CN104185275B  A kind of indoor orientation method based on WLAN  Google Patents
A kind of indoor orientation method based on WLAN Download PDFInfo
 Publication number
 CN104185275B CN104185275B CN201410458932.0A CN201410458932A CN104185275B CN 104185275 B CN104185275 B CN 104185275B CN 201410458932 A CN201410458932 A CN 201410458932A CN 104185275 B CN104185275 B CN 104185275B
 Authority
 CN
 China
 Prior art keywords
 vector
 rssi
 region
 characteristic
 data
 Prior art date
Links
 238000000034 method Methods 0.000 claims abstract description 6
 230000000875 corresponding Effects 0.000 claims description 22
 230000004807 localization Effects 0.000 claims description 19
 239000000203 mixtures Substances 0.000 claims description 19
 239000000284 extracts Substances 0.000 claims description 10
 238000007621 cluster analysis Methods 0.000 claims description 7
 238000000605 extraction Methods 0.000 claims description 7
 230000001174 ascending Effects 0.000 claims description 6
 241001269238 Data Species 0.000 claims description 4
 238000004458 analytical methods Methods 0.000 claims 1
 230000001808 coupling Effects 0.000 claims 1
 238000010168 coupling process Methods 0.000 claims 1
 238000005859 coupling reactions Methods 0.000 claims 1
 230000000694 effects Effects 0.000 abstract description 10
 238000005070 sampling Methods 0.000 abstract description 9
 230000005540 biological transmission Effects 0.000 abstract description 5
 238000004891 communication Methods 0.000 abstract description 4
 230000002159 abnormal effect Effects 0.000 abstract description 3
 238000004364 calculation method Methods 0.000 description 5
 238000005516 engineering process Methods 0.000 description 5
 238000004422 calculation algorithm Methods 0.000 description 4
 238000005192 partition Methods 0.000 description 3
 230000001419 dependent Effects 0.000 description 2
 235000013399 edible fruits Nutrition 0.000 description 2
 238000004220 aggregation Methods 0.000 description 1
 230000002776 aggregation Effects 0.000 description 1
 238000001514 detection method Methods 0.000 description 1
 238000005315 distribution function Methods 0.000 description 1
 238000009114 investigational therapy Methods 0.000 description 1
 238000011002 quantification Methods 0.000 description 1
 238000004642 transportation engineering Methods 0.000 description 1
 238000004450 types of analysis Methods 0.000 description 1
Abstract
The invention discloses a kind of indoor orientation method based on WLAN, belong to indoor wireless communication and network technique field.Method includes：The RSSI data predictions for each AP that sampled point is collected, a peacekeeping bivector is therefrom extracted respectively as characteristic vector；Feature vector clusters are analyzed area to be targeted is divided into multiple positioning subregions；For every group of characteristic vector, respective disaggregated model is respectively trained out；Poll highest subregion set is chosen from all subregions with reference to " ballot " mechanism based on disaggregated model；Positioned using twowheeled and reduce subregion range of convergence, improve positioning precision.Fully digging utilization of the invention RSSI spatial distribution characteristic, solves the problems such as largescale indoor positioning search package space is excessive, and computation complexity is high；New location model is established, is solved in existing WLAN indoor orientation methods, the problems such as can not effectively learning and adapt to RSSI signals nonlinear caused by the reason such as non line of sight transmission effects, RSSI attenuation laws be abnormal, nongaussian statistical property.
Description
Technical field
The present invention relates to a kind of localization method in WLAN indoor positionings field, belongs to indoor wireless communication and network technology neck
Domain.
Background technology
In recent years, as the continuous improvement of people's living standard, people are also growing day by day to the demand of locationbased service,
Such as in personal scheduling, asset management, emergency relief, security monitoring, sacurity dispatching, intelligent transportation, digital map navigation, travel guide
All manysided widespread demands to positioning；It is that such as emergency relief, disaster relief emergency command are dispatched particularly in reply emergency
Under special applications scene, location information is more particularly important.
With the further investigation of general fit calculation machine and Distributed Communication Technology, indoor wireless communication and network technology are rapidly sent out
Exhibition, has derived and has been based on WLAN (Wireless Local Area Networks, WLAN), Bluetooth (bluetooth),
The indoor positioning modes such as WSN (wireless sensor network, wireless sensor network), and based on fingerprint and probability
The indoor orientation method of method.
Based on WLAN, Bluetooth, WSN etc. location technology, by carrying out mesh generation, and portion indoors indoors
Affix one's name to substantial amounts of AP (Access Point, access points), the RSSI for multiple AP that terminal detection receives in each grid
(Received Signal Strength Indication, received signal strength indicator), due to diverse location receive it is each
The signal intensity that individual signal node is sent is different, using the RSSI of each node received in each grid as the network
Characteristic quantity to complete to position.
Indoor positioning based on fingerprint, by gathering the RSSI of different AP in room area, and by corresponding wireless access
The address and coordinate of point are stored in database, and terminal user measures surrounding wireless signal strength, by it and are stored in advance in number
Matching positioning is carried out in right amount according to the RSSI in storehouse, so as to obtain being positioned the coordinate information of terminal user.
Probabilistic method draws the RSSI signal probabilities distribution in each reference point using the existing training sample in reference point.
Probability Distribution Fitting is typically carried out using Gaussian function, draws the average and bandwidth of the gaussian probability distribution of each reference point.Generally
Rate method takes full advantage of the statistical nature of signal distributions, and positioning precision typically will height compared with weighted nearest neighbor method.
However, they equally exist the problem of respective.Indoor orientation method based on fingerprint, in actual applications, for
Largescale indoor positioning, Existential Space matching hunting zone is larger, and computation complexity is high, and memory space requirement is larger not
, RSSI signals in actual applications be present in the reference point that some is fixed in foot, and the indoor orientation method based on probabilistic method
Nongaussian, nonlinear, multimodal characteristic is presented in probability distribution so that the probabilitydistribution function and actual probability distribution fitted
Differ larger, so as to cause positioning when larger matching error.
The content of the invention
The technical problem to be solved in the present invention is：What deficiency of prior art overcome, there is provided a kind of room based on WLAN
Interior localization method, matching hunting zone can be reduced, the forecast model that and can is tallied with the actual situation, is reduced to a certain extent
Computation complexity and time complexity.
The technical problem to be solved in the present invention is：Matching hunting zone is reduced, establishes a kind of prediction to tally with the actual situation
Model, there is provided a kind of indoor orientation method based on WLAN, comprise the following steps realization：
Step 1：The RSSI data predictions for each AP that sampled point is collected, therefrom extract a peacekeeping two dimension to
Amount is respectively as characteristic vector.
Carrying out necessary pretreatment to the RSSI data scanned includes：The data that RSSI is less than 100dB are deleted, are deleted
Nonlocating AP data.The data for deleting nonlocating AP refer to, delete the RSSI for the AP for being unsuitable for positioning.It is unsuitable for positioning
AP feature is intensity too low (RSSI is less than 95dB) or less stable (variance is more than 20), can increase meter using these AP
Complexity is calculated, reduces positioning precision, therefore excluded.
Extracted using Different Extraction Method from initial data it is a variety of can the accurate quantification RSSI regularities of distribution characteristic vector.
Comprise the following steps：
(1) by all AP scanned according to MAC Address ascending sort, all original numbers that will be scanned when gathering offline
Corresponding sampling point number on position mark is gathered according to according to it；
(2) respective characteristic vector can be extracted according to following two methods：
A. by the AP combination of two after sequence, i.e., AP is divided into according to MAC AddressGroup, every group of AP are expressed as (AP_{i},
AP_{j}) (wherein, 0<i<J≤m, m represent all AP number), corresponding A P is extracted from the initial data for marked sampled point
The RSSI of combination is vectorial and corresponding sampled point；
B. each AP is separately as one group, will all offline gathered datas be divided into m groups according to AP MAC Address, every group
AP is expressed as AP_{i}(wherein, 0<I≤m, m represent all AP number), extracted from the initial data for marked sampled point pair
Answer AP RSSI onedimensional vectors and corresponding sampled point.
Step 2：Feature vector clusters are analyzed, area to be targeted is divided into multiple positioning subregions, per subregions
Reflect a kind of RSSI distribution characteristics.
Using the characteristic vector constructed in step 1 as input, measuring similarity function is used as using the distance between characteristic vector
Carry out cluster analysis.Optionally, cluster analysis is using the Xmeans algorithms that can find clusters number automatically.Xmeans clusters are calculated
Method improves Kmeans algorithms, and cluster numbers K need not be preassigned in the initial computing of algorithm, need to only specify K value model
Enclose [K1, K2] (K1<K2), algorithm will find an optimal cluster numbers K in specified scope, realize clustering.X
For means algorithms using bayesian information criterion as guidance, the cluster centre for constantly traveling through inhomogeneity cluster is to represent different signal spies
Sign, signal characteristic reflect the clustering phenomena of the signal distributions in a certain region.
Step 3：For every group of characteristic vector combination cluster result, respective disaggregated model is respectively trained out；It is based on
Disaggregated model chooses poll highest subregion set with reference to " ballot " mechanism from all subregions.Including：
Offline phase, the characteristic vector constructed for the two kinds of structures method proposed in step 2, it is respectively trained out every
SVMs (Support Vector Machine, SVM) classification mould corresponding to every feature vectors of kind building method
Type.SVM is built upon VC dimensions (VC dimension) theory and structural risk minimization (structural of statistical learning
Risk minimization) on principle basis.SVM by nicety of grading (to the classification correctness of specific sample) and point
Class ability (carrying out errorfree misclassification to arbitrary sample) trades off, to make grader obtain best Generalization Ability.Feature
It is worth the input as SVM classifier, is the abstractdesription to data, therefore the selection of characteristic value is extremely important, can be accurate
Reflect that data characteristicses to be sorted will directly affect final classifying quality.
Online stage, from real time data extraction characteristic of division vector, read the corresponding svm classifier that offline phase trains
Model, according to the supporting vector polynomial expansion entry value, the probability that vector to be sorted corresponds to different zones is calculated, with reference to
" ballot " mechanism chooses poll highest set of regions R from all areas.
The concrete operations in tuning online stage include：
(1) the svm classifier model trained is read, calculates supporting vector polynomial expansion entry value；
(2) RSSI currently collected, extraction characteristic of division vector are read；
(3) by Polynomial kernel function by characteristic of division DUAL PROBLEMS OF VECTOR MAPPING to higher dimensional space, it is and more according to the supporting vector
Item formula expansion entry value calculates the probability that vector to be sorted corresponds to different zones；
(4) for each AP groups (AP_{i},AP_{j}), judge whether every subregions for marking off are eligible, if there is more
Subregions are eligible, then SVM models think that current device is likely to be in the union of these subregions；
The qualified region refers to, as AP groups (AP_{i},AP_{j}) in the prediction probability of a certain subregion it is more than a certain threshold
Value ε (0<ε<1) when, it is qualified to be considered as the region；
(5) poll highest set of regions R is chosen from all areas with reference to " ballot " mechanism, specific steps include：
If AP groups (AP_{i},AP_{j}) sample data be identified as by SVM predictions in a certain region, then the region ticket
Number plus 1.From geometrically showing as choosing localization region of the most region of capped number as coarseness, the ticket in each region
Number should arrive 0Between.
Step 4：Positioned using twowheeled and reduce set of regions scope, improve positioning precision.Specifically include：
(1) the svm classifier model trained is read, calculates supporting vector polynomial expansion entry value；
(2) RSSI currently collected, extraction characteristic of division vector are read, and characteristic of division is standardized；
(3) by Polynomial kernel function by characteristic of division DUAL PROBLEMS OF VECTOR MAPPING to higher dimensional space, it is and more according to the supporting vector
Item formula expansion entry value calculates the probability that vector to be sorted corresponds to different zones, and the coarseness therefrom obtained in selecting step three is determined
The probability of regional in the region R of position；
(4) for each AP_{i}, judge whether every subregions for marking off are eligible, and the subregion is step 3
In the coarseness localization region R subset obtained, eligible if there is more subregions, then SVM models are thought currently to set
It is standby to be likely to be in the union of these subregions；
The qualified region refers to, works as AP_{i}It is more than a certain threshold epsilon (0 in some region of prediction probability<ε<1)
When, it is qualified to be considered as the region；
(5) poll highest set of regions R ' is chosen from R with reference to " ballot " mechanism, specific steps include：If AP_{i}Sample
Notebook data is identified as in a certain region by SVM predictions, then the region poll adds 1.It is coated to from geometrically showing as choosing
As finegrained localization region, the poll in each region should be 0 between m in lid number most region.
The beneficial effect of technical scheme provided by the invention is：Of the invention fully digging utilization RSSI spatial distribution is special
Sign, reduces the localization region deviation caused by region division is improper；New location model is established, it is fixed in existing WLAN rooms to solve
Position method in, can not effectively learn and adapt to RSSI signals because non line of sight transmission effects, multipath transmisstion effect and RSSI decline
Subtract nonlinear, the nongaussian statistical properties caused by reason such as rule exception, and largescale indoor positioning, search for package space
It is excessive, the problems such as computation complexity is high.
Brief description of the drawings
Technical scheme in order to illustrate the embodiments of the present invention more clearly, make required in being described below to embodiment
Accompanying drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for
For those of ordinary skill in the art, on the premise of not paying creative work, other can also be obtained according to these accompanying drawings
Accompanying drawing.
Fig. 1 is the inventive method implementation process figure；
Fig. 2 is the cluster flow chart of the inventive method；
Fig. 3 is another cluster flow chart of the inventive method；
Fig. 4 is the training flow chart of the inventive method；
Fig. 5 is a kind of coarseness positioning flow figure of the inventive method；
Fig. 6 is a kind of fine granularity positioning flow figure of the inventive method.
Embodiment
Specific embodiments of the present invention are described further with reference to flow chart and specific embodiment.
Fig. 2 is the cluster flow chart of the inventive method, and the flow belongs to a part for offline phase.It can specifically include such as
Lower step：
201st, smart mobile phone high frequency sweep periphery AP signals, the data format scanned such as table 1 are used in each school punctuate
It is shown.It should be noted that the number of data of each school punctuate collection is not fixed, it is different because of acquisition time length.It is if current
Position fails to collect corresponding AP RSSI, is filled up with 100dB.
The scan data form of table 1
School punctuate numbering  AP1  AP2  AP3  AP4  AP5  AP6  AP7 
1  85  97  63  100  100  90  72 
1  83  92  65  100  98  85  69 
2  70  73  95  82  63  100  100 
……  ……  ……  ……  ……  ……  ……  …… 
202nd, from all AP of extracting data of collection, according to MAC Address ascending sort.It is intended that positioning stage
The SVM algorithm used and vector are order dependent, it is therefore necessary to artificially determine that a kind of vector puts in order.In the embodiment of the present invention
In, the MAC ascending orders arrangement using AP is used as sort method.
203rd, by the AP combination of two after sequence into (AP_{i},AP_{j}) (wherein, 0<i<J≤m, m represent all AP number),
AP is divided into according to MAC AddressGroup.Go out the two of corresponding A P combinations from each RSSI extracting datas that marked sampled point
RSSI vectors are tieed up, as classification initial data.Such as table 2, shown in table 3.
Table 2 extracts data format
School punctuate numbering  AP1  AP2 
1  85  97 
1  83  92 
2  70  73 
……  ……  …… 
Table 3 extracts data format
School punctuate numbering  AP2  AP3 
1  97  63 
1  92  65 
2  73  95 
……  ……  …… 
204th, using the vector that step 203 constructs as input, using the distance between vector as measuring similarity function, use
It can find that the Xmeans algorithms of clusters number carry out cluster analysis automatically.The each twodimentional AP combinations of record are to whole positioning respectively
The dividing condition in region.
The specific implementation process of Xmeans algorithm cluster analyses is as follows：
Step1. clusters number k scopes [k is specified_{min},k_{max}], and initialize k=k_{min}.K scope is according to actual area to be measured
The size selection in domain, the scope per subregions is in 200m^{2}To 700m^{2}, [k is calculated in this approach_{min},k_{max}]；
Step2. k AP number of combinations strong point u is randomly selected in the set of eigenvectors EV extracted from step 202_{1},u_{2},
u_{3}...u_{k}As initial cluster center；Set of eigenvectors EV such as tables 2, shown in table 3, select k characteristic vector and be used as just
Beginning center；
Step3. for each AP number of combinations strong point x in set of eigenvectors EV^{i}, according to belonging to similarity judges it
Class cluster,Wherein s (arg_{1},arg_{2}) it is Similarity Measure function；
Step4. above procedure is repeated, all data points are all assigned to most like class cluster, so as to by all AP groups
Data point all Preliminary divisions are into corresponding class cluster；
Step5. for each class cluster, its cluster centre is recalculated,Its
In, c^{(i)}Represent data point x^{i}It is preliminary to assert affiliated type；c^{(i)}=j is referred to：If data point x^{i}Belong to class cluster j, then (c^{(i)}=j)=1, otherwise (c^{(i)}=j)=0；The center represents the weighted average center position of each class cluster；
Step6. calculation criterion function,Wherein x^{i}It is the data in data set
Point, u_{j}It is class cluster j cluster centre；K refers to the number of cluster centre；
Step7. if criterion function, which no longer changes, turns to Step8, illustrate that the cluster result has been stablized；Otherwise jump to
Step3, restart cluster；
Step8. to having gathered the bayesian information criterion before and after each class cluster carries out further division and computation partition
BIC_{pre},BIC_{post}；Bayesian information criterion (Bayesian Information Criterions, BIC) is bayesian theory
An important component, the different models in same data set can be evaluated based on posterior probability, are suitable as
Selection complexity is relatively low and the reference frame of preferable model is described to data set.
Wherein for Clustering Model corresponding to clusters number k, the calculation formula of bayesian information criterion：Wherein, EV is the set of the characteristic vector extracted in step 202；R is in EV
Comprising characteristic vector number, here characteristic vector number be equal to AP groups collect all positions RSSI combine
Number；P represents number of parameters, referred to as Schwarz criterions, and its calculation formula is p=k+kd in the present invention, wherein, d EV
The dimension of middle characteristic vector, i.e. d=2；It is considered as the punishment to Clustering Model complexity；It is cluster mould
Type M_{k}Maximum posteriori loglikelihood estimation on characteristic vector set EV, its calculation formula such as following formula institute
Wherein,u_{(i)}For class cluster i cluster centre；
Step9. if BIC_{pre}＞ BIC_{post}, whether than original score higher, fraction height then receives point if watching results model
Split, turn to Step10, otherwise make k=k+1 and jump to Step8；
Step10. if k ＞ k_{max}, then need to restart cluster, turn to Step7；Otherwise make k=k+1 and jump to
Step2, calculate the cluster situation of one class of increase；
Step11. the maximum dividing modes of BIC are chosen as cluster result；
It is assumed that M is model set corresponding to different clusters number k, then have It is as optimal poly
Class model.Each type is expressed as a kind of signal characteristic, and signal characteristic reflects the aggregation of the signal distributions in a certain region
Phenomenon.
Fig. 3 is another cluster flow chart of the inventive method, and the flow belongs to a part for offline sample phase.Specifically
It may include steps of：
301st, smart mobile phone high frequency sweep periphery AP signals, the data format scanned such as table 1 are used in each school punctuate
It is shown.It should be noted that the number of data of each school punctuate collection is not fixed, it is different because of acquisition time length.It is if current
Position fails to collect corresponding AP RSSI, is filled up with 100dB.
302nd, AP is divided into m groups according to MAC Address (m represents all AP number).From marked each of sampled point
RSSI extracting datas go out corresponding A P onedimensional RSSI vectors, as classification initial data.Such as table 4, shown in table 5.
Table 4 extracts data format
School punctuate numbering  AP1 
1  85 
1  83 
2  70 
……  …… 
Table 5 extracts data format
School punctuate numbering  AP2 
1  97 
1  92 
2  73 
……  …… 
303rd, using the vector that step 302 constructs as input, carried out using the distance between vector as measuring similarity function
Cluster analysis, cluster analysis is using the Xmeans algorithms that can find clusters number automatically, and clustering method is similar with step 203, area
The onedimensional vector of single AP be not changed in the bivector for combining the characteristic vector in step 203 by AP.Signal mode is anti
The clustering phenomena of the signal distributions in a certain region has been reflected, has recorded dividing conditions of each AP to whole localization region respectively.Need
It should be noted that for same localization region, divisions of the different AP to the region can be different, and its reason is these AP
Deployed position spatially wide apart, it is abnormal by non line of sight transmission effects, multipath transmisstion effect and RSSI attenuation laws
It is different, it is thus possible to difference to be produced to division result.
Fig. 4 is the training flow chart of the inventive method, and the flow belongs to a part for offline sample phase.It can specifically wrap
Include following steps：
401st, the twodimentional RSSI vectors of AP combination of two are extracted in each calibration point, such as table 2, shown in table 3, and will be demarcated
The numbering of point replaces with the class number after corresponding cluster.
402nd, single AP onedimensional RSSI vectors are extracted in each calibration point, such as table 4, shown in table 5, and by calibration point
Numbering replaces with the class number after corresponding cluster.
403rd, to step 401,402 obtained vectors carry out SVM training respectively, calculate the characteristic of division of SVMs
Value.The followup validity for judging initialisation range and reducing orientation range that is calculated as of characteristic of division value provides data support.
The characteristic of division that the present embodiment is chosen is exactly the RSSI vectors that step 401,402 obtains.
Fig. 5 is a kind of coarseness positioning flow figure of the inventive method, and the flow belongs to a part for online stage.Specifically
It may include steps of：
501st, each twodimentional AP groups (AP trained is loaded_{i},AP_{j}) svm classifier model, read and currently collect
RSSI, according to AP MAC Address ascending sort, then extract every group of AP combination and be used as class vector.It should be noted that SVM is calculated
Method is order dependent with vector, it is therefore necessary to artificially determines that a kind of vector puts in order.In this example, using AP MAC ascending orders
Arrangement is used as sort method.
502nd, the AP for extracting step 501 combines the class vector to be formed, and it is carried out using corresponding SVM models pre
Survey, obtain every group of AP respectively under its corresponding region partition mode in the probability of regional.Because current location is likely to be at
The edge in multiple regions, or because some in twodimentional AP combination or some AP are due to by non line of sight transmission effects, more
The reasons such as footpath propagation effect and RSSI attenuation laws are abnormal cause RSSI to fluctuate, it is possible that multiple regions are all satisfactory
Situation, it can be chosen in accordance with the following methods：
The characteristic vector that each AP samplings extract, multiple prediction knots are might have after corresponding SVM model predictions
Fruit meets the requirements, and each prediction result corresponds to the subregions that the AP groups mark off area to be targeted.Behalf in above formula
Qualified prediction result number, that is, represent SVM qualification of model current devices and be likely to be in several subregions；area_{k}Table
Show kth of qualified region, that is, representing SVM qualification of model current device may be in which subregions.Area(AP_{i},
AP_{j}) represent AP groups (AP_{i},AP_{j}) determined by region where current location, that is, represent SVM models and think that current device may
Union in these subregions.The choosing method of satisfactory subregion is existed if predicting current signature vector
The probability of certain subregions is not less than some threshold epsilon (0<ε<1) subregion, is considered as to meet the requirements.In this example, selection
ε=1/n, n represent the subregion number that AP combinations mark off.
503rd, all AP groups (AP to being obtained in step 502_{i},AP_{j}) Area (AP_{i},AP_{j}) use " ballot " mode meter
Calculate positioning result.May be in a certain region if the sample data of certain AP combinations is identified as by step 502 prediction, should
Region poll adds 1.Travel through the Area (AP of all AP combinations_{i},AP_{j}) and vote, the most region of poll is selected as the thick of positioning
Granularity localization region, the poll in each region arrive 0Between.From geometrically showing as choosing by Area (AP_{i},AP_{j}) covering
Localization region of the most region of number as coarseness.If the poll of all areas is both less than a certain threshold xi, it is believed that positioning
Failure, terminate positioning；If there is multiple regions poll at most and be more than ξ, then the union in these regions is sought, as positioning
Coarseness localization region.Because normal positioning at least needs 3 AP to participate in calculating, therefore in this example, ξ values are 4.
Fig. 6 is a kind of fine granularity positioning flow figure of the inventive method, and the flow belongs to a part for online stage.Specifically
It may include steps of：
601st, the svm classifier model for each AP that loading trains before, the RSSI currently collected is read, by the RSSI
An onedimensional class vector is formed, each AP trained before use svm classifier model is predicted to it, asked respectively
Go out each AP under its corresponding region partition mode in the probability of regional, therefrom choose the coarseness positioning that previous step is obtained
The probability of regional in the R of region.Because current location is likely to be at the edge in multiple regions, or due to the AP due to by
RSSI is caused to fluctuate to reasons such as non line of sight transmission effects, multipath transmisstion effect and RSSI attenuation laws exceptions, it is possible that
Multiple regions are all more conform with the situation of requirement, can be chosen in accordance with the following methods：
The characteristic vector that each AP samplings extract, multiple prediction knots are might have after corresponding SVM model predictions
Fruit meets the requirements, and each prediction result corresponds to the subregions that the AP marks off area to be targeted.Behalf accords with above formula
The prediction result number of conjunction condition, that is, represent SVM qualification of model current devices and be likely to be in several subregions；area_{k}Represent
Kth of qualified region, that is, representing SVM qualification of model current device may be in which subregions, and the region is necessary
It is the coarseness localization region R obtained in step 4 subset.Area(AP_{i}) represent AP_{i}Area where identified current location
Domain, that is, represent SVM models and think that current device is likely to be at the union of these subregions.The selection of satisfactory subregion
Method is, if the probability for predicting current signature vector in certain subregions is not less than some threshold epsilon (0<ε<1), it is considered as
The subregion meets the requirements.In this example, ε=1/n of selection, n represent the subregion number that the AP is marked off.
602nd, to all AP in step 601_{i}Area (AP_{i}) calculated using " ballot " mode, i.e., if certain AP sample
Data are identified as in a certain region by step 601 prediction, then the region poll adds 1.Travel through all AP Area (AP_{i})
And vote, the most region of poll is selected as the fine granularity localization region positioned, and the poll in each region is 0 between m.From
Geometrically show as choosing by Area (AP_{i}) the most region of degree of covering as coarseness localization region.If all areas
The poll in domain is both less than a certain threshold xi, it is believed that positioning failure, terminates positioning；It is at most and big if there is the poll in multiple regions
In ξ, then the union in these regions is sought, obtains its center point coordinate and radius, as final localization region.Due to normal positioning
3 AP are at least needed to participate in calculating, therefore in this example, ξ values are 4.
Nonelaborated part of the present invention belongs to techniques well known.
It is described above, part embodiment only of the present invention, but protection scope of the present invention is not limited thereto, and is appointed
What those skilled in the art the invention discloses technical scope in, the change or replacement that can readily occur in should all be covered
Within protection scope of the present invention.
Claims (4)
1. a kind of indoor orientation method based on WLAN, it is characterised in that realize that step is as follows：
Step 1：The RSSI data predictions for each AP that sampled point is collected, therefrom extract a peacekeeping bivector point
Zuo Wei not characteristic vector；
Step 2：Feature vector clusters are analyzed, area to be targeted is divided into multiple positioning subregions；
Step 3：For every group of characteristic vector combination cluster result, respective disaggregated model is respectively trained out；Based on classification
Models coupling " ballot " mechanism chooses poll highest subregion set from all subregions；
Step 4：Positioned using twowheeled and reduce subregion range of convergence, improve positioning precision；
The step 1 extracts a peacekeeping bivector respectively as characteristic vector from preprocessed data, including：
(1) by all AP scanned according to MAC Address ascending sort；
(2) a peacekeeping bivector is extracted as characteristic vector according to following two methods：
A. by the AP combination of two after sequence, AP is divided into according to MAC AddressGroup, every group of AP are expressed as (AP_{i},AP_{j}), its
In, 0<i<J≤m, m represent all AP number, and the vector that these AP combination compositions are gone out from pretreated extracting data is made
It is characterized vector；
B. each AP is separately as one group, will all offline gathered datas be divided into m groups, every group of AP table according to AP MAC Address
It is shown as AP_{i}, wherein, 0<I≤m, m represent all AP number, from pretreated extracting data go out these AP composition to
Amount is used as characteristic vector；
The step 3, specific implementation process include offline phase and online stage；
Offline phase, the characteristic vector constructed for the two kinds of structures method proposed in step 1, is respectively trained out every kind of structure
Make SVMs (Support Vector Machine, SVM) disaggregated model corresponding to every feature vectors of method；
Online stage, from real time data extraction characteristic of division vector, the svm classifier model that offline phase trains is read, according to
Supporting vector polynomial expansion entry value, the probability that vector to be sorted corresponds to different zones is calculated, with reference to " ballot " mechanism from institute
Have and poll highest set of regions R is chosen in region；
The voting mechanism refers to, if AP groups (AP_{i},AP_{j}) sample data by SVM prediction be identified as in a certain region
Interior, then the region poll adds 1；Travel through the EV (AP of all AP groups_{i},AP_{j}) and vote, the most region of poll is selected as positioning
Coarseness localization region, the poll in each region should arrive 0Between, EV is characterized vector set；
The step 4, positioned using twowheeled and reduce set of regions scope, be implemented as：
(1) the svm classifier model trained is read, calculates supporting vector polynomial expansion entry value；
(2) RSSI currently collected, extraction characteristic of division vector are read, and characteristic of division is standardized；
(3) by Polynomial kernel function by characteristic of division DUAL PROBLEMS OF VECTOR MAPPING to higher dimensional space, and according to the supporting vector multinomial
Deploy entry value and calculate the probability that vector to be sorted corresponds to different zones, the coarseness positioning area therefrom obtained in selecting step three
The probability of regional in the R of domain；
(4) for each AP_{i}, judge whether every subregions for marking off are eligible, and the subregion is obtained in step 3
Coarseness localization region R subset, it is eligible if there is more subregions, then SVM models think current device may
In union in these subregions；
(5) poll highest set of regions R ' is chosen from R with reference to " ballot " mechanism, specific steps include：If AP_{i}Sample number
It is identified as according to by SVM predictions in a certain region, then the region poll adds 1, is voted according to each AP localization region, choosing
Fine granularity localization region of the most region of booking number as positioning, the poll in each region should be 0 between m.
2. the indoor orientation method according to claim 1 based on WLAN, it is characterised in that：The step 1 is by sampled point
The each AP collected RSSI data predictions, including：The too low data of RSSI are deleted, nonlocating AP data is deleted, fills out
Mend the RSSI data not being scanned；
Deletion data too low RSSI refer to, data of the RSSI intensity less than a certain threshold value are deleted；The deletion is nonfixed
Position AP data refer to, delete the RSSI for the AP for being unsuitable for positioning, and the feature for being unsuitable for positioning AP is that intensity is too low, i.e., RSSI is small
In 95dB or less stable, i.e. variance is more than 20.
3. the indoor orientation method according to claim 1 based on WLAN, it is characterised in that：In the step 2, to spy
Vector clusters analysis is levied, area to be targeted is divided into multiple positioning subregions, concretely comprised the following steps：With the spy constructed in step 1
Sign vector is input, and cluster analysis is carried out using the distance between characteristic vector as measuring similarity function, and cluster analysis uses
The Xmeans algorithms of clusters number can be found automatically.
4. the indoor orientation method according to claim 1 based on WLAN, it is characterised in that：The tuning online stage
Concrete operations include：
(1) the svm classifier model trained is read, calculates supporting vector polynomial expansion entry value；
(2) RSSI currently collected, extraction characteristic of division vector are read；
(3) by Polynomial kernel function by characteristic of division DUAL PROBLEMS OF VECTOR MAPPING to higher dimensional space, and according to the supporting vector multinomial
Deploy entry value and calculate the probability that vector to be sorted corresponds to different zones；
(4) for each AP groups (AP_{i},AP_{j}), judge whether every subregions for marking off are eligible, if there is more height
Region is eligible, then SVM models think that current device is likely to be in the union of these subregions.
Priority Applications (1)
Application Number  Priority Date  Filing Date  Title 

CN201410458932.0A CN104185275B (en)  20140910  20140910  A kind of indoor orientation method based on WLAN 
Applications Claiming Priority (1)
Application Number  Priority Date  Filing Date  Title 

CN201410458932.0A CN104185275B (en)  20140910  20140910  A kind of indoor orientation method based on WLAN 
Publications (2)
Publication Number  Publication Date 

CN104185275A CN104185275A (en)  20141203 
CN104185275B true CN104185275B (en)  20171117 
Family
ID=51965929
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

CN201410458932.0A CN104185275B (en)  20140910  20140910  A kind of indoor orientation method based on WLAN 
Country Status (1)
Country  Link 

CN (1)  CN104185275B (en) 
Families Citing this family (17)
Publication number  Priority date  Publication date  Assignee  Title 

CN104463929B (en) *  20141216  20170718  重庆邮电大学  Indoor WLAN signal mapping and mapping method based on Image EdgeDetection signal correlation 
CN104853434A (en) *  20150113  20150819  中山大学  Indoor positioning method based on SVM and K mean value clustering algorithm 
CN105044662B (en) *  20150527  20190301  南京邮电大学  A kind of fingerprint cluster multipoint joint indoor orientation method based on WIFI signal intensity 
CN105334493B (en) *  20151009  20171226  北京航空航天大学  A kind of indoor orientation method based on WLAN 
CN109212464B (en) *  20160606  20200114  中科劲点（北京）科技有限公司  Method and equipment for estimating terminal distance and position planning 
CN106131958A (en) *  20160809  20161116  电子科技大学  A kind of based on channel condition information with the indoor Passive Location of support vector machine 
CN106131959B (en) *  20160811  20190514  电子科技大学  A kind of dualpositioning method divided based on WiFi signal space 
CN106412838B (en) *  20160910  20191018  华南理工大学  A kind of bluetooth indoor orientation method based on statistical match 
CN106643736B (en) *  20170106  20200522  中国人民解放军信息工程大学  Indoor positioning method and system 
CN107087256A (en) *  20170317  20170822  上海斐讯数据通信技术有限公司  A kind of fingerprint cluster method and device based on WiFi indoor positionings 
CN107290714B (en) *  20170704  20200221  长安大学  Positioning method based on multiidentification fingerprint positioning 
CN108712723B (en) *  20180508  20190531  深圳市名通科技股份有限公司  AP similarity determines method, terminal and computer readable storage medium 
CN108924756B (en) *  20180630  20200818  天津大学  Indoor positioning method based on WiFi dualband 
CN109286900B (en) *  20180829  20200717  桂林电子科技大学  WiFi sample data optimization method 
CN109640262B (en) *  20181130  20210105  哈尔滨工业大学（深圳）  Positioning method, system, equipment and storage medium based on mixed fingerprints 
CN110213710A (en) *  20190419  20190906  西安电子科技大学  A kind of highperformance indoor orientation method, indoor locating system based on random forest 
CN110519692B (en) *  20190912  20201002  中南大学  Positioning and partitioning method based on Bayesk mean clustering 
Citations (3)
Publication number  Priority date  Publication date  Assignee  Title 

CN103402256A (en) *  20130711  20131120  武汉大学  Indoor positioning method based on WiFi (Wireless Fidelity) fingerprints 
CN103533647A (en) *  20131024  20140122  福建师范大学  Radio frequency map selfadaption positioning method based on clustering mechanism and robust regression 
CN103648106A (en) *  20131231  20140319  哈尔滨工业大学  WiFi indoor positioning method of semisupervised manifold learning based on category matching 
Family Cites Families (2)
Publication number  Priority date  Publication date  Assignee  Title 

CA2638754C (en) *  20010905  20130402  Newbury Networks, Inc.  Position detection and location tracking in a wireless network 
US8224349B2 (en) *  20100225  20120717  At&T Mobility Ii Llc  Timed fingerprint locating in wireless networks 

2014
 20140910 CN CN201410458932.0A patent/CN104185275B/en active IP Right Grant
Patent Citations (3)
Publication number  Priority date  Publication date  Assignee  Title 

CN103402256A (en) *  20130711  20131120  武汉大学  Indoor positioning method based on WiFi (Wireless Fidelity) fingerprints 
CN103533647A (en) *  20131024  20140122  福建师范大学  Radio frequency map selfadaption positioning method based on clustering mechanism and robust regression 
CN103648106A (en) *  20131231  20140319  哈尔滨工业大学  WiFi indoor positioning method of semisupervised manifold learning based on category matching 
NonPatent Citations (2)
Title 

A Novel Algorithm for Enhancing Accuracy of Indoor Position Estimation;Xiaoqing Lu et al.;《Proceeding of the 11th World Congress on Intelligent Control and Automation》;20140704;全文 * 
A Novel ClusteringBased Approach of Indoor Location Fingerprinting;ChungWei Lee et al.;《2013 IEEE 24th International Symposium on Personal Indoor and Mobile Radio Communications (PIMRC)》;20131125;全文 * 
Also Published As
Publication number  Publication date 

CN104185275A (en)  20141203 
Similar Documents
Publication  Publication Date  Title 

Soh et al.  Adaptive deep learningbased air quality prediction model using the most relevant spatialtemporal relations  
Campos et al.  WiFi multifloor indoor positioning considering architectural aspects and controlled computational complexity  
EP3241370B1 (en)  Analyzing semantic places and related data from a plurality of location data reports  
Mohamed et al.  Accurate realtime map matching for challenging environments  
CN104469676B (en)  The localization method and system of a kind of mobile terminal  
CN100579307C (en)  Wireless node location method and device for featuring definition of search region to optimize location computation  
Haghani et al.  Data collection of freeway travel time ground truth with bluetooth sensors  
CN104010364B (en)  For determining the method and system in the geographical location of the estimation of base station  
CN104703143B (en)  A kind of indoor orientation method based on WIFI signal intensity  
CN102170697B (en)  Indoor positioning method and device  
US20160295547A1 (en)  Methods and systems of assigning estimated positions and attributes to wireless access points in a positioning system  
CN102521973B (en)  A kind of mobile phone switches the road matching method of location  
CN103119470B (en)  Carry out the system and method for the positionbased service of indoor navigation  
CN104635203B (en)  Radio interference source directionfinding and positioning method based on particle filter algorithm  
US8504035B2 (en)  System and method for population tracking, counting, and movement estimation using mobile operational data and/or geographic information in mobile network  
Kjærgaard et al.  Hyperbolic location fingerprinting: A calibrationfree solution for handling differences in signal strength (concise contribution)  
CN102098780B (en)  Positioning method and device  
CN104853317B (en)  The structure and update method of fingerprint base in a kind of WiFi indoor positionings  
Diaz et al.  Bluepass: An indoor bluetoothbased localization system for mobile applications  
CN103476115B (en)  A kind of WiFi fingerprint positioning method based on AP collection similarity  
Li et al.  Fingerprint and assistant nodes based WiFi localization in complex indoor environment  
CN103197280B (en)  Access point (AP) location estimation method based on radiofrequency signal strength  
CN103874118B (en)  Radio Map bearing calibrations in WiFi indoor positionings based on Bayesian regression  
US8190172B2 (en)  Indoor/outdoor decision apparatus and indoor/outdoor decision method  
US9313617B2 (en)  Identification of points of interest and positioning based on points of interest 
Legal Events
Date  Code  Title  Description 

C06  Publication  
PB01  Publication  
C10  Entry into substantive examination  
SE01  Entry into force of request for substantive examination  
GR01  Patent grant  
GR01  Patent grant 