fingerprint positioning method based on interpolation, should Method only needs the RSSI value of a small amount of sample node of known location in measurement and positioning region just can be to the unknown bits in region, location Put a RSSI value and carry out Best unbiased estimator, finally use matching algorithm to calculate the physical location of node, efficiently solve Indoor environment changes and causes loaded down with trivial details fingerprint reconstruction.

Indoor fingerprint positioning method based on interpolation of the present invention, comprises the steps:

Step 1, Kalman filtering；

Step 11, with pending RSSI (received signal strength) value of known location for input, due to RSSI value easily by dry Disturb effect of noise, and Kalman filtering is a kind of efficient algorithm of Gaussian process optimal filter, treated obtain target RSSI Value；

Step 2, Cokriging interpolation algorithm；

Step 21, with step 11 obtain RSSI value as master variable, LQI (communication link quality) value of node be auxiliary variable As input, use an interpolation algorithm in collaborative gram the RSSI value of the unknown position point in region, location is carried out optimum, unbiased, Linear Estimation, concrete steps include for:

Step 211, uses Cokriging interpolation algorithm to find the dependence between sample point, closes first with pairing It is Pair_i=(d_i,RSSI_i)、Pair_j=(d_j,LQI_j)(d_iAnd d_jRepresent the distance between two sensor nodes, RSSI respectively_i And LQI_jRepresent the RSSI value acquired in sensor node and corresponding LQI value) carry out autocorrelation and the sky cared for each other Between model, concrete steps include for:

Step 2111, according to the sample point measured distribution situation spatially, analyzes the semivariable function of autocorrelation, As shown in formula (1)

$\mathrm{\γ}\left(h\right)=\frac{1}{2n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\left(z\right(x_a)-z(x_a+h\left)\right)}^2---\left(1\right)$

In formula (1), n represents that relation is to Pair_iQuantity, z (x_a) and z (x_a+ h) represent the genus of two nodes of pairing respectively Property value (RSSI and LQI), h represents two euclidean distance between node pair, and γ (h) represents semivariable function, by formula (1) obtain relation pair Pair_iAnd Pair_jRespective semivariable function value；

Step 2112, according to the sample point measured distribution situation spatially, analyzes the semivariable function of cross correlation, As formula (2) shows

${\mathrm{\γ}}_{ij}\left(h\right)=\frac{1}{2n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\[z_i\left(x_a\right)-z_i(x_a+h)\]\×\[z_j\left(x_a\right)-z_j(x_a+h)\]---\left(2\right)$

In formula (2), n represents the quantity of Pair, z_i(x_a) and z_i(x_a+ h) respectively represent two nodes RSSI value, z_j(x_a) And z_j(x_a+ h) represent the LQI value of two nodes respectively, and distance is h, γ_ijH () represents mutual semivariable function, pass through formula (2) each Pair is obtained_iSemivariable function value；

Step 2113, with step 2111 obtain relation to Pair_iAnd Pair_jRespective autocorrelative semivariable function and The semivariable function of the cross-correlation that step 2112 obtains is input, then uses method of least square to be fitted setting up relevant half Mutation model；

Step 212, the experience semivariation model of the autocorrelative and cross-correlation obtained with step 2113 is for input, to the unknown The RSSI value of location point is predicted, it was predicted that shown in process equation below

$Z_{0i}=\underset{1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{Z}_i+\underset{1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\λ}}_j{Z}_j---\left(3\right)$

Z in formula (3)_0iRepresent the RSSI value of position to be estimated, Z_iAnd Z_jRepresent respectively apart from the joint that position to be estimated is nearest Point measures RSSI and the LQI value obtained, m and n represents sample size corresponding to RSSI and LQI (with bit attribute m=n) respectively, λ_i And λ_jRepresenting the weight coefficient of RSSI and LQI respectively, concrete calculating process is as follows:

$\underset{1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i=1;\underset{1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\λ}}_j=0---\left(4\right)$

$\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{\mathrm{\γ}}_{11}(x_{1i}-x_i)+\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\λ}}_j{\mathrm{\γ}}_{12}(x_{2j}-x_1)+u_1={\mathrm{\γ}}_{11}(x_0-x_i)---\left(5\right)$

$\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{\mathrm{\γ}}_{21}(x_{1i}-x_1)+\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\λ}}_j{\mathrm{\γ}}_{22}(x_{2j}-x_1)+u_2={\mathrm{\γ}}_{12}(x_0-x_j)---\left(6\right)$

γ in formula (4), formula (5) and formula (6)₁₁And γ₂₂It is Z respectively_iAnd Z (RSSI)_j(LQI) the ideal from variation function Model, γ₁₂And γ₂₁Represent is the ideal model of the cross-variogram of two variablees, wherein γ₁₂=γ₂₁, the weight system of sample Number λ_iAnd λ_jSum is 1 respectively, m and n is respectively the quantity of sample RSSI and LQI, u₁And u₂For Lagrange coefficient, weight Coefficient lambda_iAnd λ_jAll can pass through system of linear equations (4), (5) and (6) to solve；

Step 3, RSSI fingerprint is set up；

Step 31, uses the Cokriging interpolation algorithm of step 2 can set up a beaconing nodes corresponding at positioning area One-dimensional degree fingerprint vector in territory；

Step 32, all beaconing nodes broadcast singals are set up corresponding one-dimensional degree fingerprint vector, then are referred to all one-dimensional degree Stricture of vagina vector carries out union operation, and concrete steps include:

Step 321, the length and width in region, location is respectively L and W, and is divided into m row and n row, the length of each sub-box It is respectively C with width_l=L/m, C_w=W/N；

Step 322, after step 321 will position region D division, calculates each sub-box by Cokriging interpolation algorithm Corresponding RSSI value, thus establishes one-dimensional degree fingerprint f corresponding for beaconing nodes A_A, f_AIt it is the fingerprint of a m × n；

Step 323, repeats step 322, builds one-dimensional degree fingerprint f respectively beaconing nodes B, C, D, E_B、f_C、f_DAnd f_E, Afterwards one-dimensional degree fingerprint is done union and sets up the various dimensions fingerprint vector in region, location, then i-th (i ∈ 1,2,3 ..., m × n}) 5 dimensional vectors that individual grid is corresponding are exactly five one-dimensional degree fingerprints unions in the RSSI value of this position, by 321,322,323 3 Individual step just can set up complete region, location various dimensions fingerprint vector

F_finger=f_A∪f_B∪f_C∪f_D∪f_E (5)

F in formula (5)_i(i ∈ A, B, C, D, E}) represent that beaconing nodes i broadcast singal is at the built-in vertical one-dimensional in region, location Fingerprint vector, F_fingerRepresent a union of 5 one-dimensional fingerprint vector that beaconing nodes A, B, C, D, E are corresponding；

Step 4, mates location algorithm；

Step 41, the multidimensional fingerprint vector obtained with step 323 and the multidimensional RSSI vector that need to position accessed by node For input, vector similarity matching algorithm is used to obtain the grid that similarity front number vector from high to low is corresponding, i.e. Being the scope of node locating, multidimensional RSSI vector is as follows

$T_5=\left[\begin{array}{c}RSS{I}_a\\ {\mathrm{RSSI}}_b\\ {\mathrm{RSSI}}_c\\ {\mathrm{RSSI}}_d\\ {\mathrm{RSSI}}_e\end{array}\right]---\left(6\right)$

RSSI in formula (6)_i(i ∈ { a, b, c, d, e}) represent corresponding to beaconing nodes i broadcast singal time location node institute The RSSI value obtained, T₅Represent the location 5 dimension RSSI fingerprint vector accessed by node；

Step 5, K-central point algorithm cluster extracts positioning result；

Step 51, with front number the grid of step 41 acquisition for input, uses K-key store algorithm to these grid Classify, extract the cluster head (bunch geometric center) of most bunches of the number containing grid and export as the actual location result of node.

The invention have the advantage that

(1) Cokriging interpolation algorithm effectively reduces and sets up workload loaded down with trivial details required for conventional fingerprint builds；

(2) K-central point clustering algorithm reduces range of error effectively, finally reduces the error of location.

Accompanying drawing explanation

The sample pair relationhip figure of Fig. 1 present invention

The cross-variogram fitted figure of Fig. 2 present invention

The prediction procedure chart of Fig. 3 present invention

Fig. 4 sensor of the invention node deployment figure

The location zoning plan of Fig. 5 present invention

The destination node location administrative division map of Fig. 6 present invention

The sub-clustering location figure of Fig. 7 present invention

Fig. 8 is the system schematic realizing the inventive method.

Detailed description of the invention

Further illustrate the present invention below in conjunction with the accompanying drawings.

Indoor fingerprint location localization method based on interpolation of the present invention, comprises the steps: step 1, Kalman Filtering；

Step 11, the present invention is positioned by the method setting up region, location RSSI vector fingerprint.But sensor is adopted The RSSI signal of collection instability, be easily disturbed effect of noise.And Kalman filtering is the one of Gaussian process optimal filter Efficient algorithm, treated obtains target RSSI value；

Step 2, Cokriging interpolation algorithm；

Step 21, Cokriging (Cokriging) algorithm the regionalized variable in finite region can be carried out optimum, Unbiased ground is estimated, therefore, it is possible to the RSSI value prediction being perfectly suitable in region.Cokriging algorithm refers to except main transformer Outside amount, also introduce the one multivariable Krieger algorithm of Cooperative Area variable simultaneously；Co-regionalized variable refers to same In one spatial domain, existing spatial coherence has again one group of variable of statistic correlation feature.The Cokriging that the present invention uses is calculated Method with RSSI as master variable, LQI (link communication quality) be that auxiliary variable calculates；Because RSSI and LQI is in spatial distribution It is proportionate, therefore meets the variable requirement of this algorithm.The two variable, in addition to carrying out autocorrelation prediction, also carries out two changes The prediction of crossing dependency between amount, therefore Cokriging algorithm has higher degree of accuracy in theory.Cokriging is calculated Method concrete implementation process is as follows:

Step 211 sets up dependence；

If using Cokriging interpolation algorithm prediction unknown position RSSI value, first we need to set up between sample Interdependent rule.The present invention uses the sample node R SSI value of collection and LQI value to carry out unknown position RSSI in region, location The prediction of value；First carrying out the pairing between sample, all nodes match two-by-two, set up between sensor node after pairing Pass between distance d and RSSI

System is to Pair_i=(d_i,RSSI_i) and Pair_j=(d_j,LQI_j).Pairing process is as it is shown in figure 1, utilize relation pair Pair carries out the spatial modeling of autocorrelation and cross correlation, specifically comprises the following steps that

Step 2111, according to the sample point distribution situation spatially measured, analyzes the semivariable function of autocorrelation, As shown in formula (1).

$\mathrm{\γ}\left(h\right)=\frac{1}{2n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\left(z\right(x_a)-z(x_a+h\left)\right)}^2---\left(1\right)$

In formula (1), n represents that relation is to Pair_iQuantity, z (x_a) and z (x_a+ h) represent two nodes of pairing respectively RSSI value, h represents two euclidean distance between node pair, and γ (h) represents semivariable function.Each Pair is obtained by formula (1)_iSemivariation Functional value.

Step 2112, according to the sample point distribution situation spatially measured, analyzes the semivariable function of cross correlation, As formula (2) shows.

${\mathrm{\γ}}_{ij}\left(h\right)=\frac{1}{2n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\[z_i\left(x_a\right)-z_i(x_a+h)\]\×\[z_j\left(x_a\right)-z_j(x_a+h)\]---\left(2\right)$

In formula (2), n represents Pair_iQuantity, z_i(x_a) and z_i(x_a+ h) respectively represent two nodes RSSI value, z_j(x_a) And z_j(x_a+ h) represent the LQI value of two nodes respectively, and distance is h, γ_ijH () represents mutual semivariable function, pass through formula (2) each Pair is obtained_iSemivariable function value.

Through formula (1) and formula (2), we have obtained relation to Pair to step 2113_iAutocorrelative semivariance γ (h) (comprising two respective semivariances of attribute of RSSI and LQI) and the semivariance γ of cross-correlation_ij(h) and corresponding two joints Dot spacing, from the relation of h, then uses least square fitting, and relevant semivariation model is set up in matching.Such as Fig. 2 intermediate cam shape Represent Pair_iRelation pair, curve is the result of cross-variogram model of fit, mates an existing and error according to matched curve Minimum model, then generates prediction surface according to statistical model, is finally successfully established dependence.

Step 212 predicted estimate；

The RSSI value of unknown point position is carried out by the semivariation model utilizing autocorrelative semivariation model and cross-correlation Excellent unbiased esti-mator, as shown in formula (3).

$Z_{0i}=\underset{1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{Z}_i+\underset{1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\λ}}_j{Z}_j---\left(3\right)$

Z in formula (3)_0iRepresent the RSSI value of position to be estimated, Z_iAnd Z_jRepresent respectively apart from the joint that position to be estimated is nearest Point measures RSSI and the LQI value obtained, m and n represents sample size corresponding to RSSI and LQI (with bit attribute m=n) respectively.For Meeting the requirement that Cokriging interpolation algorithm Optimal Unbiased difference is estimated, the present invention must is fulfilled for following two condition, as Shown in formula (4).The present invention uses apart from nearest 4 sample points pair of point to be predicted

$\underset{1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i=1,\underset{1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\λ}}_j=0---\left(4\right)$

Position to be estimated carries out the prediction of RSSI value, m=n=4, and introduce two lagrange formulas carry out derivation can Obtain formula (5) and (6).

$\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{\mathrm{\γ}}_{11}(x_{1i}-x_i)+\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\λ}}_j{\mathrm{\γ}}_{12}(x_{2j}-x_1)+u_1={\mathrm{\γ}}_{11}(x_0-x_i)---\left(5\right)$

$\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{\mathrm{\γ}}_{21}(x_{1i}-x_1)+\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\λ}}_j{\mathrm{\γ}}_{22}(x_{2j}-x_1)+u_2={\mathrm{\γ}}_{12}(x_0-x_j)---\left(6\right)$

γ in formula (5) and formula (6)₁₁And γ₂₂It is Z respectively_iAnd Z_jThe ideal model from variation function, γ₁₂And γ₂₁Generation Table is the ideal model of cross-variogram of two variablees, wherein γ₁₂=γ₂₁.Solve system of linear equations formula (4), (5) and (6) Weight coefficient λ can be obtained_iAnd λ_jAnd two Lagrange's multipliers u₁And u₂Value, finally be can get survey region by (3) formula Interior any point Z_0iInterpolate estimation, it was predicted that process use be predicted apart from four known sample points that point to be predicted is nearest, as Shown in Fig. 3.

Step 3RSSI fingerprint is set up；

Step 31, conventional fingerprint build be a complexity and loaded down with trivial details process, need to location region in measurement point by One ground is measured.The present invention uses Cokriging interpolation algorithm to build the fingerprint in region, location, and the method only needs limited sample Point carries out linear unbiased estimation to unknown position point RSSI value in region, location；Once indoor environment changes, and RSSI refers to Stricture of vagina reconstruct is the most very convenient.

The present invention uses the sensor node composition sensor network of 25 known location, wherein 5 anchors for location Node, remaining 20 ordinary nodes build the RSSI fingerprint in region, location, as shown in Figure 4,5 anchor nodes in figure for assisting Broadcast singal successively, the built-in vertical 5 dimension fingerprint vector in region, location.

Step 32, first anchor node A receive the signal of anchor node A broadcast to other node broadcasts signals, residue node And obtain signal intensity RSSI value；Position in known node is distributed in region, location in the present invention is known, and node is equal Can get signal intensity RSSI value and communication link quality LQI value, therefore we just can be with the Pair in establishment step 211_i And Pair_jRelation pair；Use Cokriging interpolation algorithm that the RSSI value of unknown point position in region, location is carried out linearly again Best unbiased estimator；Finally just it is successfully established anchor node A one-dimensional RSSI vector fingerprint in region, location.Remain four anchor joints Point uses same method to build vector fingerprint, then in region, location, 5 dimension RSSI fingerprint vector just complete to build.Region, location Multi-C vector fingerprint build that to implement process as follows:

Step 321, the length and width in region, location is respectively L and W, and is divided into m row n row, the length of each sub-box and Wide respectively C_l=l/m, C_w=w/n, as shown in Figure 5.

Step 322, after previous step will position region division, utilizes the Pair that Node node is set up in this step_i And Pair (RSSI)_j(LQI) with the relation of h, calculate, by Cokriging interpolation algorithm, the RSSI value that each sub-box is corresponding. Thus establish one-dimensional degree fingerprint f corresponding for beaconing nodes A_A, f_AIt it is the fingerprint of a m × n.

Step 323, by previous step, we set up the RSSI fingerprint vector of anchor node A, equally to B, C, D, E anchor node weight Carry out step 2 operation again and just can set up one-dimensional degree fingerprint f respectively_B、f_C、f_DAnd f_E, finally to the f in region, location_A、f_B、 f_C、f_DAnd f_EFive RSSI fingerprint vector do union, as shown in formula (7).

F_finger=f_A∪f_B∪f_C∪f_D∪f_E (7)

Step 4, mates location algorithm；

Step 41, the multidimensional RSSI vector fingerprint in step 323 successfully constructs region, location, for this step to mesh Mark node realizes location.After multidimensional RSSI fingerprint vector successfully constructs, it is fixed all to be withdrawn by the ordinary node in region, location Region, position, destination node enters in region, location, and five anchor nodes are successively to destination node broadcast singal, as shown in Figure 6, red Triangle represents destination node.Destination node first gets orderly 5 and ties up RSSI fingerprint vector T₅, as shown in formula (8), Carry out coupling based on vector similarity algorithm again, extract Number the sub-box that vector similarity is the highest, can obtain The orientation range of positioning result.

$T_5=\left\{\begin{array}{c}RSS{I}_A\\ {\mathrm{RSSI}}_B\\ {\mathrm{RSSI}}_C\\ {\mathrm{RSSI}}_D\\ {\mathrm{RSSI}}_E\end{array}\right\}---\left(8\right)$

Step 5, K-central point algorithm cluster extracts positioning result；

Step 51, because there may be error, so it is the highest to extract vector similarity in step 41 in experimentation Number sub-box exists some error sub-boxes.The present invention uses K-central cluster algorithm range of error to be rejected, step Rapid 41 Number the sub-boxes extracted are distributed as it is shown in fig. 7, diagonal line hatches part represents the grid that error is bigger, these grid Quantity fewer, the grid of full shadow represents the grid that precision is higher, and these grid quantity are the most.We use k-center Point clustering algorithm Number result is clustered, by obtain one containing grid number maximum bunch, the cluster head conduct of this bunch The physical location of destination node.If it is sufficiently fine that m × n grid is divided by we, then the position error of node controls connecing Within the scope of being subject to.

The result of destination node location as shown in the triangle position in Fig. 7, this algorithm using the geometric center of grid as Positioning result.Because we are divided into m × n sub-box location region D, as long as divide is enough fine, then position error Just caning be controlled in can be in received scope.

Claims (1)

1. an indoor fingerprint positioning method based on interpolation, it is characterised in that comprise the steps:

Step 1, Kalman filtering；

Step 11, with the pending received signal strength RSSI value of known location for input, is easily disturbed noise due to RSSI value Impact, and Kalman filtering is a kind of efficient algorithm of Gaussian process optimal filter, treated obtains target RSSI value；

Step 2, Cokriging interpolation algorithm；

Step 21, using step 11 obtain RSSI value as master variable, the communication link quality LQI value of node be that auxiliary variable is as defeated Enter, use an interpolation algorithm in collaborative gram that the RSSI value of the unknown position point in region, location is carried out optimum, unbiased, linearly estimated Meter, concrete steps include for:

Step 211, uses Cokriging interpolation algorithm to find the dependence between sample point, first with pair relationhip Pair_i=(d_i,RSSI_i)、Pair_j=(d_j,LQI_j) carry out autocorrelation and the spatial modeling cared for each other, d_iAnd d_jRespectively Represent the distance between two sensor nodes, RSSI_iAnd LQI_jRepresent the RSSI value acquired in sensor node and corresponding LQI Value, concrete steps include for:

Step 2111, according to the sample point measured distribution situation spatially, analyzes the semivariable function of autocorrelation, such as public affairs Shown in formula (1)

$\mathrm{\γ}\left(h\right)=\frac{1}{2n}\underset{i=1}{\overset{n}{\Σ}}{\left(z\right(x_a)-z(x_a+h\left)\right)}^2---\left(1\right)$

In formula (1), n represents that relation is to Pair_iQuantity, z (x_a) and z (x_a+ h) represent the property value of two nodes of pairing respectively RSSI and LQI, h represent two euclidean distance between node pair, and γ (h) represents semivariable function, obtain relation to Pair by formula (1)_iWith Pair_jRespective semivariable function value；

Step 2112, according to the sample point measured distribution situation spatially, analyzes the semivariable function of cross correlation, such as public affairs Formula (2) is shown

${\mathrm{\γ}}_{ij}\left(h\right)=\frac{1}{2n}\underset{i=1}{\overset{n}{\Σ}}\[z_i\left(x_a\right)-z_i(x_a+h)\]\×\[z_j\left(x_a\right)-z_j(x_a+h)\]---\left(2\right)$

In formula (2), n represents the quantity of Pair, z_i(x_a) and z_i(x_a+ h) respectively represent two nodes RSSI value, z_j(x_a) and z_j (x_a+ h) represent the LQI value of two nodes respectively, and distance is h, γ_ijH () represents mutual semivariable function, by formula (2) Obtain each Pair_iSemivariable function value；

Step 2113, with step 2111 obtain relation to Pair_iAnd Pair_jRespective autocorrelative semivariable function and step The semivariable function of 2112 cross-correlation obtained is input, then uses method of least square to be fitted setting up relevant semivariation Model；

Step 212, the experience semivariation model of the autocorrelative and cross-correlation obtained with step 2113 is for input, to unknown position The RSSI value of point is predicted, it was predicted that shown in process equation below

$Z_{0i}=\underset{1}{\overset{m}{\Σ}}{\mathrm{\λ}}_i{Z}_i+\underset{1}{\overset{n}{\Σ}}{\mathrm{\λ}}_j{Z}_j---\left(3\right)$

Z in formula (3)_0iRepresent the RSSI value of position to be estimated, Z_iAnd Z_jRepresent respectively apart from the node measurement that position to be estimated is nearest RSSI and the LQI value obtained, m and n represents sample size corresponding to RSSI and LQI respectively, with bit attribute m=n, λ_iAnd λ_jRespectively Representing the weight coefficient of RSSI and LQI, concrete calculating process is as follows:

$\underset{1}{\overset{m}{\Σ}}{\mathrm{\λ}}_i=1;\underset{1}{\overset{n}{\Σ}}{\mathrm{\λ}}_j=0---\left(4\right)$

$\underset{i=1}{\overset{m}{\Σ}}{\mathrm{\λ}}_i{\mathrm{\γ}}_{11}(x_{1i}-x_i)+\underset{j=1}{\overset{n}{\Σ}}{\mathrm{\λ}}_j{\mathrm{\γ}}_{12}(x_{2j}-x_1)+u_1={\mathrm{\γ}}_{11}(x_0-x_i)---\left(5\right)$

$\underset{i=1}{\overset{m}{\Σ}}{\mathrm{\λ}}_i{\mathrm{\γ}}_{21}(x_{1i}-x_1)+\underset{j=1}{\overset{n}{\Σ}}{\mathrm{\λ}}_j{\mathrm{\γ}}_{22}(x_{2j}-x_1)+u_2={\mathrm{\γ}}_{12}(x_0-x_j)---\left(6\right)$

γ in formula (4), formula (5) and formula (6)₁₁And γ₂₂It is Z respectively_iAnd Z (RSSI)_j(LQI) the ideal mode from variation function Type, γ₁₂And γ₂₁Represent is the ideal model of the cross-variogram of two variablees, wherein γ₁₂=γ₂₁, the weight coefficient of sample λ_iAnd λ_jSum is 1 respectively, m and n is respectively the quantity of sample RSSI and LQI, u₁And u₂For Lagrange coefficient, weight system Number λ_iAnd λ_jAll can pass through system of linear equations (4), (5) and (6) to solve；

Step 3, RSSI fingerprint is set up；

Step 31, uses the Cokriging interpolation algorithm of step 2 can set up a beaconing nodes corresponding in region, location One-dimensional degree fingerprint vector；

Step 32, all beaconing nodes broadcast singals set up corresponding one-dimensional degree fingerprint vector, then to all one-dimensional degree fingerprints to Amount carries out union operation, and concrete steps include:

Step 321, the length and width in region, location is respectively L and W, and is divided into m row and n row, the length of each sub-box and width It is respectively C_l=L/m, C_w=W/N；

Step 322, after step 321 will position region D division, calculates each sub-box by Cokriging interpolation algorithm corresponding RSSI value, thus establish one-dimensional degree fingerprint f corresponding for beaconing nodes A_A, f_AIt it is the fingerprint of a m × n；

Step 323, repeats step 322, builds one-dimensional degree fingerprint f respectively beaconing nodes B, C, D, E_B、f_C、f_DAnd f_E, the most right One-dimensional degree fingerprint does union and sets up the various dimensions fingerprint vector in region, location, then 5 dimensional vectors that i-th grid is corresponding are exactly five Individual one-dimensional degree fingerprint in the union of the RSSI value of this position, i ∈ 1,2,3 ..., m × n}, by 321,322,323 3 steps Just can set up complete region, location various dimensions fingerprint vector,

F_finger=f_A∪f_B∪f_C∪f_D∪f_E (5)

F in formula (5)_iRepresent that beaconing nodes i broadcast singal is positioning the built-in vertical one-dimensional fingerprint vector in region, i ∈ A, B, C, D, E}, F_fingerRepresent a union of 5 one-dimensional fingerprint vector that beaconing nodes A, B, C, D, E are corresponding；

Step 4, mates location algorithm；

Step 41, the multidimensional fingerprint vector obtained with step 323 and need to position the multidimensional RSSI vector accessed by node be defeated Enter, use vector similarity matching algorithm to obtain the grid that similarity front number vector from high to low is corresponding, be i.e. joint The scope of point location, multidimensional RSSI vector is as follows

$T_5=\left[\begin{array}{c}{\mathrm{RSSI}}_a\\ {\mathrm{RSSI}}_b\\ {\mathrm{RSSI}}_c\\ {\mathrm{RSSI}}_d\\ {\mathrm{RSSI}}_e\end{array}\right]---\left(6\right)$

RSSI in formula (6)_iLocation RSSI value acquired in node during beaconing nodes i broadcast singal corresponding to expression, i ∈ a, b, C, d, e}, T₅Represent the location 5 dimension RSSI fingerprint vector accessed by node；

Step 5, K-central point algorithm cluster extracts positioning result；

Step 51, with front number the grid of step 41 acquisition for input, uses K-key store algorithm to carry out these grid Classification, extracts the cluster head (bunch geometric center) of most bunches of the number containing grid and exports as the actual location result of node, cluster head The geometric center being bunch.