CN106093852A - A kind of method improving WiFi fingerprint location precision and efficiency - Google Patents

Abstract

The present invention relates to a kind of method improving WiFi fingerprint location precision and efficiency, in off-line training step, build the fingerprint database for tuning on-line, and to fingerprint database using K means clustering algorithm classify, wherein, the signal strength values of the sampled point in fingerprint database processes through data smoothing；In the tuning on-line stage, using K the fingerprint that k nearest neighbor algorithm is sought closest with actual measurement fingerprint, the position of the average correspondence sampled point of K fingerprint is the estimation position of tested point.The present invention can improve positioning precision and location efficiency.

Description

A kind of method improving WiFi fingerprint location precision and efficiency

Technical field

The present invention relates to indoor positioning technologies field based on WiFi, particularly relate to a kind of raising WiFi fingerprint location essence Degree and the method for efficiency.

Background technology

Indoor locating system is one of focus of current areas of information technology, fast along with Internet of Things and wireless communication technology Speed development, location Based service presents wide application prospect in fields such as health care, public safety, commercial production.

Global positioning system (Global Positioning System, GPS) is the location skill being widely used present stage Art, it is widely used in various location-based service.But GPS alignment system cannot position in indoor, because this location Method needs the satellite of more than three to provide location information, is generally only applicable to spacious unsheltered outdoor environment, Under the indoor environment more closed, GPS alignment system cannot obtain the information needed for location by satellite.As can be seen here, GPS alignment system is only applicable to outdoor positioning, and cannot meet the location requirement in diversified indoor environment.Now, based on The WiFi location technology of WLAN heats up rapidly, and the most most widely used is exactly WiFi fingerprint location technology, and WiFi refers to Stricture of vagina location technology is the technology in wireless location technology with degree of precision and exploitativeness, and it need not extra hardware and sets Execute, cheap, therefore there is the strongest practicality.

WiFi fingerprint location comes from database-located technology, and it needs to be pre-created fingerprint database, in fingerprint database Deposit is signal intensity and the position coordinates of off-line.Owing to the multipath transmisstion of signal has dependency to environment, at not coordination The multipath characteristics putting its channel the most all differs, and presents the strongest particularity.Indoor Position Techniques Based on Location Fingerprint effectively utilizes Multipath effect, combines multipath characteristics with positional information, owing to the multi-path influence of channel has uniquely at same location point Property, can be using multidiameter configuration as fingerprint in data base.Tested point obtains the wireless signal that access point sends in same environment, will The wireless signal strength received mates with fingerprint in data base, finds out most like result and positions.

WiFi fingerprint location technology specifically location implement time in two stages: off-line training step and tuning on-line rank Section.

Off-line training step: first dispose wireless aps in localizing environment, determine sampling point position so that each sampled point The signal that wireless aps is launched can be received.Placing signal receiving device (mobile device) at each sampled point, record is received from These signal strength values and coordinate information are stored in fingerprint database, thus by the signal intensity (RSSI value) of each AP Uniquely identify this sampled point.After the sampling of all sampled points is terminated, build complete signal strength information and correspondence position The fingerprint database of relation, i.e. fingerprint map.

The tuning on-line stage: measure the signal strength information obtaining each AP at tested point in real time, and by itself and location fingerprint Information in data base is mated, and measured data and pre-stored data is carried out the matching analysis, thus estimates the position of terminal to be measured Put.

But, the positioning precision of traditional WiFi fingerprinting localization algorithm is the highest, location efficiency is relatively low.

Summary of the invention

The technical problem to be solved is to provide a kind of method improving WiFi fingerprint location precision and efficiency, energy Enough improve positioning precision and location efficiency.

The technical solution adopted for the present invention to solve the technical problems is: provide one improve WiFi fingerprint location precision with The method of efficiency, in off-line training step, builds the fingerprint database for tuning on-line, and to fingerprint database uses K- Means clustering algorithm is classified, and wherein, the signal strength values of the sampled point in fingerprint database processes through data smoothing； In the tuning on-line stage, using K the fingerprint that k nearest neighbor algorithm is sought closest with actual measurement fingerprint, the average correspondence of K fingerprint is adopted The position of sampling point is the estimation position of tested point.

Described structure specifically includes following steps for the fingerprint database of tuning on-line:

Selected certain environment indoor, as region, location, in this region, location, is disposed n WAP and chooses L Individual sampled point, records the position coordinates of each sampled point；

At each sampled point, the mobile terminal with WiFi signal detection function is utilized to carry out signal strength detection, repeatedly Gather the RSSI value of each WAP, then the data collected are smoothed, obtain this sampled point mean value smoothing After fingerprint；Travel through L sampled point, obtain L fingerprint, be stored in fingerprint database.

Described specifically include following steps to fingerprint database using K-means clustering algorithm carry out classification:

Fingerprint database is carried out K-means cluster, and using Euclidean distance as the interpretational criteria of similarity, distance is less Fingerprint is gathered in a subclass, apart from bigger fingerprint away from each other；

Previous step is performed a plurality of times, until cluster terminates, fingerprint database becomes the sample fingerprint sky with K subclass Between.

Described K-means clustering algorithm particularly as follows:

Input L fingerprint and cluster number K, wherein, K≤L；Arbitrarily select K fingerprint as initial from L fingerprint Cluster centre；

For remaining fingerprint, calculate each fingerprint distance to each cluster centre, after finding minimum range, by fingerprint Assign in the cluster of correspondence, obtain new cluster result, complete fingerprint distribution；

Calculating new cluster centre, and compare with last cluster centre, if both are identical, cluster terminates, Otherwise update cluster centre and return previous step and perform cluster.

The tuning on-line stage specifically includes following steps:

Actual measurement fingerprint is mated with the fingerprint database after training, calculates actual measurement fingerprint and each cluster centre Distance, and find out the cluster corresponding to minimum range；

Calculate the distance of actual measurement fingerprint and each fingerprint in the cluster corresponding to minimum range；

Arrange according to order from small to large according to the distance value obtained, retain K minimum distance, and by individual for this K away from Electing reference fingerprint as from corresponding fingerprint, the sample point coordinate of its correspondence is as reference coordinate；

Calculate the average estimation position as actual measurement fingerprint of this K reference coordinate.

Beneficial effect

Owing to have employed above-mentioned technical scheme, the present invention compared with prior art, has the following advantages that and actively imitates Really: the present invention uses mean value smoothing method to reduce the undulatory property of fingerprint sequence in off-line phase, and uses K-means clustering algorithm WiFi fingerprint is carried out classification process, in the tuning on-line stage, it is proposed that based on K-means cluster k nearest neighbor algorithm thus carry High location efficiency.

Accompanying drawing explanation

Fig. 1 is fingerprint database training flow chart in the present invention；

Fig. 2 is that in the present invention, K-means clusters flow chart；

Fig. 3 is k nearest neighbor algorithm flow chart based on K-means cluster in the present invention.

Detailed description of the invention

Below in conjunction with specific embodiment, the present invention is expanded on further.Should be understood that these embodiments are merely to illustrate the present invention Rather than restriction the scope of the present invention.In addition, it is to be understood that after having read the content that the present invention lectures, people in the art The present invention can be made various changes or modifications by member, and these equivalent form of values fall within the application appended claims equally and limited Scope.

Embodiments of the present invention relate to a kind of method improving WiFi fingerprint location precision and efficiency, on off-line training rank Section, builds the fingerprint database for tuning on-line, and to fingerprint database using K-means clustering algorithm classify, Wherein, the signal strength values of the sampled point in fingerprint database processes through data smoothing；In the tuning on-line stage, use k nearest neighbor K the fingerprint that algorithm is sought closest with actual measurement fingerprint, the position of the average correspondence sampled point of K fingerprint is tested point Estimate position.

Off-line training step

In off-line training step, in localizing environment, first dispose WAP (AP), determine sampling point position so that Each sampled point can receive the signal that wireless aps is launched.Place signal receiving device at each sampled point afterwards (to move and set Standby), record is received from the signal intensity (RSSI value) of each AP, finally these signal strength values and coordinate information is stored in finger In stricture of vagina data base, the most uniquely identifying this sampled point, the data of all sampled points are all stored in data base, form fingerprint number According to storehouse, for tuning on-line.

Ideally, the signal intensity RSSI value received can be successively decreased as regularity along with the increase of propagation distance, but Being in actual applications, wireless signal is effected by environmental factors in communication process, as the multipath of indoor signal, reflection, Wall and the absorption etc. of door, cause signal to produce inconsistent attenuation relation so that arbitrary sampled point collect every The RSSI value of individual wireless aps is not unique, there is larger fluctuation, and the precision of off-line training step fingerprint database is affected very by this Greatly, it is therefore desirable to take some effective and feasible measures farthest to reduce the fluctuation of RSSI value, to reduce finger print data The error in data in storehouse, improves positioning precision during tuning on-line.

For arbitrary sampled point, when recording the RSSI value of each wireless aps it can be seen that the RSSI of any one wireless aps Value is not the most unique, in the RSSI value the most difference that the different time periods receives, the most widely different.Therefore, Can not be only with certain rssi measurement value once as standard, the finger print data as a certain sampled point is stored in data base, so makes The position error become can be very big, and the method that now should use repetitive measurement, in arbitrary sample point, multi collect is each wireless Then the data collected are smoothed by the RSSI value of AP, thus reduce the fluctuation of RSSI value, to improve positioning precision.Letter Number smooth method has a lot, such as averaging method, median method, mode method etc..Present embodiment uses mean value smoothing method number According to smoothing processing.

Mean value smoothing method presets a standard deviation threshold method X_D, in arbitrary sample point, each nothing that it is collected Multiple RSSI value of line AP calculate its standard deviation S_D, standard deviation S_DThe biggest, it was demonstrated that the fluctuation of RSSI value is the most obvious, i.e. by environment Disturb the biggest, if data now being taken average be stored in fingerprint database as the RSSI value of this certain wireless aps of sample point, Error can be the biggest.First data are divided into two parts by mean value smoothing method, as shown in formula (1) and (2), wherein, and S₁Represent less than institute There is data mean value S_avThe average of part data, S₂Represent more than all data mean value S_avThe average of part data；Its Secondary, weigh S with a₁And S₂Proportion shared in the signal intensity RSSI received, as shown in formula (4), at S_D>X_DTime, S₁Account for The data of the relatively strong part of larger specific gravity, i.e. signal account for larger specific gravity in the data gathered, otherwise then account for less proportion；Finally, The RSSI value utilizing formula (3) can calculate arbitrary certain wireless aps of sample point is stored in fingerprint database.This value is by S₁And S₂Press Calculate according to different specific weight, at signal relatively strength S₁Account for larger specific gravity, at signal more weak place S₂Account for larger specific gravity, therefore the method Can effectively reduce the fluctuation of RSSI value, improve the data precision of off-line training step fingerprint database.

$S_1=\frac{1}{n1}\underset{i=1}{\overset{n1}{\&Sigma;}}{S}_i,S_i<S_{av}---\left(1\right)$

$S_2=\frac{1}{n2}\underset{i=1}{\overset{n2}{\&Sigma;}}{S}_i,S_i\&GreaterEqual;S_{av}---\left(2\right)$

RSSI value after mean value smoothing method processes is:

RSSI=(1-a) * S₁+a*S₂ (3)

Wherein:

$\begin{array}{c}a=\left\{\begin{array}{cc}0.5*(1-\frac{S_D-X_D}{S_D}),& S_D\&GreaterEqual;X_D\\ 0.5*(1+\frac{X_D-S_D}{X_D}),& S_D<X_D\end{array}\right.\\ 0<a<1\end{array}---\left(4\right)$

K-means cluster alternatively referred to as K-mean cluster.K-means clustering method is that a kind of use is extensive, it is many to be applicable to Planting the clustering algorithm of data type, algorithm is simple, it is achieved quickly.K-means cluster is typical clustering algorithm based on distance, Using distance as the tolerance of similarity, its algorithm basic thought be according to existing sample between similarity be divided into K Individual subclass, by sample collection bigger for similarity together, the less sample of similarity is away from each other.K-means clustering algorithm Can efficiently classify so that whole WiFi fingerprint database is divided into different groups, thus reduces tuning on-line stage position Fingerprint search scope, improves location efficiency.

The first stage of WiFi fingerprint algorithm is to set up WiFi fingerprint database under off-line state.Under WiFi environment, Assume in certain specific room area, n the i.e. RSSI value of WiFi signal intensity can be detected, select in this region, location Determining L sampled point, it is known that use two-dimensional space coordinate, (x y) represents the positional information of sampled point.Can adopt at each sampled point Collection arrives this n RSSI value (rssi¹,rssi²,…rssiⁿ), using this array as the fingerprint of this sampled point, each fingerprint with The position one_to_one corresponding of its sampled point, becomes the relation mapped one by one.So this fingerprint database can be expressed as formula (5) institute Show.

$LR=\left[\begin{array}{cccccc}{x}_1& y_1& {\mathrm{rssi}}_1^1& {\mathrm{rssi}}_1^2& \mathrm{...}& {\mathrm{rssi}}_1^n\\ x_2& y_2& {\mathrm{rss}}_2^1& {\mathrm{rssi}}_2^2& \mathrm{...}& {\mathrm{rssi}}_2^n\\ .& .& .& .& & .\\ .& .& .& .& \mathrm{...}& .\\ .& .& .& .& & .\\ x_L& y_L& {\mathrm{rssi}}_L^1& {\mathrm{rssi}}_L^2& \mathrm{...}& {\mathrm{rssi}}_L^n\end{array}\right]---\left(5\right)$

Wherein, LR comprises positional information and RSSI sequence, then finger print information just can separately shown for location sets L and Fingerprint set R, as shown in formula (6) and formula (7).

$L=\left[\begin{array}{cc}{x}_1& y_1\\ x_2& y_2\\ .& .\\ .& .\\ .& .\\ x_L& y_L\end{array}\right]---\left(6\right)$

$R=\left[\begin{array}{cccc}{\mathrm{rssi}}_1^1& {\mathrm{rssi}}_1^2& \mathrm{...}& {\mathrm{rssi}}_1^n\\ {\mathrm{rssi}}_2^1& {\mathrm{rssi}}_2^2& \mathrm{...}& {\mathrm{rssi}}_2^n\\ .& .& & .\\ .& .& \mathrm{...}& .\\ .& .& & .\\ {\mathrm{rssi}}_L^1& {\mathrm{rssi}}_L^2& \mathrm{...}& {\mathrm{rssi}}_L^n\end{array}\right]---\left(7\right)$

When the WiFi fingerprint of off-line training step training fingerprint database, it is necessary first to finger print data is carried out pre-place Reason, employing mean value smoothing method the most set forth above smooths fingerprint set R, obtainsUse K-means clustering algorithm to finger afterwards Stricture of vagina setCarry out classification process.The flow process of training fingerprint database is as shown in Figure 1.

Concrete training step is following four step:

Step1. selected certain environment indoor is as region, location, in this region, location, disposes n wireless aps and chooses L sampled point, records the position coordinates of each sampled point.

Step2. at each sampled point, the mobile terminal with WiFi signal detection function is utilized to carry out signal intensity inspection Survey, the RSSI value of each wireless aps of multi collect, then the data collected are smoothed, obtain this sampled point average and put down Fingerprint after cunning1≤i≤L；Travel through L sampled point, obtain L fingerprint, be stored in fingerprint database.

Step3. fingerprint database is carried out K-means cluster, using Euclidean distance as the interpretational criteria of similarity, distance Less fingerprint is gathered in a subclass, apart from bigger fingerprint away from each other.

Step4. Step3 is performed a plurality of times, until cluster terminates, fingerprint database becomes the sample fingerprint with K subclass Space.The K-means clustering method being wherein previously mentioned in Step3, its execution process is divided into following five steps:

1. L fingerprint of inputWith cluster number K, (K≤L)；From L fingerprint Arbitrarily select K fingerprint as initial cluster centre C=(C₁,C₂,…C_K)。

2., for remaining (L-K) individual fingerprint, calculate each fingerprint distance Dis tan ce=to each cluster centre {d_ij| i=1,2 ..., (L-K)；J=1,2 ..., K}, wherein d_ijRepresent that i-th fingerprint, to the distance of jth cluster centre, is looked for To min (Dis tan ce), i-th fingerprint is assigned in jth cluster, obtain new cluster result.

3. repeat the 2nd step, remaining fingerprint is assigned, form K cluster G₁,G₂,…,G_j,…G_K, each Class G_jAll comprising its cluster centre and belong to such fingerprint member, the total number of fingerprint is n_j。

4. according to formulaCalculate new cluster centre, wherein rssi_iRepresent G_jThe i-th of apoplexy due to endogenous wind RSSI value.Calculate each Lei Lei center, obtain new cluster centre

If 5. C^*=C, the cluster centre of the most adjacent twice is identical, i.e. classification tends towards stability, and cluster terminates, current G₁, G₂,…,G_j,…G_KRepresent the cluster ultimately formed.Otherwise make C=C^*, i.e. update class center, return second step and continue executing with Cluster process.The flow process of K-means clustering algorithm is as shown in Figure 2.

The tuning on-line stage

K-nearest neighbor algorithm (KNNSS) is the innovatory algorithm of nearest neighbor algorithm (NNSS).Nearest neighbor algorithm is easily achieved, algorithm Simply, the actual measurement fingerprint recorded by on-line stage mates with fingerprint database, the fingerprint seeking closest, and this fingerprint correspondence is adopted The position of sampling point is the estimation position of tested point.The reference fingerprint that nearest neighbor algorithm selects is the most single, and positioning result is unstable Fixed, it is easily generated bigger error.

For the problem that reference fingerprint is single, present embodiment uses k nearest neighbor algorithm (K-Nearest Neighbor in Signal Space, KNNSS).In KNNSS algorithm, it not to choose sampled point the estimating as tested point that single fingerprint is corresponding Meter position, but select K the fingerprint closest with actual measurement fingerprint, calculating the average of this K sampled point, this average is The estimation position of tested point.Although but KNNSS reduces position error on the basis of NNSS, but during its each tuning on-line Needing to compare actual measurement fingerprint with all fingerprints in fingerprint database to seek range difference, this position fixing process is the most oversize, is System operand is too big, and location efficiency is the lowest.Therefore, present embodiment is in order to, while improving positioning precision, it is fixed also to improve Position efficiency, devises a kind of k nearest neighbor algorithm based on K-means cluster.Its position fixing process is as it is shown on figure 3, be specially following five Individual step:

Step1. will actual measurement fingerprint r=(rssi₁,rssi₂,…,rssi_n) carry out with the fingerprint database after training Join, calculate the distance of r and each cluster centre, be designated as D=[D₁,D₂,…,D_K]。

Step2. find the class that minima min (D) in D is corresponding, be designated as G_MIN。

Step3. actual measurement fingerprint r=(rssi is calculated₁,rssi₂,…,rssi_n) and G_MINIn the distance of each fingerprintIt is designated asWhereinRepresent this apoplexy due to endogenous wind i-th fingerprint, n_gTable Show the number of fingerprint in such.

Step4. willArrange according to order from small to large, reject the most significantly greater distance value, protect Staying remaining K distance, and elect the fingerprint of this K distance correspondence as reference fingerprint, the sample point coordinate of its correspondence is as ginseng Examine coordinate.

Step5. the average estimation position as actual measurement fingerprint of this K reference coordinate, computational methods such as formula (8) institute are calculated Show.

$\begin{array}{cc}{x}_{estimate}=\frac{1}{K}\underset{i=1}{\overset{K}{\&Sigma;}}{x}_i& y_{estimate}=\frac{1}{K}\underset{i=1}{\overset{K}{\&Sigma;}}{y}_i\end{array}---\left(8\right)$

It is seen that, the present invention uses mean value smoothing method to reduce the undulatory property of fingerprint sequence in off-line phase, and uses K-means clustering algorithm carries out classification process to WiFi fingerprint, in the tuning on-line stage, it is proposed that K based on K-means cluster Nearest neighbor algorithm thus improve location efficiency.

Claims (5)

1. the method improving WiFi fingerprint location precision and efficiency, it is characterised in that in off-line training step, structure is used for The fingerprint database of tuning on-line, and to fingerprint database using K-means clustering algorithm classify, wherein, fingerprint number Process through data smoothing according to the signal strength values of the sampled point in storehouse；The tuning on-line stage, use k nearest neighbor algorithm seek with reality Surveying K the fingerprint that fingerprint is closest, the position of the average correspondence sampled point of K fingerprint is the estimation position of tested point.

Raising WiFi fingerprint location precision the most according to claim 1 and the method for efficiency, it is characterised in that described structure Fingerprint database for tuning on-line specifically includes following steps:

Certain environment selected indoor is as region, location, in this region, location, disposes n WAP and chooses that L is individual to be adopted Sampling point, records the position coordinates of each sampled point；

At each sampled point, the mobile terminal with WiFi signal detection function is utilized to carry out signal strength detection, multi collect Then the data collected are smoothed by the RSSI value of each WAP, after obtaining this sampled point mean value smoothing Fingerprint；Travel through L sampled point, obtain L fingerprint, be stored in fingerprint database.

Raising WiFi fingerprint location precision the most according to claim 1 and the method for efficiency, it is characterised in that described to finger Stricture of vagina data base uses K-means clustering algorithm carry out classification and specifically includes following steps:

Fingerprint database is carried out K-means cluster, using Euclidean distance as the interpretational criteria of similarity, the fingerprint that distance is less It is gathered in a subclass, apart from bigger fingerprint away from each other；

Previous step is performed a plurality of times, until cluster terminates, fingerprint database becomes the sample fingerprint space with K subclass.

Raising WiFi fingerprint location precision the most according to claim 1 and the method for efficiency, it is characterised in that described K- Means clustering algorithm particularly as follows:

Input L fingerprint and cluster number K, wherein, K≤L；Arbitrarily select K fingerprint as initial cluster from L fingerprint Center；

For remaining fingerprint, calculate each fingerprint distance to each cluster centre, after finding minimum range, fingerprint is assigned to In corresponding cluster, obtain new cluster result, complete fingerprint distribution；

Calculating new cluster centre, and compare with last cluster centre, if both are identical, cluster terminates, otherwise Update cluster centre and return previous step and perform cluster.

Raising WiFi fingerprint location precision the most according to claim 1 and the method for efficiency, it is characterised in that tuning on-line Stage specifically includes following steps:

Will actual measurement fingerprint with training after fingerprint database mate, calculate actual measurement fingerprint and each cluster centre away from From, and find out the cluster corresponding to minimum range；

Calculate the distance of actual measurement fingerprint and each fingerprint in the cluster corresponding to minimum range；

Arrange according to order from small to large according to the distance value obtained, retain K minimum distance, and by this K apart from right The fingerprint answered elects reference fingerprint as, and the sample point coordinate of its correspondence is as reference coordinate；

Calculate the average estimation position as actual measurement fingerprint of this K reference coordinate.