CN106304331A - A kind of WiFi fingerprint indoor orientation method - Google Patents

Abstract

The invention discloses a kind of WiFi fingerprint indoor orientation method, comprise the following steps: processed offline step, including: (11), region, delimitation location, set up the location fingerprint storehouse LFDB:(12 of described L sampled point), described location fingerprint storehouse LFDB is carried out k mean cluster, using Euclidean distance as the interpretational criteria of similarity, fingerprint base is divided into the fingerprint base with k subclass；Tuning on-line step, including: (21), calculating actual measurement fingerprint and the distance at each class center, it is designated as；(22), find the class corresponding to minima in DIS, be designated as；(23), calculate actual measurement fingerprint with in the distance of each fingerprint；(24), selected reference fingerprint；(25) weight coefficient of each reference fingerprint, is calculated；(26) position coordinates of actual measurement fingerprint, is calculated.The RSSI value of off-line data sample phase collection is clustered by this method k mean algorithm, reduces the amount of calculation of fingerprint matching process；In the tuning on-line stage, employ the method processed of dividing and ruling, reduce position error.

Description

A kind of WiFi fingerprint indoor orientation method

Technical field

The present invention relates to a kind of localization method, specifically, relate to a kind of WiFi fingerprint indoor orientation method.

Background technology

For the research of indoor locating system, had typical indoor locating system such as RADAT system, HORUS system, LANDMARC system etc..RADAR indoor locating system is based on RSSI (Received Signal Strength Indication) method of indoor propagation model, by the dependency of sampled point in analysis room Yu RSSI, introduces what wall brought Decay, and estimate decay factor by the method for linear regression, it is used for compensating indoor propagation loss model.Compensated by this Model, calculates the angle between RSSI receptor and emitter, orients the position of receptor in conjunction with triangle polyester fibre algorithm.

HORUS wireless location system, including sub-clustering module, discrete space estimation module, relevant and processing module, continuously sky Between estimator.HORUS system has the positioning precision higher than RADAR system, but owing to structure is complicated, processing module is more, The computation complexity of alignment system is bigger.

LANDMARC alignment system is dynamic positioning identification system based on active RFID verification, and it uses location with reference to mark Signing auxiliary positioning, these reference label serve as the location reference point of system.

Current WiFi fingerprinting localization algorithm has two megastages, Database and positioning stage.In position estimation procedure In, the nearest neighbor algorithm currently used, using fingerprint maximum to actual measurement fingerprint and similarity in fingerprint base as positioning result, algorithm Fairly simple, but owing to the position of reference is single, do not get rid of the interference of outlier, there is a lot of noises, can be to position The coupling of fingerprint causes maximum error to affect.

Summary of the invention

The present invention is to solve the nearest neighbor algorithm that existing WiFi fingerprint indoor positioning uses, will actual measurement fingerprint and fingerprint base The fingerprint of middle similarity maximum is as positioning result, and method comparison is simple, but owing to the position of reference is single, does not get rid of nothing Close the interference of item, there is a lot of noises, the coupling of location fingerprint can be caused the problem that maximum error affects, it is proposed that be a kind of WiFi fingerprint indoor orientation method, can solve the problems referred to above.

In order to solve above-mentioned technical problem, the present invention is achieved by the following technical solutions:

A kind of WiFi fingerprint indoor orientation method, comprises the following steps:

Processed offline step, including:

(11), delimit region, location, and lay n WiFi discharger, in region, described location in region, location Interior selected L sampled point, calculates described n the WiFi discharger received by each sampled point and launches intensity RSSI of signal, N RSSI value (rssi can be observed at each sampled point¹,rssi²..., rssiⁿ), and this n RSSI value is adopted as this The fingerprint of sampling point, the position coordinates of sampled point be (x, y), the position one_to_one corresponding of each fingerprint and its sampled point, set up institute State the location fingerprint storehouse LFDB of L sampled point:

$LFDB={\left[\begin{array}{c}{x}_1,y_1,{{\mathrm{rssi}}_1}^1,{{\mathrm{rssi}}_1}^2,...,{{\mathrm{rssi}}_1}^n\\ x_2,y_2,{\mathrm{rssi}}_2^1,{\mathrm{rssi}}_2^2,...,{\mathrm{rssi}}_2^n\\ \mathrm{......}\\ x_L,y_L,{\mathrm{rssi}}_L^1,{\mathrm{rssi}}_L^2,...,{\mathrm{rssi}}_L^n\end{array}\right]}_{L\×(n+2)};$

(12), described location fingerprint storehouse LFDB is carried out k-mean cluster, accurate as the evaluation of similarity using Euclidean distance Then, fingerprint base is divided into the fingerprint base KFp with k subclass, and wherein, n, L, k are positive integer；

Tuning on-line step, including:

(21), will actual measurement fingerprint lf=(rssi₁,rssi₂,…,rssi_n) carry out with the fingerprint base KFp after cluster Join, calculate the distance of lf and each class center, be designated as DIS=[d₁,d₂,…,d_k]；

(22), find the class corresponding to minima in DIS, be designated as G_SPECIAL；

(23) actual measurement fingerprint lf and G, is calculated_SPECIALIn the distance of each fingerprint, be designated asIts Middle p represents G_SPECIALIn the number of fingerprint, for positive integer；

(24), the data in DIS are arranged according to order from small to large, take front m value, and by described m value correspondence Fingerprint selected as reference fingerprint, the position coordinates of its correspondence is as reference coordinate；

(25) the weight coefficient ω of each reference fingerprint, is calculated_i(i=1,2 ..., m)；

(26), according to weight coefficient and the position coordinates (x of reference coordinate calculating actual measurement fingerprint of each reference fingerprint_estimate, y_estimate)。

Further, in step (12), fingerprint base is carried out k-mean cluster and comprises the following steps:

(121), L fingerprint of inputWith cluster number k (0 ＜ k≤L), from L fingerprint arbitrarily select k fingerprint as initial cluster centre

(122), for remaining L-k fingerprint, each fingerprint distance Distance=to each cluster centre is calculated {d_uvU=1,2 ..., (L-k)；V=1,2 ..., k}, wherein d_uvRepresent the u fingerprint to v in remaining L-k fingerprint The distance of cluster centre, finds the minima in Distance, by right for the minima institute in the u fingerprint classification to Distance In the cluster answered, obtain new cluster result, remaining fingerprint is assigned, form k cluster G₁,G₂,…,G_v,…G_k, Each cluster G_vAll comprise its cluster centre, and belong to such fingerprint member and number n thereof_v；

(123), recalculate the cluster centre of each class, obtain new cluster centre

(124), new cluster centre is compared with previous cluster centre, if both are identical, the most adjacent twice Cluster centre is identical, i.e. classification tends towards stability, and cluster terminates, current G₁,G₂,…,G_v,…G_kRepresent the cluster ultimately formed, Otherwise make C=C^*, i.e. update class center, return step (122), continue executing with cluster process.

Further, in step (123), according to formulaRecalculate the cluster of each class Center, wherein rssi_qRepresent G_vThe q-th RSSI value of apoplexy due to endogenous wind, wherein, 0 ＜ q ＜ n_v。

Further, in step (25), the calculating of weight coefficient is relevant to the fingerprint of fingerprint base, by maximum to similarity M fingerprint carry out related operation, calculate the percentage contribution of each fingerprint, and this percentage contribution be mapped to the position of correspondence Information, carries out the estimation of position:

Assume the fingerprint space Fp maximum with actual measurement fingerprint similarity^*There is a m fingerprint:

${\mathrm{Fp}}^{*}={\left[\begin{array}{c}{\mathrm{rssi}}_1^1,{\mathrm{rssi}}_1^2,...,{\mathrm{rssi}}_1^n\\ {\mathrm{rssi}}_2^1,{\mathrm{rssi}}_2^2,...,{\mathrm{rssi}}_2^n\\ .\\ .\\ .\\ {\mathrm{rssi}}_m^1,{\mathrm{rssi}}_m^2,...,{\mathrm{rssi}}_m^n\end{array}\right]}_{m\×n}$

Calculate Fp^*In the weight coefficient ω of each fingerprint_i, by Fp^*Carry out transposition (Tp^*)^T；Calculate each reference point respectively to refer to The average of stricture of vagina and standard deviation:

$\stackrel{\‾}{{\mathrm{rssi}}_i}=\frac{1}{n}\underset{a=1}{\overset{n}{\Σ}}{\mathrm{rssi}}_i^a$

$s_i=\sqrt{\frac{1}{n-1}\underset{a=1}{\overset{n}{\Σ}}{({\mathrm{rssi}}_i^a-\stackrel{\‾}{{\mathrm{rssi}}_i})}^2}$

Wherein, i=1,2 ..., m,

OrderThen weight coefficient is

${\mathrm{\ω}}_i={\mathrm{\υ}}_i/\underset{i=1}{\overset{n}{\Σ}}{\mathrm{\υ}}_i.$

Further, in step (26), the position coordinates of actual measurement fingerprint is:

$x_{estimate}=\underset{i=1}{\overset{m}{\Σ}}{\mathrm{\ω}}_i{x}_i$

$y_{estimate}=\underset{i=1}{\overset{m}{\Σ}}{\mathrm{\ω}}_i{y}_i;$

Wherein, x_i, y_iReference coordinate for i-th reference fingerprint.

Compared with prior art, advantages of the present invention and good effect are: present invention k-mean algorithm is to off-line data The RSSI value of sample phase collection clusters, and rejects the RSSI information of Outliers, utilizes and remaining aligns standby RSSI Set up new fingerprint base, reduce the amount of calculation of fingerprint matching process；In the tuning on-line stage, employ the method processed of dividing and ruling, Physical location is progressively got rid of, thus reduces the screening amount of data, range shorter will be tested simultaneously.Use actual measurement fingerprint with The thought of fingerprint base coupling, improves nearest neighbor algorithm, adds signal attenuation model, is weighted data processing, this Unknown position is being estimated by sample, reduces position error.

After reading in conjunction with the accompanying the detailed description of embodiment of the present invention, the other features and advantages of the invention will become more Add clear.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing In having technology to describe, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is only this Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to Other accompanying drawing is obtained according to these accompanying drawings.

Fig. 1 is a kind of embodiment flow chart of WiFi fingerprint indoor orientation method proposed by the invention.

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under not making creative work premise Embodiment, broadly falls into the scope of protection of the invention.

Embodiment one, the present embodiment proposes a kind of WiFi fingerprint indoor orientation method, as it is shown in figure 1, include following step Rapid:

Processed offline step, including:

S11, region, delimitation location, and lay n WiFi discharger, in region, described location in region, location Selected L sampled point, calculates described n the WiFi discharger received by each sampled point and launches intensity RSSI of signal, Each sampled point can observe n RSSI value (rssi¹,rssi²..., rssiⁿ), and using this n RSSI value as this sampling The fingerprint of point, the position coordinates of sampled point be (x, y), the position one_to_one corresponding of each fingerprint and its sampled point, set up described L The location fingerprint storehouse LFDB of individual sampled point:

$LFDB={\left[\begin{array}{c}{x}_1,y_1,{{\mathrm{rssi}}_1}^1,{{\mathrm{rssi}}_1}^2,...,{{\mathrm{rssi}}_1}^n\\ x_2,y_2,{\mathrm{rssi}}_2^1,{\mathrm{rssi}}_2^2,...,{\mathrm{rssi}}_2^n\\ \mathrm{......}\\ x_L,y_L,{\mathrm{rssi}}_L^1,{\mathrm{rssi}}_L^2,...,{\mathrm{rssi}}_L^n\end{array}\right]}_{L\×(n+2)};$

S12, described location fingerprint storehouse LFDB is carried out k-mean cluster, accurate as the evaluation of similarity using Euclidean distance Then, fingerprint base is divided into the fingerprint base KFp with k subclass, and wherein, n, L, k are positive integer；With k-mean algorithm to data The RSSI value of off-line sample phase collection clusters, reject Outliers RSSI information, utilize remaining align standby RSSI sets up new fingerprint base, reduces the amount of calculation of fingerprint matching process；

Tuning on-line step, including:

S21, general actual measurement fingerprint lf=(rssi₁,rssi₂,…,rssi_n) mate with the fingerprint base KFp after cluster, Calculate the distance of lf and each class center, be designated as DIS=[d₁,d₂,…,d_k]；

S22, find the class corresponding to minima in DIS, be designated as G_SPECIAL；

S23, calculating actual measurement fingerprint lf and G_SPECIALIn the distance of each fingerprint, be designated asIts Middle p represents G_SPECIALIn the number of fingerprint, for positive integer；

S24, the data in DIS are arranged according to order from small to large, take front m value, and be worth correspondence by described m Fingerprint is selected as reference fingerprint, and the position coordinates of its correspondence is as reference coordinate；

S25, calculate the weight coefficient ω of each reference fingerprint_i(i=1,2 ..., m)；

S26, calculate the position coordinates (x of actual measurement fingerprint according to the weight coefficient of each reference fingerprint and reference coordinate_estimate, y_estimate)。

In the tuning on-line stage, employ the method processed of dividing and ruling, physical location is progressively got rid of, thus reduces number According to screening amount, range shorter will be tested simultaneously.The thought using actual measurement fingerprint to mate with fingerprint base, is carried out nearest neighbor algorithm Improve, add signal attenuation model, be weighted data processing, so unknown position estimated, reducing location Error.

As a preferred embodiment, in step S12, fingerprint base is carried out k-mean cluster and comprises the following steps:

S121, L fingerprint of inputWith cluster number k (0 ＜ k≤L), from L Individual fingerprint arbitrarily select k fingerprint as initial cluster centre

S122, for remaining L-k fingerprint, calculate each fingerprint distance Distance=to each cluster centre {d_uvU=1,2 ..., (L-k)；V=1,2 ..., k}, wherein d_uvRepresent the u fingerprint to v in remaining L-k fingerprint The distance of cluster centre, finds the minima in Distance, by right for the minima institute in the u fingerprint classification to Distance In the cluster answered, obtain new cluster result, remaining fingerprint is assigned, form k cluster G₁,G₂,…,G_v,…G_k, Each cluster G_vAll comprise its cluster centre, and belong to such fingerprint member and number n thereof_v；

S123, recalculate the cluster centre of each class, obtain new cluster centre

S124, new cluster centre is compared with previous cluster centre, if both are identical, the most adjacent twice poly- Class center is identical, i.e. classification tends towards stability, and cluster terminates, current G₁,G₂,…,G_v,…G_kRepresent the cluster ultimately formed, no Then make C=C^*, i.e. update class center, return step S122, continue executing with cluster process.

In step S123, according to formulaRecalculate the cluster centre of each class, wherein rssi_qRepresent G_vThe q-th RSSI value of apoplexy due to endogenous wind, wherein, 0 ＜ q ＜ n_v.It is arbitrarily k fingerprint conduct of selection during due to initial calculation Initial cluster centre, according to Euclidean distance as the interpretational criteria of similarity after calculating is carried out, assembles away from similar fingerprint A subclass, apart from bigger fingerprint away from each other, the data inside subclass there occurs change, and therefore cluster centre is sent out accordingly Changing, in order to improve precision, the cluster centre recalculating each class that should adapt.

In step S25, the calculating of weight coefficient is relevant to the fingerprint of fingerprint base, by m the fingerprint maximum to similarity Carry out related operation, calculate the percentage contribution of each fingerprint, and this percentage contribution is mapped to corresponding positional information, carry out The estimation of position:

Assume the fingerprint space Fp maximum with actual measurement fingerprint similarity^*There is a m fingerprint:

${\mathrm{Fp}}^{*}={\left[\begin{array}{c}{\mathrm{rssi}}_1^1,{\mathrm{rssi}}_1^2,...,{\mathrm{rssi}}_1^n\\ {\mathrm{rssi}}_2^1,{\mathrm{rssi}}_2^2,...,{\mathrm{rssi}}_2^n\\ .\\ .\\ .\\ {\mathrm{rssi}}_m^1,{\mathrm{rssi}}_m^2,...,{\mathrm{rssi}}_m^n\end{array}\right]}_{m\×n}$

Calculate Fp^*In the weight coefficient ω of each fingerprint_i, by Fp^*Carry out transposition (Tp^*)^T；Calculate each reference point respectively to refer to The average of stricture of vagina and standard deviation:

$\stackrel{\‾}{{\mathrm{rssi}}_i}=\frac{1}{n}\underset{a=1}{\overset{n}{\Σ}}{\mathrm{rssi}}_i^a$

$s_i=\sqrt{\frac{1}{n-1}\underset{a=1}{\overset{n}{\Σ}}{({\mathrm{rssi}}_i^a-\stackrel{\‾}{{\mathrm{rssi}}_i})}^2}$

Wherein, i=1,2 ..., m,

OrderThen weight coefficient is

${\mathrm{\ω}}_i={\mathrm{\υ}}_i/\underset{i=1}{\overset{n}{\Σ}}{\mathrm{\υ}}_i.$

In the application, the selection of weight coefficient is relevant to the fingerprint of fingerprint base, is by k maximum to similarity Fingerprint carries out the computing being correlated with, and calculates the percentage contribution of each fingerprint, and this percentage contribution is mapped to the position letter of correspondence Breath, carries out the estimation of position.

In step S26, the position coordinates of actual measurement fingerprint is:

$x_{estimate}=\underset{i=1}{\overset{m}{\Σ}}{\mathrm{\ω}}_i{x}_i$

$y_{estimate}=\underset{i=1}{\overset{m}{\Σ}}{\mathrm{\ω}}_i{y}_i;$

Wherein, x_i, y_iReference coordinate for i-th reference fingerprint.

Certainly, described above is not limitation of the present invention, and the present invention is also not limited to the example above, and this technology is led Change that the those of ordinary skill in territory is made in the essential scope of the present invention, retrofit, add or replace, also should belong to this Bright protection domain.

Claims (5)

1. a WiFi fingerprint indoor orientation method, it is characterised in that comprise the following steps:

Processed offline step, including:

(11), delimit region, location, and lay n WiFi discharger in region, location, select in region, described location Determine L sampled point, calculate described n the WiFi discharger received by each sampled point and launch intensity RSSI of signal, often Individual sampled point can observe n RSSI value (rssi¹,rssi²..., rssiⁿ), and using this n RSSI value as this sampled point Fingerprint, the position coordinates of sampled point be (x, y), the position one_to_one corresponding of each fingerprint and its sampled point, set up described L The location fingerprint storehouse LFDB of sampled point:

$LFDB={\left[\begin{array}{c}{x}_1,y_1,rss{i_1}^1,rss{i_1}^2,...,rss{i_1}^n\\ x_2,y_2,{\mathrm{rssi}}_2^1,{\mathrm{rssi}}_2^2,...,{\mathrm{rssi}}_2^n\\ ......\\ x_L,y_L,rss{i}_L^1,rss{i}_L^2,...,rss{i}_L^n\end{array}\right]}_{L\×(n+2)};$

(12), described location fingerprint storehouse LFDB is carried out k-mean cluster, using Euclidean distance as the interpretational criteria of similarity, refer to Stricture of vagina storehouse is divided into the fingerprint base KFp with k subclass, and wherein, n, L, k are positive integer；

Tuning on-line step, including:

(21), will actual measurement fingerprint lf=(rssi₁,rssi₂,…,rssi_n) mate with the fingerprint base KFp after cluster, meter Calculate the distance of lf and each class center, be designated as DIS=[d₁,d₂,…,d_k]；

(22), find the class corresponding to minima in DIS, be designated as G_SPECIAL；

(23) actual measurement fingerprint lf and G, is calculated_SPECIALIn the distance of each fingerprint, be designated asWherein p table Show G_SPECIALIn the number of fingerprint, for positive integer；

(24), the data in DIS are arranged according to order from small to large, take front m value, and by finger corresponding for described m value Stricture of vagina is selected as reference fingerprint, and the position coordinates of its correspondence is as reference coordinate；

(25) the weight coefficient ω of each reference fingerprint, is calculated_i(i=1,2 ..., m)；

(26), according to weight coefficient and the position coordinates (x of reference coordinate calculating actual measurement fingerprint of each reference fingerprint_estimate, y_estimate)。

WiFi fingerprint indoor orientation method the most according to claim 1, it is characterised in that in step (12), by fingerprint base Carry out k-mean cluster to comprise the following steps:

(121), L fingerprint of inputWith cluster number k (0 ＜ k≤L), from L Fingerprint arbitrarily select k fingerprint as initial cluster centre

(122), for remaining L-k fingerprint, each fingerprint distance Distance={d to each cluster centre is calculated_uv|u =1,2 ..., (L-k)；V=1,2 ..., k}, wherein d_uvIn representing that in remaining L-k fingerprint, the u fingerprint to the v clusters The distance of the heart, finds the minima in Distance, by gathering corresponding to the minima in the u fingerprint classification to Distance Apoplexy due to endogenous wind, obtains new cluster result, is assigned by remaining fingerprint, forms k cluster G₁,G₂,…,G_v,…G_k, Mei Geju Class G_vAll comprise its cluster centre, and belong to such fingerprint member and number n thereof_v；

(123), recalculate the cluster centre of each class, obtain new cluster centre

(124), new cluster centre is compared with previous cluster centre, if both are identical, the cluster of the most adjacent twice Center is identical, i.e. classification tends towards stability, and cluster terminates, current G₁,G₂,…,G_v,…G_kRepresent the cluster ultimately formed, otherwise Make C=C^*, i.e. update class center, return step (122), continue executing with cluster process.

WiFi fingerprint indoor orientation method the most according to claim 2, it is characterised in that in step (123), according to formulaRecalculate the cluster centre of each class, wherein rssi_qRepresent G_vThe q-th RSSI value of apoplexy due to endogenous wind, Wherein, 0 ＜ q ＜ n_v。

4. according to the WiFi fingerprint indoor orientation method described in any one of claim 1-3, it is characterised in that in step (25), The calculating of weight coefficient is relevant to the fingerprint of fingerprint base, carries out related operation by m the fingerprint maximum to similarity, calculates every The percentage contribution of individual fingerprint, and this percentage contribution is mapped to the positional information of correspondence, carry out the estimation of position:

Assume the fingerprint space Fp maximum with actual measurement fingerprint similarity^*There is a m fingerprint:

${\mathrm{Fp}}^{*}={\left[\begin{array}{c}{\mathrm{rssi}}_1^1,{\mathrm{rssi}}_1^2,...,{\mathrm{rssi}}_1^n\\ {\mathrm{rssi}}_2^1,{\mathrm{rssi}}_2^2,...,{\mathrm{rssi}}_2^n\\ \·\\ \·\\ \·\\ {\mathrm{rssi}}_m^1,{\mathrm{rssi}}_m^2,...,{\mathrm{rssi}}_m^n\end{array}\right]}_{m\×n}$

Calculate Fp^*In the weight coefficient ω of each fingerprint_i, by Fp^*Carry out transposition (Tp^*)^T；Calculate each reference point fingerprint respectively Average and standard deviation:

$\stackrel{\‾}{{\mathrm{rssi}}_i}=\frac{1}{n}\underset{a=1}{\overset{n}{\Σ}}{\mathrm{rssi}}_i^a$

$s_i=\sqrt{\frac{1}{n-1}\underset{a=1}{\overset{n}{\Σ}}{({\mathrm{rssi}}_i^a-\stackrel{\‾}{{\mathrm{rssi}}_i})}^2}$

Wherein, i=1,2 ..., m,

OrderThen weight coefficient is

${\mathrm{\ω}}_i={\mathrm{\υ}}_i/\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\υ}}_i.$

WiFi fingerprint indoor orientation method the most according to claim 4, it is characterised in that in step (26), surveys fingerprint Position coordinates be:

$x_{estimate}=\underset{i=1}{\overset{m}{\Σ}}{\mathrm{\ω}}_i{x}_i$

$y_{estimate}=\underset{i=1}{\overset{m}{\Σ}}{\mathrm{\ω}}_i{y}_i;$

Wherein, x_i, y_iReference coordinate for i-th reference fingerprint.