CN105960021A - Improved position fingerprint indoor positioning method - Google Patents

Abstract

The invention discloses an improved position fingerprint indoor positioning method. The method comprises an offline stage and an online stage. The method is characterized in that the offline stage comprises the steps of establishing a position fingerprint database by use of reference points and APs, and classifying data in the fingerprint database through a K-means clustering algorithm and obtaining the class cluster to which the to-be-tested point belongs; and the online stage comprises the steps of matching the to-be-tested point with the fingerprint of the class cluster by use of a Bayesian probability algorithm, thereby calculating the position coordinate of the to-be-tested point. According to the improved position fingerprint indoor positioning method, the K-means clustering algorithm is combined with the Bayesian probability algorithm; the data in the fingerprint database is classified by use of the K-means clustering algorithm, thereby obtaining the class cluster to which the to-be-tested point belongs; therefore, the target range is greatly reduced; the to-be-tested point is matched with the fingerprint of the class cluster by use of the Bayesian probability algorithm; and therefore, according to the position fingerprint positioning algorithm, the positioning precision is ensured, the complexity of the algorithm is reduced, the efficiency is improved, and the algorithm has practical value.

Description

A kind of location fingerprint indoor orientation method of improvement

Technical field

The present invention relates to wireless location technology field, in particular, particularly relate to the position of a kind of improvement Put fingerprint indoor orientation method.

Background technology

Along with the fast development of radio communication, wireless location technology increasingly obtains the concern of people, especially It is indoor positioning technologies, it is desirable in megastore, public place, underground parking, Longer velocity tunnel etc. know the position at self place at any time, and can quickly arrive at. The indoor positioning technologies that existing comparison is popular has following several, and the time of advent (TOA), the time of advent is poor (TDOA), angle (AOA), location fingerprint location (LFP) are arrived.Above location technology is respectively arranged with pluses and minuses, Wherein location fingerprint setting accuracy is the highest, but owing to its algorithm complex is high, the time of cost is longer, Higher to environmental requirement, these are all present stage problem demanding prompt solutions.

The most conventional fingerprint positioning method mainly has nearest neighbor algorithm, Bayesian probability method, BP neural Network technique etc..Their position fixing process is the most all divided into off-line phase and on-line stage.Collect during off-line Massive Sample data form data base, by the comparison one by one in data base of the data in site undetermined time online, Obtain final positioning result.But, to nearest neighbor method, its arithmetic speed is very fast, but accuracy is not Height, its innovatory algorithm k nearest neighbor and K weighting nearest neighbour method there is also the problem that k value cannot determine, pattra leaves Although this probabilistic method precision is high, but it is high to there is also algorithm complex, the problem such as cause positioning time oversize..

Summary of the invention

The present invention is for the shortcoming overcoming above-mentioned technical problem, it is provided that the location fingerprint room of a kind of improvement Inner position method.

The location fingerprint indoor orientation method of the improvement of the present invention, is divided into off-line phase and on-line stage, It is particular in that, off-line phase sets up location fingerprint data base first with reference point and AP, so By K-means clustering algorithm, data in fingerprint database are classified afterwards, and obtain class described in tested point Bunch；On-line stage utilizes Bayesian probability method to be mated by the fingerprint of tested point with described class bunch, with Calculate the position coordinates of tested point.

The location fingerprint indoor orientation method of the improvement of the present invention, off-line phase is divided into off-line build storehouse and gather Class processes, and is realized by following steps:

A). off-line builds storehouse, arranges n reference point, m AP in region, location, all reference points Location sets is: L={l₁,l₂,…l_n, wherein l_i={ x_i,y_iIt is the position coordinates of i-th reference point, I=1,2 ..., n；

Set up and the set L R={r of location fingerprint set one to one₁,r₂,...,r_n, wherein r_i={ rssi_i1,rssi_i2,...,rssi_im, i=1,2 ..., n；J=1,2 ..., m；rssi_ijRepresent what i-th reference point received Signal strength values from jth AP；Like this, set up containing signal strength values and position coordinates Location fingerprint data base { the rssi of reference point_i1,rssi_i2,...,rssi_im,x_i,y_i}；

B). clustering processing, cluster, according to data to be clustered centered by positioning in region K point Divide with cluster centre distance minimum principle, and update the value of cluster centre, until obtaining best Cluster result；After cluster terminates, form K Ge Leicu center；

On-line stage is realized by following steps:

C). obtain class bunch belonging to tested point, calculate the distance of tested point and K Ge Leicu center respectively, away from It is class bunch belonging to tested point from reckling, if the number of reference point is s in affiliated class bunch；

D). obtain the finger print information of tested point, calculate the signal intensity of tested point and all AP, to be measured The finger print information of point is A:A={rssi_a1,rssi_a2,...,rssi_am, wherein rssi_ajRepresent what tested point received Signal strength values from jth AP；

E). ask for posterior probability, if the posterior probability of reference point is P (l in class bunch belonging to tested point_i| A), I=1,2 ..., s, s are the number of reference point in described class bunch；According to Bayes theorem, the meter of posterior probability It is converted into:

$P\left(l_i\right|A)=\frac{P\left(A\right|l_i)P\left(l_i\right)}{P\left(A\right)}=\frac{P\left(A\right|l_i)P\left(l_i\right)}{\underset{i=1}{\overset{n}{\Σ}}P\left(A\right|l_i)P\left(l_i\right)}---\left(1\right)$

Wherein, P (A | l_i) represent at known location coordinate l_iThe conditional probability that location fingerprint is A at place, P(l_i) represent that tested point occurs in l_iThe probability of position, the probability that tested point occurs in region, location is obeyed all Even distribution, if P is (l_i)=C, C are constant；

From probabilistic knowledge:

$P\left(A\right|l_i)=P\left({\mathrm{rssi}}_{a1}\right|l_i)P\left({\mathrm{rssi}}_{a2}\right|l_i)...P\left({\mathrm{rssi}}_{am}\right|l_i)=\underset{j=1}{\overset{m}{\Π}}P\left({\mathrm{rssi}}_{aj}\right|l_i)---\left(2\right)$

P(rssi_aj|l_i) represent at known location coordinate l_iPlace's signal strength values is rssi_ajProbability, each position The signal intensity received at fingerprint obeys Gauss normal distribution；

Formula (2) is substituted in formula (1) P (l can be tried to achieve_i|A)；

F). calculate tested point position, the posterior probability P (l that will obtain_i| A) as the reference of fingerprint database The weight of point, substitutes into formula (3) and estimates the position of tested point:

$(\hat{x},\hat{y})=\underset{i=1}{\overset{s}{\Σ}}P\left(l_i\right|A)\×(x_i,y_i)---\left(3\right)$

Wherein, (x_i,y_i) it is the coordinate of reference point in class bunch belonging to tested point.

The location fingerprint indoor orientation method of the improvement of the present invention, the clustering processing described in step b) is led to Cross following steps to realize:

B-1). initialize clustering parameter, if the data set participating in clustering processing is D=(D₁,D₂,...,D_n), D₁、D₂、...、D_nThe location fingerprint data base of the most corresponding n reference point；If the number of cluster is K, Here arranging cluster number consistent with the number of AP, K class bunch is expressed as C_j, 0≤j≤k ＜ n；

B-2). choosing cluster centre, choosing the location fingerprint from the nearest reference point of each AP coordinate is Cluster centre, to avoid owing to randomness chooses the error caused；

B-3). reference point is classified, and calculates element D in data set D_iWith the distance of K cluster centre, i.e. d_j=| | D_i-X_j| |, wherein j=1,2 ..., k；If meeting min (d_j)=| | D_i-X_j| |, then judge D_i∈C_j；

B-4). calculate new cluster centre, after each cluster puts new reference point under, all pass through formula (4) cluster centre that calculating makes new advances:

$X_j^{\left(2\right)}=\frac{\underset{i=1}{\overset{N_j}{\Σ}}{D}_i}{N_j}---\left(4\right)$

Wherein, N_jRepresent the number belonging to jth class bunch；

B-5). iterative step b-3) to b-4), all reference points are classified one by one, until new poly- The change at class center is less than given threshold epsilon, i.e.The center of K class bunch of output, Calculate and stop.

The location fingerprint indoor orientation method of the improvement of the present invention, the off-line storehouse described in step a) was set up Cheng Zhong, due to change, the impact of walking about of crowd of external environment, can make the signal intensity detected send out Raw fluctuation, therefore when position fingerprint database is set up, the method using multiple repairing weld, by repeatedly Measure and average as final signal strength values.

The location fingerprint indoor orientation method of the improvement of the present invention, the calculating of tested point position in step f) During, when the quantity of reference point is more, according to signal intensity by treating reference to weak selected part by force Point is weighted.

The invention has the beneficial effects as follows: the location fingerprint indoor orientation method of the improvement of the present invention, pass through K-means and bayesian algorithm are combined, and utilize K-means clustering algorithm by fingerprint database Data are classified, and obtain class bunch belonging to tested point, greatly reduce target zone, then use Bayesian probability Tested point is mated by method with the fingerprint of this type of bunch, makes location fingerprint location algorithm on the one hand ensure Setting accuracy, on the other hand reduces algorithm complex, improves efficiency, have practical value.

Accompanying drawing explanation

Fig. 1 is the schematic diagram of the location fingerprint indoor orientation method of the improvement of the present invention.

Detailed description of the invention

The invention will be further described with embodiment below in conjunction with the accompanying drawings.

The present invention adopts the following technical scheme that the location fingerprint indoor positioning algorithms of a kind of improvement is divided into off-line Stage and on-line stage.

1, off-line phase.Off-line phase is divided into off-line to build storehouse and clustering processing.

1) off-line builds storehouse

The location sets gathering all reference points in region, location is: L={l₁,l₂,…l_n, wherein, n is location The number of region reference point, l_i={ x_i,y_i(i=1,2 ... n) be the position coordinates of i-th reference point.With L Location fingerprint collection is combined into one to one: R={r₁,r₂,…r_n, wherein, r_i={ rssi_i1,rssi_i2,…rssi_im(i=1,2 ... n；J=1,2 ... m), in formula, rssi_ijRepresent that i-th reference point connects The signal strength values from jth AP received, m represents the number of AP.Set up one like this Location fingerprint data base, record is signal strength values and corresponding position coordinates {rssi_i1,rssi_i2…rssi_im,x_i,y_i}.It is noted herein that, due to the change of external environment, crowd The factor such as walk about can make the signal intensity detected fluctuate, and therefore builds at position fingerprint database Immediately, the method using multiple repairing weld, averaged by repetitive measurement and carry out the record of data.

2) clustering processing

First, clustering processing is to be realized by K-means algorithm, and algorithm idea is: with K in space Cluster centered by Dian, divide with cluster centre distance minimum principle according to data to be clustered, And updating the value of new cluster centre, iteration is carried out until obtaining best cluster result then.

K-means clustering algorithm iterative step:

Input: data set D=(D₁,D₂…D_n), number k of class

Output: k class bunch C_j(0≤j≤k, k ＜ N)

Step1: initialize each clustering parameter.(1) arrange cluster number k, here arrange cluster number with The number of AP is consistent；(2) initial cluster center (X is set₁,X₂…X_k), here cluster centre not with Machine generates, according to the feature of the general distribution of wireless signal strength, from emission source more close to, corresponding signal intensity The strongest, therefore choose from the location fingerprint treating reference point that each AP coordinate is nearest be cluster centre, this Sample avoids owing to randomness chooses the error caused；(3) iteration stopping condition ε is set.

Step2: to data set D=(D₁,D₂…D_n) classify.

(1) D is calculated_iThe distance of (i=1,2 ... N) and K cluster centre, i.e. d_j=| | D_i-X_j| | (j=1,2 ... k), if meeting min (d_j)=| | D_i-X_j| | (1≤j≤k), then judge D_i∈C_j。

(2) new cluster centre is calculated,N_jRepresent the number belonging to jth class bunch.

Step3: iteration step2, classifies one by one by data set, until the change of new cluster centre is little In given threshold value, i.e.The center of k class bunch of output, stops calculating.

In actual applications, owing to each location fingerprint can receive the signal intensity from different AP, And cluster and may be only available for the signal that same AP launches.Therefore can choose at tested point with established standards The AP that the signal intensity that receives is the strongest is as the criterion.

2, on-line stage

On-line stage utilizes Bayesian probability method to be mated by the fingerprint of tested point with affiliated class bunch.Pattra leaves This probabilistic method is by calculating each posterior probability treating reference point, choose posterior probability maximum one Or several reference modes, thus calculate the physical coordinates of tested point.Compared with additive method, Probabilistic method can have the impact of bigger error in effective removal sampled signal, and denoising is with the obvious advantage.

Concretely comprise the following steps:

(1) class bunch at tested point place is obtained.Calculate respectively tested point and k cluster centre away from From, distance reckling is class bunch belonging to tested point.

(2) fingerprint matching.Making the finger print information recorded at tested point is A, A={rssi_a1,rssi_a2,…rssi_am}, Its posterior probability treating reference point in affiliated class bunch is P (l_i| A) (i=1,2 ... s), s are affiliated class bunch Treat the number of reference point.According to Bayes theorem, posterior probability can be converted into:Wherein, P (A | l_i) represent at known location coordinate l_iPlace The conditional probability that location fingerprint is A, P (l_i) represent occur in l at tested point_iThe probability of position.Typically It is uniformly distributed assuming that tested point is obeyed at the probability that region, location occurs, can be by P (l_i) it is seen as a constant. It addition, from probabilistic knowledge:

$P\left(A\right|l_i)=P\left({\mathrm{rssi}}_{a1}\right|l_i)P\left({\mathrm{rssi}}_{a2}\right|l_i)...P\left({\mathrm{rssi}}_{am}\right|l_i)=\underset{j=1}{\overset{m}{\Π}}P\left({\mathrm{rssi}}_{aj}\right|l_i)$

The signal intensity received at each location fingerprint is obeyed Gauss normal distribution, substitutes into formula and try to achieve P(l_i| A), as the weight treating reference point of fingerprint database, then estimate the position of tested point Put,

Finally giving tested point position is

Preferably, in practical operation, as when reference point quantity is more, can according to signal intensity by Treat that reference point is weighted to weak selected part by force.

The foregoing is only the preferred embodiments of the present invention, it is impossible to limit the practical range of the present invention with this, Therefore any improvement done according to the claims in the present invention is the most within the scope of the present invention.

Claims (5)

1. the location fingerprint indoor orientation method improved, is divided into off-line phase and on-line stage, its Being characterised by, off-line phase sets up location fingerprint data base first with reference point and AP, then passes through Data in fingerprint database are classified by K-means clustering algorithm, and obtain class bunch described in tested point；Online Stage utilizes Bayesian probability method to be mated by the fingerprint of tested point with described class bunch, treats to calculate The position coordinates of measuring point.

The location fingerprint indoor orientation method of improvement the most according to claim 1, it is characterised in that: Off-line phase is divided into off-line to build storehouse and clustering processing, is realized by following steps:

A). off-line builds storehouse, arranges n reference point, m AP in region, location, all reference points Location sets is: L={l₁,l₂,…l_n, wherein l_i={ x_i,y_iIt is the position coordinates of i-th reference point, I=1,2 ..., n；

Set up and the set L R={r of location fingerprint set one to one₁,r₂,...,r_n, wherein r_i={ rssi_i1,rssi_i2,...,rssi_im, i=1,2 ..., n；J=1,2 ..., m；rssi_ijRepresent what i-th reference point received Signal strength values from jth AP；Like this, set up containing signal strength values and position coordinates Location fingerprint data base { the rssi of reference point_i1,rssi_i2,...,rssi_im,x_i,y_i}；

B). clustering processing, cluster, according to data to be clustered centered by positioning in region K point Divide with cluster centre distance minimum principle, and update the value of cluster centre, until obtaining best Cluster result；After cluster terminates, form K Ge Leicu center；

On-line stage is realized by following steps:

C). obtain class bunch belonging to tested point, calculate the distance of tested point and K Ge Leicu center respectively, away from It is class bunch belonging to tested point from reckling, if the number of reference point is s in affiliated class bunch；

D). obtain the finger print information of tested point, calculate the signal intensity of tested point and all AP, to be measured The finger print information of point is A:A={rssi_a1,rssi_a2,...,rssi_am, wherein rssi_ajRepresent what tested point received Signal strength values from jth AP；

E). ask for posterior probability, if the posterior probability of reference point is P (l in class bunch belonging to tested point_i| A), I=1,2 ..., s, s are the number of reference point in described class bunch；According to Bayes theorem, the meter of posterior probability It is converted into:

$P\left(l_i|A\right)=\frac{P\left(A|l_i\right)P\left(l_i\right)}{P\left(A\right)}=\frac{P\left(A|l_i\right)P\left(l_i\right)}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}P\left(A|l_i\right)P\left(l_i\right)}---\left(1\right)$

Wherein, P (A | l_i) represent at known location coordinate l_iThe conditional probability that location fingerprint is A at place, P(l_i) represent that tested point occurs in l_iThe probability of position, the probability that tested point occurs in region, location is obeyed all Even distribution, if P is (l_i)=C, C are constant；

From probabilistic knowledge:

$P\left(A\right|l_i)=P\left({\mathrm{rssi}}_{a1}\right|l_i)P\left({\mathrm{rssi}}_{a2}\right|l_i)...P\left({\mathrm{rssi}}_{am}\right|l_i)=\underset{j=1}{\overset{m}{\Π}}P\left({\mathrm{rssi}}_{aj}\right|l_i)---\left(2\right)$

P(rssi_aj|l_i) represent at known location coordinate l_iPlace's signal strength values is rssi_ajProbability, each position The signal intensity received at fingerprint obeys Gauss normal distribution；

Formula (2) is substituted in formula (1) P (l can be tried to achieve_i|A)；

F). calculate tested point position, the posterior probability P (l that will obtain_i| A) as the reference of fingerprint database The weight of point, substitutes into formula (3) and estimates the position of tested point:

$(\hat{x},\hat{y})=\underset{i=1}{\overset{s}{\Σ}}P\left(l_i\right|A)\×(x_i,y_i)---\left(3\right)$

Wherein, (x_i,y_i) it is the coordinate of reference point in class bunch belonging to tested point.

The location fingerprint indoor orientation method of improvement the most according to claim 2, it is characterised in that: Clustering processing described in step b) is realized by following steps:

B-1). initialize clustering parameter, if the data set participating in clustering processing is D=(D₁,D₂,...,D_n), D₁、D₂、...、D_nThe location fingerprint data base of the most corresponding n reference point；If the number of cluster is K, Here arranging cluster number consistent with the number of AP, K class bunch is expressed as C_j, 0≤j≤k ＜ n；

B-2). choosing cluster centre, choosing the location fingerprint from the nearest reference point of each AP coordinate is Cluster centre, to avoid owing to randomness chooses the error caused；

B-3). reference point is classified, and calculates element D in data set D_iWith the distance of K cluster centre, i.e. d_j=| | D_i-X_j| |, wherein j=1,2 ..., k；If meeting min (d_j)=| | D_i-X_j| |, then judge D_i∈C_j；

B-4). calculate new cluster centre, after each cluster puts new reference point under, all pass through formula (4) cluster centre that calculating makes new advances:

$X_j^{\left(2\right)}=\frac{\underset{i=1}{\overset{N_j}{\Σ}}{D}_i}{N_j}---\left(4\right)$

Wherein, N_jRepresent the number belonging to jth class bunch；

B-5). iterative step b-3) to b-4), all reference points are classified one by one, until new poly- The change at class center is less than given threshold epsilon, i.e.The center of K class bunch of output, Calculate and stop.

4. according to the location fingerprint indoor orientation method of the improvement described in Claims 2 or 3, its feature It is: during off-line storehouse described in step a) is set up, due to the change of external environment, the walking of crowd Dynamic impact, can make the signal intensity detected fluctuate, therefore when position fingerprint database is set up, The method using multiple repairing weld, is averaged as final signal strength values by repetitive measurement.

5. according to the location fingerprint indoor orientation method of the improvement described in Claims 2 or 3, its feature It is: in step f) during the calculating of tested point position, when the quantity of reference point is more, according to Signal intensity is by treating that reference point is weighted to weak selected part by force.