CN103648106A - WiFi indoor positioning method of semi-supervised manifold learning based on category matching - Google Patents

WiFi indoor positioning method of semi-supervised manifold learning based on category matching Download PDF

Info

Publication number
CN103648106A
CN103648106A CN201310750528.6A CN201310750528A CN103648106A CN 103648106 A CN103648106 A CN 103648106A CN 201310750528 A CN201310750528 A CN 201310750528A CN 103648106 A CN103648106 A CN 103648106A
Authority
CN
China
Prior art keywords
rss
radio map
algorithm
classification
online
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201310750528.6A
Other languages
Chinese (zh)
Other versions
CN103648106B (en
Inventor
谭学治
周才发
马琳
邓仲哲
何晨光
迟永钢
魏守明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Harbin University of Technology Robot Group Co., Ltd.
Original Assignee
Harbin Institute of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Harbin Institute of Technology filed Critical Harbin Institute of Technology
Priority to CN201310750528.6A priority Critical patent/CN103648106B/en
Publication of CN103648106A publication Critical patent/CN103648106A/en
Application granted granted Critical
Publication of CN103648106B publication Critical patent/CN103648106B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Abstract

The invention discloses a WiFi indoor positioning method of semi-supervised manifold learning based on category matching, and relates to an indoor positioning method. The WiFi indoor positioning method disclosed by the invention is used for solving the problems that a Radio Map database is large and the like in an existing WiFi indoor positioning method. The WiFi indoor positioning method comprises the following steps: 1. collecting Radio Map; 2. carrying out intrinsic dimension analysis on the Radio Map; 3. carrying out clustering analysis on the Radio Map; 4. carrying out dimensionality reduction on the Radio Map; 5. adding RSS in the Radio Map to obtain Radio Mapul; and 6. carrying out dimensionality reduction on the Radio Mapul to obtain a characteristic transformation matrix V, and forming an online positioning database through the Radio Map * and V. The WiFi indoor positioning method also comprises the following steps: 1. online testing RSS; 2. carrying out dimensionality reduction on the RSS to obtain RSS *; 3. outputting a positioning result; and 4. updating the database. The WiFi indoor positioning method disclosed by the invention is applied to the field of network technology.

Description

A kind of WiFi indoor orientation method of the semi-supervised manifold learning based on classification coupling
Technical field
The present invention relates to a kind of indoor orientation method, be specifically related to a kind of WiFi indoor orientation method of the semi-supervised manifold learning based on classification coupling.
Background technology
Along with WLAN is popularized the extensive of worldwide develop rapidly and mobile terminal device, technology and application that many indoor positioning are relevant have been there is in recent years.Due to the complexity of multipath effect, signal attenuation and indoor positioning environment, the indoor orientation method of the signal propagation model based on traditional is difficult to reach high-precision indoor positioning requirement.Based on the time of advent (Time of Arrival), the time of advent poor (Time Difference of Arrival) and arrive angle (Angles of Arrival) although etc. localization method can substantially meet positioning precision demand, yet all need locating terminal to have extra hardware device support, there is larger limitation, thereby cause the indoor locating system based on above-mentioned a few class localization methods not popularized.
At present, the WiFi indoor orientation method based on WLAN location fingerprint (Finger Print) is widely applied.The network establishing method of the method is with low cost, and it uses 2.4GHz ISM (Industrial Science Medicine) common frequency band and without add location survey specialized hardware on existing utility.Access point (the Access Point that only need to receive by wireless network card and the corresponding software measurement of mobile terminal, AP) signal strength signal intensity (Received Signal Strength, RSS), build thus network signal coverage diagram (Radio Map), and then predict the coordinate of mobile subscriber present position by matching algorithm, or relative position.
Yet the Radio Map setting up by which includes huge data message, and along with locating area expands, Radio Map may (select) to be index situation and increase according to position matching mode and algorithm.Obtain related data characteristic information as much as possible and can promote positioning precision for whole system, but process a large amount of characteristic informations and increase algorithm expense, location algorithm cannot effectively operation on the limited mobile terminal of disposal ability, some characteristic information may be not act on and even have negative effect for location simultaneously, cause matching efficiency to reduce, thereby cause the realization of mating location algorithm to become more complicated, and positioning precision decline.
When the number increase of AP and reference point (Reference Point) increase of location, the data message of Radio Map increases.The information of the AP number now, representing in Radio Map has represented the dimension of data.Therefore,, when AP number increases, Radio Map has just become high dimensional data.For alleviating the burden of processing high dimensional data, dimension-reduction algorithm is one of effective solution.High dimensional data may comprise a lot of features, and these features are all being described same things, and these features are to be closely connected to a certain extent.As when same object being taken pictures from all angles simultaneously, the data that obtain just contain overlapping information.If can obtain nonoverlapping expression of some simplification of these data, will greatly improve the efficiency of data processing operation and improve to a certain extent accuracy.The object of dimension-reduction algorithm is also to improve the treatment effeciency of high dimensional data just.
Except can reduced data can efficiently processing, dimension reduction method can also be realized data visualization.Due to a lot of statistical very poor for the accuracy of optimal solution with machine learning algorithm, the visual application of dimensionality reduction can make user can actually see the space structure of high dimensional data and the ability of algorithm output, has very strong using value.
There are at present a lot of dimension-reduction algorithms based on different objects, include linearity and nonlinear reductive dimension algorithm.Wherein PCA (Principal Component Analysis) and LDA (Linear Discriminant Analysis) are typical linear dimension-reduction algorithms.This class algorithm has good result for the high dimensional data with linear structure, but the result not had for the high dimensional data of nonlinear organization.Typical nonlinear reductive dimension algorithm is with manifold learning (Manifold Learning) algorithm.On Science magazines in 2000, with first phase, 3 pieces have been delivered relevant for having proposed 2 kinds of classical manifold learning arithmetic: LLE (Local Linear Embedding) and ISOMAP (Isometric Mapping) in manifold learning arithmetic.Thus, the manifold learning arithmetic of the various criterions based on different is suggested also some manifold learning arithmetic and is applied to image processing method face.LDE(Local Discriminant Embedding) algorithm is more late proposition in manifold learning arithmetic, and it is a kind of typical manifold learning arithmetic based on feature extraction, and not only based on visual target.
For above-mentioned dimension-reduction algorithm, can not improve to the RSS data of online acquisition or newly-increased RSS the precision of dimensionality reduction.Due to the variation of indoor positioning environment, not in the same time, particularly after long-time, the correlation between RSS data will reduce.In existing algorithm, there is a class algorithm RSS data at different time can be added in corresponding data with existing simultaneously, thereby strengthen the correlation between different pieces of information, improve dimensionality reduction precision.This class algorithm is referred to as semi supervise algorithm conventionally.According to the feature of semi supervise algorithm, propose semi-supervised discriminating and embed (Semi-supervised Discriminant Embedding, SDE) algorithm.
According to the feature of SDE algorithm, adopt the mode of classification coupling that the RSS newly obtaining is online added to existing Radio Map, then carry out part and differentiate embedding dimensionality reduction, thereby semi-supervised local discriminating thing propose based on classification coupling embeds algorithm (Classification Matching Based Semi-supervised Discriminant Embedding, CM-SDE).Adopt CM-SDE algorithm to carry out dimensionality reduction to Radio Map, draw the Radio Map after dimensionality reduction, by the Radio Map indoor positioning to after dimensionality reduction, thereby propose the WiFi indoor positioning algorithm based on CM-SDE algorithm.
Summary of the invention
The present invention is that will to solve the Radio Map database existing in existing WiFi indoor orientation method large, and being difficult to of causing because online positioning stage computation complexity is high apply on-line stage and obtain RSS data, be difficult to the problems such as requirement of real-time that realize and be difficult to guarantee location at mobile terminal, and provide a kind of WiFi indoor orientation method of the semi-supervised manifold learning based on classification coupling.
The WiFi indoor orientation method off-line phase position fixing process of the semi-supervised manifold learning based on classification coupling is realized according to the following steps:
One, room area to be positioned is arranged to AP, make wireless signal cover room area to be positioned, complete WiFi network struction;
In room area rule to be positioned, choose and record the corresponding coordinate of reference point, measure and record successively the RSS signal of all AP that reference point receives as position feature information, build Radio Map, and store Radio Map;
Two, adopt GMST intrinsic dimension algorithm for estimating to analyze the intrinsic dimension of the Radio Map building in step 1, the intrinsic dimension obtaining, as one of input parameter of CM-SDE algorithm, determines the dimension after Radio Map dimensionality reduction;
Three, adopt KFCM algorithm to carry out cluster analysis to Radio Map, realize the classification mark of the Radio Map setting up, and as one of input parameter of CM-SDE, and corresponding initial cluster center and classification mark are provided;
Four, the intrinsic dimension in step 2 and the classification mark in step 3, as input parameter, adopt CM-SDE algorithm to the Radio Map dimensionality reduction building in step 1, draw the RadioMap after corresponding dimensionality reduction *, RadioMap *as being used for online positioning stage in coupling location database;
Five, the unmarked RSS online positioning stage test of different user being obtained, adopts the mode of classification coupling to be added in existing Radio Map, obtains comprising accordingly unmarked Radio map RadioMap ul, the cluster centre that mates renewal by classification is as classification input parameter new in CM-SDE algorithm;
Six, the cluster centre of the renewal in step 5, as input parameter, adopts CM-SDE to RadioMap uldimensionality reduction obtain eigentransformation matrix V ', V ' and RadioMap *common formation On-line matching location database, locates for on-line stage; Wherein, described line phase orientation is specially:
Six (one), on-line testing RSS;
Six (two), adopting V is RSS by RSS dimensionality reduction *;
Six (three), adopt KNN algorithm to mate location output positioning result;
Six (four), user's locating terminal location database is upgraded;
Completed a kind of off-line phase implementation of WiFi indoor orientation method of the semi-supervised manifold learning based on classification coupling.
The WiFi indoor orientation method on-line stage position fixing process of the semi-supervised manifold learning based on classification coupling is realized by following step:
One, on-line testing RSS;
Two, the RSS in the site undetermined obtaining is adopted the conversion of eigentransformation matrix dimensionality reduction obtain RSS*;
Three, adopt KNN algorithm to RSS* and Radio Map* match bit, treat the particular location coordinate of anchor point and predict and carry out the renewal of online data, its implementation procedure is:
(1) on-line stage, the RSS=[AP that test point place receives 1, AP 2..., AP n], with eigentransformation matrix V ' multiply each other, thereby draw the RSS '=[AP after dimensionality reduction 1, AP 2... AP d], wherein d represents intrinsic dimension;
(2) adopt KNN algorithm to realize mating of RSS ' and Radio Map*, adopt and the mean value of the coordinate of K the reference point that RSS ' is nearest as test point (x ', y '), its expression formula is:
( x ′ , y ′ ) = 1 K Σ i = 1 K ( x i , y i )
In formula, (x ', y ') is the coordinate of test point prediction, (x i, y i) be the coordinate of i Neighbor Points, K is the number of neighbour in KNN algorithm;
Four, user's locating terminal location database is upgraded, and has completed the WiFi on-line stage indoor orientation method of the semi-supervised manifold learning based on classification coupling.
Invention effect:
The CM-SDE algorithm and the background that for the present invention, propose need to be asked, by the research park 2A of Harbin Institute of Technology the indoor positioning region that 12 floor corridor form positioned.Adopt association's V450 notebook computer and in conjunction with the RSS at all reference points of NetStumbler software test place, form Radio Map, and adopt CM-SDE algorithm to Radio Map dimensionality reduction and adopt KNN algorithm to realize indoor positioning.In the emulation of accompanying drawing 4, the dimension of the Radio Map after dimensionality reduction is 1/3rd of original Radio Map dimension.Simulation result from accompanying drawing 4, adopt the positioning performance of the algorithm of CM-SDE algorithm and initial KNN to compare, but the location complexity of CM-SDE is only 1/3rd, and CM-SDE algorithm can be applied the density that the new RSS obtaining in real time carrys out Radio Map effectively, thereby effectively improve positioning precision.
Accompanying drawing explanation
Fig. 1 is offline database location implementing procedure figure in the present invention; Solid arrow represents the transfer of data between step;
Fig. 2 is online database location implementing procedure figure in the present invention; Solid arrow represents the transfer of data between step;
Fig. 3 is structure and the experimental situation schematic diagram of the indoor positioning network based on WiFi;
Fig. 4 is sampling network trrellis diagram;
Fig. 5 adopts CM-SDE and KNN algorithm positioning performance comparison diagram; Wherein,
Figure BDA0000451371200000042
represent CM-SDE,
Figure BDA0000451371200000043
represent KNN.
Embodiment
Embodiment one: the WiFi indoor orientation method off-line phase position fixing process of the semi-supervised manifold learning based on classification coupling of present embodiment is realized according to the following steps:
One, room area to be positioned is arranged to AP, make wireless signal cover room area to be positioned, complete WiFi network struction;
In room area rule to be positioned, choose and record the corresponding coordinate of reference point, measure and record successively the RSS signal of all AP that reference point receives as position feature information, build Radio Map, and store Radio Map;
Two, adopt GMST intrinsic dimension algorithm for estimating to analyze the intrinsic dimension of the Radio Map building in step 1, the intrinsic dimension obtaining, as one of input parameter of CM-SDE algorithm, determines the dimension after Radio Map dimensionality reduction;
Three, adopt KFCM algorithm to carry out cluster analysis to Radio Map, realize the classification mark of the Radio Map setting up, and as one of input parameter of CM-SDE, and corresponding initial cluster center and classification mark are provided;
Four, the intrinsic dimension in step 2 and the classification mark in step 3, as input parameter, adopt CM-SDE algorithm to the Radio Map dimensionality reduction building in step 1, draw the RadioMap after corresponding dimensionality reduction *, RadioMap *as being used for online positioning stage in coupling location database;
Five, the unmarked RSS online positioning stage test of different user being obtained, adopts the mode of classification coupling to be added in existing Radio Map, obtains comprising accordingly unmarked Radio map RadioMap ul, the cluster centre that mates renewal by classification is as classification input parameter new in CM-SDE algorithm;
Six, the cluster centre of the renewal in step 5, as input parameter, adopts CM-SDE to RadioMap uldimensionality reduction obtain eigentransformation matrix V ', V ' and RadioMap *common formation On-line matching location database, locates for on-line stage; Wherein, described line phase orientation is specially:
Six (one), on-line testing RSS;
Six (two), adopting V is RSS by RSS dimensionality reduction *;
Six (three), adopt KNN algorithm to mate location output positioning result;
Six (four), user's locating terminal location database is upgraded;
Completed a kind of off-line phase implementation of WiFi indoor orientation method of the semi-supervised manifold learning based on classification coupling.
Described off-line phase and on-line stage all complete at locating terminal.
Embodiment two: present embodiment is different from embodiment one: adopt GMST intrinsic dimension algorithm for estimating to analyze the intrinsic dimension of the Radio Map building in step 1 in step 2, its computing formula is:
Figure BDA0000451371200000051
geodesic distance minimal spanning tree algorithm
In above formula
Figure BDA0000451371200000052
in a represent the slope of the linear fit expression formula y=ax+b of minimum spanning tree.
Other step and parameter are identical with embodiment one.
Embodiment three: present embodiment is different from embodiment one or two: generate the Radio map RadioMap after new dimensionality reduction in step 4 *with the V ' expression formula in step 6 be:
Radio Map*=V′·X
X is the Radio Map that needs dimensionality reduction.
Other step and parameter are identical with embodiment one or two.
Embodiment four: present embodiment is different from one of embodiment one to three: the realization of classification coupling in step 5, its core process completes in two steps:
The first step, finds the category attribute of unmarked RSS, by following formula, completes category attribute mark:
x i ∈ G i ⇔ arg min 1 ≤ j ≤ c D ( x i , v j )
Second step: RSS is carried out to Threshold detection: by calculating and judge the relation of generalized symbol value and threshold value, thereby realize the renewal of Radio Map and classification flag data, the renewal of generalized symbol value and cluster centre respectively lower two formulas completes:
S j = Σ j = 1 N sgn ( RSS ij - v ij )
v i = 1 | G i | Σ x k ∈ G i x k
Wherein, described threshold value V t=0.9N.
Other step and parameter are identical with one of embodiment one to three.
Embodiment five: the WiFi on-line stage indoor orientation method of the semi-supervised manifold learning based on classification coupling is realized by following step:
One, on-line testing RSS;
Two, the RSS in the site undetermined obtaining is adopted the conversion of eigentransformation matrix dimensionality reduction obtain RSS*;
Three, adopt KNN algorithm to RSS* and Radio Map* match bit, treat the particular location coordinate of anchor point and predict and carry out the renewal of online data, its implementation procedure is:
(1) on-line stage, the RSS=[AP that test point place receives 1, AP 2..., AP n], with eigentransformation matrix V ' multiply each other, thereby draw the RSS '=[AP after dimensionality reduction 1, AP 2... AP d], wherein d represents intrinsic dimension;
(2) adopt KNN algorithm to realize mating of RSS ' and Radio Map*, adopt and the mean value of the coordinate of K the reference point that RSS ' is nearest as test point (x ', y '), its expression formula is:
( x ′ , y ′ ) = 1 K Σ i = 1 K ( x i , y i )
In formula, (x ', y ') is the coordinate of test point prediction, (x i, y i) be the coordinate of i Neighbor Points, K is the number of neighbour in KNN algorithm;
Four, user's locating terminal location database is upgraded, and has completed the WiFi on-line stage indoor orientation method of the semi-supervised manifold learning based on classification coupling;
Described off-line phase completes on server; On-line stage completes on locating terminal.
Embodiment six: present embodiment is different from embodiment five: the particular location coordinate for the treatment of anchor point in step 3 is predicted and the renewal of carrying out online data is realized by the offline database described in claim 1 and online database:
The implementation of off-line location database is: the RSS that this location of consumer positioning is recorded adopts classification matching way to join in Radio Map as Unlabeled data, and the renewal of realization to local On-line matching location database on mobile terminal, realize the local data of dynamic renewal, thereby realize offline database locate mode;
The implementation of online database is: after user has located online, the RSS value that this records online by user is uploaded to the server at online location database place, and at server end, online location database is upgraded to just online locator data and pass the locating terminal of uploading RSS data back.
Other step and parameter are identical with one of embodiment one to five.
Emulation experiment:
One, 3 pairs of these emulation experiments are described in detail by reference to the accompanying drawings: be illustrated as the plane graph signal of the research park 2A of Harbin Institute of Technology 12 floor, the indoor locating system based on WiFi is exactly based on setting up under this experimental situation.In experimental situation, altogether arrange 27 AP, the position that AP arranges is place, blue wireless transmission signal shape mark place.AP is 2 meters from room ground level.In off-line phase, on the V450 of association notebook, NetStumbler software is installed, in four of all reference points different orientation, 100 RSS values of AP are recorded in continuous sampling, and the relevant information of AP.The physical coordinates of all sampled points and corresponding physical coordinates and RSS value are stored as to the data that position fixing process calls, set up Radio Map.In experimental situation, have 900 reference points, its sampling density is 0.5 meter * 0.5 meter, as shown in Figure 4.Radio Map is as the input data of input parameter and the intrinsic dimension algorithm for estimating of CM-SDE algorithm.
Two, obtaining by following step of the intrinsic dimension of Radio Map realized:
Intrinsic dimension is for high dimensional data, to carry out the number of the independent variable of eigenspace dimension and the required minimum of space reconstruction.In concrete Practical Calculation, because the intrinsic of high dimensional data is also not obvious, not to seek to obtain definite intrinsic dimension conventionally, but seek to estimate the credible value of intrinsic dimension.Specifically, a given sample from higher dimensional space, the central task of intrinsic dimension algorithm for estimating and important content are exactly by these sample datas, to determine the intrinsic dimension of this higher-dimension structure.
The estimation of the intrinsic dimension of Radio Map is the important input parameter of CM-SDE algorithm, and whether this result that is related to dimensionality reduction can represent the feature of the higher dimensional space of Radio Map, and therefore the estimation of intrinsic dimension is most important accurately and effectively.At present, conventional intrinsic dimension algorithm for estimating is divided into two classes: partial estimation is estimated with the overall situation.Adopt Global Algorithm to estimate the intrinsic dimension of Radio Map to estimate, and as the input variable of CM-SDE algorithm.In this experiment, adopt geodesic curve minimal spanning tree algorithm (Geodesic Minimum Spanning Tree, GMST) to estimate the intrinsic dimension of Radio Map.
Below the theory of GMST algorithm is analyzed.
Geodesic curve minimum spanning tree (GMST) estimation is that the length function based on geodesic curve minimum spanning tree depends on intrinsic dimension d.GMST refers to the minimum spanning tree of the neighbour's curve being defined on data set X.The length function L of GMST (X) is the Euclidean distance sum that all edges are corresponding in geodesic curve minimum spanning tree.
GMST estimates on data set X, to construct neighbour's curve G, wherein, and each data point x in X iall with its k neighbour
Figure BDA0000451371200000081
be connected.Geodesic curve minimum spanning tree T is defined as the minimum curve on X, and it has length
Figure BDA0000451371200000082
Wherein,
Figure BDA0000451371200000087
be all subtree collections of curve G, e is an edge of tree T, g ebe the Euclidean distance that edge e is corresponding, its computing formula is shown in formula (2).
g e=||x i-x j||,x i,x j∈e (2)
In GMST estimates, some subsets
Figure BDA0000451371200000088
by all size, m forms, and the length L of the GMST of subset A (A) also needs to calculate.In theory,
Figure BDA0000451371200000083
be linear, thereby can be estimated by the function of this form of y=ax+b, by least square method, can estimate variable a and b.Can prove, by the estimated value of a and
Figure BDA0000451371200000084
can access the estimation of intrinsic dimension.The expression formula that is provided intrinsic dimension d by GMST algorithm is shown in formula (3).Intrinsic dimension d is another important input parameter of CM-SDE algorithm.
d = 1 1 - a - - - ( 3 )
Three, CM-SDE is a kind of semi-supervised manifold learning arithmetic, in the process realizing, needs all reference points to carry out class mark at CM-SDE algorithm.Consider that in current WiFi indoor positioning environment, reference point number is 1000 points nearly, artificially all reference points are not carried out to class mark, but adopt certain sorting algorithm to carry out mark to the classification of reference point.
The target of cluster is by data set X={x 1, x 2..., x nbe divided between c class and Various types of data uncorrelated mutually.Basic clustering algorithm is realized as follows:
(1) generate c cluster centre, be designated as v i, i=1,2 ..., c.
(2) by data set X={x 1, x 2..., x neach element sort out, adopt the attaching relation of arest neighbors (Nearest Neighbor) algorithm decision element, its equivalent expression is:
x i ∈ G i ⇔ arg min 1 ≤ j ≤ c D ( x i , v j ) - - - ( 4 )
In formula (4), x ibe i data point, G ibe the syntople figure that i class forms.D(x i, v j) expression calculating x iwith v jbetween Euclidean distance.
(3) renewal of cluster centre, is updated to for the cluster centre of i class:
v i = 1 | G i | Σ x k ∈ G i x k - - - ( 5 )
In formula (5), || represent to calculate the number of element in certain class.
(4) convergence verification and iteration
If meet one of convergence conditions in following four kinds of situations, iteration stopping, otherwise repeat (2)~(3), until iteration convergence or reach the maximum number of times of carrying out.Four kinds of convergence test conditions are:
Condition one: cluster centre is constant;
Condition two: the element of each cluster is constant;
Condition three: cluster centre variation converges in radius ε;
Condition four: cluster element variation converges in radius ε.
Generally speaking, above-mentioned convergence test condition can be expressed as following formula:
max||v i-v i′||≤ε,ε≥0 (6)
Wherein, v ithe cluster centre of the i class after ' expression is upgraded.
From the basic implementation analysis of above-mentioned clustering algorithm, for the Algorithm constitution of the attaching relation of decision element the core of cluster.Dissimilar clustering algorithm proposes different classification indexs, generally this function is called to the loss function (Loss Function).In basic clustering method, adopt Euclidean distance as the loss function.This patent adopts KFCM algorithm to carry out to Radio Map the classification mark that category analysis draws cluster centre and Radio Map.The theory analysis of KFCM algorithm is as follows:
The target of introducing the Fuzzy c-means Clustering of kernel function be by original data set place spatial alternation to infinite dimensional Hilbert space (Hilbert Space), more corresponding cluster analysis is done in the space after conversion.By the conversion of kernel function, the category feature between initial data is further easier to statement and distinguishes after conversion.The target function of the Fuzzy c-Means Clustering Algorithm based on kernel function is:
J KFCM = Σ i = 1 c Σ k = 1 n ( u ki ) m | | Φ ( x k ) - W i | | H 2 - - - ( 7 )
In formula (7), Φ (x k), W ibe illustrated respectively in data set and corresponding cluster centre under Hilbert space.By deriving, can show that the solution of KFCM algorithm is expressed as:
u ki = [ Σ j = 1 c ( | | Φ ( x k ) - W i | | H 2 | | Φ ( x k ) - W j | | H 2 ) 1 / m - 1 ] - 1 v j = Σ k = 1 n ( u ki ) m Φ ( x k ) Σ k = 1 n ( u ki ) m - - - ( 8 )
The key of the solution of KFCM is to calculate the loss function or the similarity function of Hilbert space.Consider to introduce in this article the FCM(Fuzzy C-Means of gaussian kernel function (Gaussian Kernel Function)) theory analysis and the realization thereof of algorithm.Gaussian kernel function is suc as formula shown in (9).
K ( x , y ) = e - λ | | x - y | | 2 - - - ( 9 )
In Hilbert space, by formula explain its corresponding loss function, this formula is further expressed as formula (10).
| | &Phi; ( x k ) - W i | | H 2 = < &Phi; ( x k ) - W i , &Phi; ( x k ) - W i > - - - ( 10 )
In formula (10), <, > represents to calculate the kernel function value of corresponding formula.And in fact, the conversion in infinite space does not exist, therefore, formula (10) is further reduced to shown in formula (11).
D ki = | | &Phi; ( x k ) - W i | | H 2 = < &Phi; ( x k ) , &Phi; ( x k ) > - 2 < &Phi; ( x k ) , W i > + < W i , W i > - - - ( 11 )
The Full-expasion of formula (11) is shown in formula (12).
D ki = < &Phi; ( x k ) , &Phi; ( x k ) > - 2 &Sigma; k = 1 n ( u ki ) m &Sigma; j = 1 n < &Phi; ( x k ) , &Phi; ( x j ) > + 1 ( &Sigma; k = 1 n ( u ki ) m ) 2 &Sigma; j = 1 n &Sigma; l = 1 n < &Phi; ( x l ) , &Phi; ( x j ) > - - - ( 12 )
In formula (12), < Φ (x k), Φ (x j) > calculates by gaussian kernel function, that is:
< &Phi; ( x k ) , &Phi; ( x j ) > = K ( x j , x k ) = e - &lambda; | | x j - x k | | 2 - - - ( 13 )
In algorithm is realized, not to generate at random cluster centre, but from data set X={x 1, x 2..., x nin random select c element as cluster centre, form and gather Y={y 1..., y c.Therefore, initialized loss function value calculate suc as formula shown in.
D ki o = | | &Phi; ( x k ) - &Phi; ( y i ) | | = K ( x k , x k ) - 2 K ( x k , y i ) + K ( y i , y i ) - - - ( 14 )
Four, using CM-SDE algorithm to realize carries out dimensionality reduction and obtains feature weight matrix process and realize by following step Radio Map:
CM-SDE algorithm is the maximized a kind of manifold learning arithmetic of divergence in class scatter based on flag data and Unlabeled data and class.Before CM-SDE algorithm is carried out to theory analysis, the given input data of CM-SDE algorithm are done to following explanation: input high dimensional data point
Figure BDA0000451371200000111
data point x iclass be labeled as y i∈ 1,2 ..., P}, wherein P represents high dimensional data to be divided into P submanifold, and the high dimensional data that is about to input is divided into P class, and the cluster centre of note P class is V={v 1, v 2..., v p.The high dimensional data of input is expressed as to the form of matrix: X=[x 1, x 2..., x m] ∈ R n * m.From the form of matrix notation, the row in matrix represent a high dimensional data point.
For the RadioMap that comprises Unlabeled data ul, all Unlabeled data X wherein u=[x u1, x u2..., x uk] ∈ R n * kcarry out classification coupling, flag data is designated as X simultaneously l=[x l1, x l2..., x lc] ∈ R n * c.For X uin all data carry out orderly classification coupling.Orderly implication is after distributing some Unlabeled datas, can affect the cluster centre of respective class, therefore can differentiate and have impact the class of next Unlabeled data.The time sequencing of the collection of main consideration signal in this patent.Suppose X uto arrange in chronological order.The ownership x of employing formula (15) compute classes u1, and adopt formula (5) to upgrade corresponding cluster centre.Then successively all Unlabeled datas are carried out to classification coupling
x ui &Element; G j &DoubleLeftRightArrow; arg min 1 &le; j &le; p D ( x ui , v j ) - - - ( 15 )
The target function of CM-SDE algorithm is:
min imize X T S w X X T S t X max imize X T S b X X T S t x - - - ( 16 )
S in formula (16) w, S b, S tin representation class, divergence, class scatter and total divergence can be calculated by formula (17) respectively:
S w = &Sigma; i = 1 P ( &Sigma; x k &Element; G i ( x k - m ( i ) ) ( x k - m ( i ) ) T ) S w = &Sigma; i = 1 P ( m - m ( i ) ) ( m - m ( i ) ) T S t = S w + S b - - - ( 17 )
In formula (17),
Figure BDA0000451371200000115
be the average of i class, l iit is the number of the sampled point of i class;
Figure BDA0000451371200000116
for the average of all sampled points, the number that N is sampled point.
For the target function shown in formula (16), can represent equally the local target function form that embeds manifold learning arithmetic of differentiating, its expression formula is:
MaximizeJ ( V ) = &Sigma; i , j | | V T x i V T x j | | 2 w ij &prime; subjectto &Sigma; i , j | | V T x i - V T x j | | 2 w ij = 1 - - - ( 18 )
In formula (18), w ijrepresent the weight allocation between homogeneous data, w ij' represent the weight allocation between inhomogeneity data, be expressed as W n * Nand W n * N'.Weight calculation process is completed by two steps.The first step: structure Neighborhood Graph.According to the class label information of high dimensional data point and neighbor relationships thereof, construct directionless figure G and G '.Wherein neighbor relationships is the criterion that adopts KNN algorithm to provide, and selects the nearest K of data point point as its neighbours, and G represents to work as x iwith x jclass label information y i=y jtime and x i, x jk nearest neighbor relation each other; G ' shows and works as x iwith x jclass label information y i≠ y jtime and x i, x jk nearest neighbor relation each other.Second step: calculate weight matrix.According to the adjacent map of first step structure, adopt class Gaussian function to carry out the calculating of weight matrix.Shown in its expression formula (19), (20) are.W in formula ijrepresent Neighbor Points x iwith x jbetween weights, || x i-x j|| 2for Neighbor Points. with x jbetween distance, adopt matrix-style to calculate distance, t is weights normalized parameter, U, L represent respectively the unmarked and number of the sampled point of mark.According to analyzing, can know W n * Nand W n * N' can be formed by three parts, respectively: flag data and the weight between flag data, the weight between flag data and Unlabeled data and the weight between Unlabeled data and Unlabeled data, be expressed as: W L &times; L = { w ij ll } , W L &times; U = { w ij lu } , W U &times; U = { w ij uu } .
W N &times; N = w ij ii = exp ( - | | x i - x j | | 2 / g ) ; if x i , x j &Element; G ; x i &Element; X l , x j &Element; X l 0 ; w ij lu = exp ( - | | x i - x j | | 2 / t ) ; if x i , x j &Element; G ; x i &Element; X l , x j &Element; X u - - - ( 19 ) 0 ; w ij uu = exp ( - | | x i - x j | | 2 / t ) ; if x i , x j &Element; G ; x i &Element; X u , x j &Element; X u 0 ;
W N &times; N &prime; = w ij &prime; ll = exp ( - | | x i - x j | | 2 / t ) ; if x i , x j &Element; G &prime; ; x i &Element; X l , x j &Element; X l 0 ; w ij &prime; lu = exp ( - | | x i - x j | | 2 / t ) ; if x i , x j &Element; G &prime; ; x i &Element; X l , x j &Element; X u - - - ( 20 ) 0 ; w ij &prime; uu = exp ( - | | x i - x j | | 2 / t ) lif x i , x j &Element; G &prime; ; x i &Element; X u , x j &Element; X u 0 ;
Character by above-mentioned computing formula and matrix can obtain:
Figure BDA0000451371200000131
Figure BDA0000451371200000132
can derive W thus n * Nand W n * N' be expressed as the form of matrix in block form, shown in (21).
W N &times; N = W L &times; L W L &times; U W L &times; U T W U &times; U W N &times; N &prime; = W L &times; L &prime; W L &times; U &prime; W L &times; U &prime; T W U &times; U &prime; - - - ( 21 )
According to the calculating formula of matrix:
Figure BDA0000451371200000134
calculating formula is expressed as the computational methods of the matrix of matrix A, and the method that calculating formula provides is consistent with the calculating formula of matrix trace, that is: || and A|| 2=tr (AA t).Formula (18) can be expressed as the account form of matrix trace thus:
J ( V ) = &Sigma; i , j { tr [ ( V T x i - V T x j ) ( V T x i - V T x j ) T w ij &prime; } } - - - ( 22 )
Formula (22) can be reduced to:
J ( V ) = &Sigma; i , j { tr [ V T ( x i - x j ) ( x i - x j ) T V ] w ij &prime; } - - - ( 23 )
Scalar nature and weights element by the calculating of trace of a matrix are real number, formula (23) can be reduced to:
J ( V ) = tr { V T &Sigma; i , j [ ( x i - x j ) w ij &prime; ( x i T - x j T ) ] V } - - - ( 24 )
According to simple mathematical relationship, formula (24) can be reduced to:
J(V)=2tr{V T[X(D′-W′ N×N)X T]V} (25)
In formula (25): X is input data, and λ and v are eigen vector, and W and W ' are respectively the weight matrix of G and G ' correspondence, and D and D ' are diagonal matrix, and its diagonal element can be represented by formula (26).
d ii = &Sigma; j w ij d ii &prime; = &Sigma; j w ij &prime; - - - ( 26 )
According to the derivation mode of formula (25), in like manner the constraints in formula (18) can be write as suc as formula (25) similar form, thus, (18) can be expressed as to form:
MaximizeJ ( V ) = 2 tr { V T [ X ( D &prime; - W N &times; N &prime; ) X T ] V } subjectto 2 tr [ X ( D - W N &times; N ) X T ] = 1 - - - ( 27 )
To formula (27) application Lagrange (Lagrange) Multiplier Method, can draw shown in formula (28):
X(D′-W′ N×N)X Tv=λX(D-W N×N)X Tv (28)
Formula (28) is carried out to generalized eigenvalue decomposition, draw characteristic value and the characteristic vector of its Eigenvalues Decomposition, be expressed as: λ=[λ 1, λ 2..., λ n] t, its characteristic of correspondence vector is: v=[v 1, v 2..., v n] t.Get front d maximum characteristic value characteristic of correspondence vector and form transformation matrix V=[v 1, v 2..., v d].Output data conversion method by CM-SDE algorithm can draw, after dimensionality reduction, data are:
z i=V Tx i (29)
In formula (29), z irepresent input high dimensional data point x ilow-dimensional output data after conversion.The implementation step of the off-line phase that summary of the invention provides from this patent is: the first step first adopts CM-SDE algorithm to carry out dimension-reduction treatment to all reference point Radio Map, obtain the Radio Map after the dimensionality reduction of corresponding reference point, as the coupling location database (RadioMap of on-line stage *).Second step is to adding the Radio Map of Unlabeled data, i.e. RadioMap ulcarry out dimension-reduction treatment, obtain eigentransformation matrix V '.Can set up thus the required database of off-line phase: RadioMap *and V '.
Five, the RSS being obtained by different user's on-line stage positioning stages is unmarked category attribute, and it adds Radio Map, and forms Radio Map ulprocess be called classification.Its implementation is as described below:
The unmarked RSS that the test of different user on-line stage positioning stage is obtained, adopts the mode of classification coupling to be added in existing Radio Map, obtains comprising accordingly unmarked Radio map RadioMap ul; By classification matching process, increase the data volume of Radio Map, and then improve the density of Radio Map, for CM-SDE algorithm provides new dimensionality reduction data, can upgrade cluster centre, for CM-SDE algorithm provides new categorical data simultaneously.Classification matching process is divided into two steps, and its implementation procedure is as described below:
The first step, finds the category attribute of unmarked RSS.Remember that one group of unmarked RSS is RSS i, mate with the cluster centre in step 3, by formula (4), complete RSS iclassification mark.
Second step: RSS is carried out to Threshold detection.For cluster centre, vi is expressed as v i=(v i1, v i2..., v iN), N is the number of AP in indoor locating system.RSS ibe expressed as RSS i=(RSS i1, RSS i2..., RSS iN).Calculate the defined generalized symbol value of following formula:
S j = &Sigma; j = 1 N sgn ( RSS ij - v ij ) - - - ( 30 )
Wherein, sgn () is defined as:
sgn ( x ) = 1 , x &GreaterEqual; 0 0 , x < 0 - - - ( 31 )
Work as S ibe greater than while setting threshold value, by RSS iadd in Radio Map, and upgrade cluster centre, otherwise give up RSS i, do not add in Radio Map.In this patent, threshold value V t=0.9N.The renewal of cluster centre is completed by formula (5):
Six, the offline database implementation of the WiFi indoor orientation method based on CM-SDE algorithm:
Offline database mode consists of three parts.The first, the foundation of the Radio Map of all reference points, and adopt CM-SDE algorithm to obtain RadioMap *.The second, then stochastical sampling U point Unlabeled data being added in original Radio Map, and obtain V ' with CM-SDE algorithm, and the location database forming is downloaded to (storage) to the mobile terminal of locating.The 3rd, realize online location and Radio Map upgrades.Described in being implemented as follows of third part:
On-line stage, the RSS=[AP that test point place receives 1, AP 2..., AP n], n represents the number of the AP that indoor locating system is arranged.By RSS and eigentransformation matrix V ' multiply each other, thereby draw the RSS '=[AP after dimensionality reduction 1, AP 2..., AP d], wherein d represents intrinsic dimension.Adopt again KNN algorithm to realize RSS ' and RadioMap *coupling.Adopt the mean value of coordinate of K the reference point nearest with RSS ' as test point (x ', y '), its expression formula is:
( x &prime; , y &prime; ) = 1 K &Sigma; i = 1 K ( x i , y i ) - - - ( 32 )
The RSS that this location of consumer positioning is recorded adopts classification matching way to join in Radio Map as Unlabeled data, and the renewal of realization to local On-line matching location database on mobile terminal, realize the dynamic local data of upgrading, thereby realize offline database locate mode, the WiFi indoor orientation method of the semi-supervised manifold learning of accompanying drawing 1 classification that is shown in coupling is realized at user's locating terminal;
Seven, the online database implementation of the WiFi indoor orientation method based on CM-SDE algorithm:
Online database mode consists of four parts.The first, the foundation of the Radio Map of all reference points, and adopt CM-SDE algorithm to obtain RadioMap *.The second, then stochastical sampling U point Unlabeled data being added in original Radio Map, and obtain V ' with CM-SDE algorithm, and the location database forming is downloaded to (storage) to the mobile terminal of locating.The 3rd, realize online location and Radio Map upgrades.Described in being implemented as follows of third part:
On-line stage, the RSS=[AP that test point place receives 1, AP 2..., AP n], n represents the number of the AP that indoor locating system is arranged.By RSS and eigentransformation matrix V ' multiply each other, thereby draw the RSS '=[AP after dimensionality reduction 1, AP 2..., AP d], wherein d represents intrinsic dimension.Adopt again KNN algorithm to realize RSS ' and RadioMap *coupling.Adopt the mean value of coordinate of K the reference point nearest with RSS ' as test point (x ', y '), its expression formula is:
( x &prime; , y &prime; ) = 1 K &Sigma; i = 1 K ( x i , y i ) - - - ( 33 )
The 4th part: after user has located online, the RSS value that this records online by user is uploaded to the server at online location database place, and at server end, online location database is upgraded to just online locator data and pass the locating terminal of uploading RSS data back, be that the off-line phase shown in accompanying drawing 2 completes on the server of online location database place, and on-line stage complete at locating terminal.

Claims (6)

1. a WiFi indoor orientation method for the semi-supervised manifold learning mating based on classification, is characterized in that the WiFi indoor orientation method off-line phase position fixing process of the semi-supervised manifold learning based on classification coupling is realized according to the following steps:
One, room area to be positioned is arranged to AP, make wireless signal cover room area to be positioned, complete WiFi network struction;
In room area rule to be positioned, choose and record the corresponding coordinate of reference point, measure and record successively the RSS signal of all AP that reference point receives as position feature information, build Radio Map, and store Radio Map;
Two, adopt GMST intrinsic dimension algorithm for estimating to analyze the intrinsic dimension of the Radio Map building in step 1, the intrinsic dimension obtaining, as one of input parameter of CM-SDE algorithm, determines the dimension after Radio Map dimensionality reduction;
Three, adopt KFCM algorithm to carry out cluster analysis to Radio Map, realize the classification mark of the Radio Map setting up, and as one of input parameter of CM-SDE, and corresponding initial cluster center and classification mark are provided;
Four, the intrinsic dimension in step 2 and the classification mark in step 3, as input parameter, adopt CM-SDE algorithm to the Radio Map dimensionality reduction building in step 1, draw the RadioMap after corresponding dimensionality reduction *, RadioMap *as being used for online positioning stage in coupling location database;
Five, the unmarked RSS online positioning stage test of different user being obtained, adopts the mode of classification coupling to be added in existing Radio Map, obtains comprising accordingly unmarked Radio map RadioMap ul, the cluster centre that mates renewal by classification is as classification input parameter new in CM-SDE algorithm;
Six, the cluster centre of the renewal in step 5, as input parameter, adopts CM-SDE to RadioMap uldimensionality reduction obtain eigentransformation matrix V ', V ' and RadioMap *common formation On-line matching location database, locates for on-line stage; Wherein, described line phase orientation is specially:
Six (one), on-line testing RSS;
Six (two), adopting V is RSS by RSS dimensionality reduction *;
Six (three), adopt KNN algorithm to mate location output positioning result;
Six (four), user's locating terminal location database is upgraded;
Completed a kind of off-line phase implementation of WiFi indoor orientation method of the semi-supervised manifold learning based on classification coupling.
2. the WiFi indoor orientation method of a kind of semi-supervised manifold learning based on classification coupling according to claim 1, it is characterized in that adopting GMST intrinsic dimension algorithm for estimating to analyze the intrinsic dimension of the Radio Map building in step 1 in step 2, its computing formula is:
Figure FDA0000451371190000011
geodesic distance minimal spanning tree algorithm
In above formula
Figure FDA0000451371190000021
in a represent the slope of the linear fit expression formula y=ax+b of minimum spanning tree.
3. the WiFi indoor orientation method of a kind of semi-supervised manifold learning based on classification coupling according to claim 1, is characterized in that generating in step 4 the Radio map RadioMap after new dimensionality reduction *with the V ' expression formula in step 6 be:
Radio Map*=V′·X
X is the Radio Map that needs dimensionality reduction.
4. the WiFi indoor orientation method of a kind of semi-supervised manifold learning based on classification coupling according to claim 1, is characterized in that the realization of classification coupling in step 5, and its core process completes in two steps:
The first step, finds the category attribute of unmarked RSS, by following formula, completes category attribute mark:
x i &Element; G i &DoubleLeftRightArrow; arg min 1 &le; j &le; c D ( x i , v j )
Second step: RSS is carried out to Threshold detection: by calculating and judge the relation of generalized symbol value and threshold value, thereby realize the renewal of Radio Map and classification flag data, the renewal of generalized symbol value and cluster centre respectively lower two formulas completes:
S j = &Sigma; j = 1 N sgn ( RSS ij - v ij )
v i = 1 | G i | &Sigma; x k &Element; G i x k
Wherein, described threshold value V t=0.9N.
5. a WiFi indoor orientation method for the semi-supervised manifold learning mating based on classification, is characterized in that the WiFi indoor orientation method on-line stage position fixing process of the semi-supervised manifold learning based on classification coupling is realized by following step:
One, on-line testing RSS;
Two, the RSS in the site undetermined obtaining is adopted the conversion of eigentransformation matrix dimensionality reduction obtain RSS*;
Three, adopt KNN algorithm to RSS* and Radio Map* match bit, treat the particular location coordinate of anchor point and predict and carry out the renewal of online data, its implementation procedure is:
(1) on-line stage, the RSS=[AP that test point place receives 1, AP 2..., AP n], with eigentransformation matrix V ' multiply each other, thereby draw the RSS '=[AP after dimensionality reduction 1, AP 2... AP d], wherein d represents intrinsic dimension;
(2) adopt KNN algorithm to realize mating of RSS ' and Radio Map*, adopt and the mean value of the coordinate of K the reference point that RSS ' is nearest as test point (x ', y '), its expression formula is:
( x &prime; , y &prime; ) = 1 K &Sigma; i = 1 K ( x i , y i )
In formula, (x ', y ') is the coordinate of test point prediction, (x i, y i) be the coordinate of i Neighbor Points, K is the number of neighbour in KNN algorithm;
Four, user's locating terminal location database is upgraded, and has completed the WiFi on-line stage indoor orientation method of the semi-supervised manifold learning based on classification coupling.
6. the WiFi indoor orientation method of a kind of semi-supervised manifold learning based on classification coupling according to claim 5, the particular location coordinate that it is characterized in that treating in step 3 anchor point is predicted and the renewal of carrying out online data is realized by the offline database described in claim 1 and online database:
The implementation of off-line location database is: the RSS that this location of consumer positioning is recorded adopts classification matching way to join in Radio Map as Unlabeled data, and the renewal of realization to local On-line matching location database on mobile terminal, realize the local data of dynamic renewal, thereby realize offline database locate mode;
The implementation of online database is: after user has located online, the RSS value that this records online by user is uploaded to the server at online location database place, and at server end, online location database is upgraded to just online locator data and pass the locating terminal of uploading RSS data back.
CN201310750528.6A 2013-12-31 2013-12-31 WiFi indoor positioning method of semi-supervised manifold learning based on category matching Active CN103648106B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310750528.6A CN103648106B (en) 2013-12-31 2013-12-31 WiFi indoor positioning method of semi-supervised manifold learning based on category matching

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310750528.6A CN103648106B (en) 2013-12-31 2013-12-31 WiFi indoor positioning method of semi-supervised manifold learning based on category matching

Publications (2)

Publication Number Publication Date
CN103648106A true CN103648106A (en) 2014-03-19
CN103648106B CN103648106B (en) 2017-03-22

Family

ID=50253244

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310750528.6A Active CN103648106B (en) 2013-12-31 2013-12-31 WiFi indoor positioning method of semi-supervised manifold learning based on category matching

Country Status (1)

Country Link
CN (1) CN103648106B (en)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103906234A (en) * 2014-04-03 2014-07-02 李晨 Indoor positioning method based on WIFI signals
CN104185275A (en) * 2014-09-10 2014-12-03 北京航空航天大学 Indoor positioning method based on WLAN
CN104469932A (en) * 2014-11-21 2015-03-25 北京拓明科技有限公司 Position fingerprint positioning method based on support vector machine
CN104507097A (en) * 2014-12-19 2015-04-08 上海交通大学 Semi-supervised training method based on WiFi (wireless fidelity) position fingerprints
CN104540221A (en) * 2015-01-15 2015-04-22 哈尔滨工业大学 WLAN indoor positioning method based on semi-supervised SDE algorithm
CN104581945A (en) * 2015-02-06 2015-04-29 哈尔滨工业大学 WLAN indoor positioning method for distance constraint based semi-supervised APC clustering algorithm
CN105517143A (en) * 2014-10-17 2016-04-20 深圳航天科技创新研究院 Method of reducing WLAN indoor positioning search dimension
CN105657823A (en) * 2015-12-16 2016-06-08 吉林大学 WIFI indoor weighted K nearest neighbor positioning algorithm based on kernel function main feature extraction
TWI554136B (en) * 2014-09-24 2016-10-11 緯創資通股份有限公司 Methods for indoor positioning and apparatuses using the same
CN106028446A (en) * 2016-07-15 2016-10-12 西华大学 Indoor parking lot location method
CN107277773A (en) * 2017-07-10 2017-10-20 广东工业大学 Combine the adaptive location method of a variety of contextual models
CN107862757A (en) * 2017-11-03 2018-03-30 广东广凌信息科技股份有限公司 A kind of movable attendance checking method and system based on Wi Fi fingerprints
WO2018094502A1 (en) * 2016-11-22 2018-05-31 Aerial Technologies Device-free localization methods within smart indoor environments
CN108519578A (en) * 2018-03-23 2018-09-11 天津大学 A kind of indoor positioning fingerprint base construction method based on intelligent perception
CN108600002A (en) * 2018-04-17 2018-09-28 浙江工业大学 A kind of mobile edge calculations shunting decision-making technique based on semi-supervised learning

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11818629B2 (en) 2016-11-22 2023-11-14 Aerial Technologies Device-free localization methods within smart indoor environments

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110164522A1 (en) * 2006-05-08 2011-07-07 Skyhook Wireless, Inc. Estimation of Position Using WLAN Access Point Radio Propagation Characteristics in a WLAN Positioning System
CN103079269A (en) * 2013-01-25 2013-05-01 哈尔滨工业大学 LDE (Linear Discriminant Analysis) algorithm-based WiFi (Wireless Fidelity) indoor locating method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110164522A1 (en) * 2006-05-08 2011-07-07 Skyhook Wireless, Inc. Estimation of Position Using WLAN Access Point Radio Propagation Characteristics in a WLAN Positioning System
CN103079269A (en) * 2013-01-25 2013-05-01 哈尔滨工业大学 LDE (Linear Discriminant Analysis) algorithm-based WiFi (Wireless Fidelity) indoor locating method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
邓志安: "基于学习算法的WLAN室内定位技术研究", 《哈尔滨工业大学博士学位论文》, 25 December 2012 (2012-12-25) *

Cited By (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103906234A (en) * 2014-04-03 2014-07-02 李晨 Indoor positioning method based on WIFI signals
CN103906234B (en) * 2014-04-03 2019-04-26 李晨 A kind of indoor orientation method based on WIFI signal
CN104185275A (en) * 2014-09-10 2014-12-03 北京航空航天大学 Indoor positioning method based on WLAN
CN104185275B (en) * 2014-09-10 2017-11-17 北京航空航天大学 A kind of indoor orientation method based on WLAN
TWI554136B (en) * 2014-09-24 2016-10-11 緯創資通股份有限公司 Methods for indoor positioning and apparatuses using the same
CN105517143A (en) * 2014-10-17 2016-04-20 深圳航天科技创新研究院 Method of reducing WLAN indoor positioning search dimension
CN105517143B (en) * 2014-10-17 2019-01-11 深圳航天科技创新研究院 A method of it reducing WLAN indoor positioning and searches for dimension
CN104469932A (en) * 2014-11-21 2015-03-25 北京拓明科技有限公司 Position fingerprint positioning method based on support vector machine
CN104469932B (en) * 2014-11-21 2018-07-06 北京拓明科技有限公司 A kind of location fingerprint localization method based on support vector machines
CN104507097A (en) * 2014-12-19 2015-04-08 上海交通大学 Semi-supervised training method based on WiFi (wireless fidelity) position fingerprints
CN104540221A (en) * 2015-01-15 2015-04-22 哈尔滨工业大学 WLAN indoor positioning method based on semi-supervised SDE algorithm
CN104540221B (en) * 2015-01-15 2018-06-22 哈尔滨工业大学 WLAN indoor orientation methods based on semi-supervised SDE algorithms
CN104581945A (en) * 2015-02-06 2015-04-29 哈尔滨工业大学 WLAN indoor positioning method for distance constraint based semi-supervised APC clustering algorithm
CN104581945B (en) * 2015-02-06 2018-09-07 哈尔滨工业大学 The WLAN indoor orientation methods of semi-supervised APC clustering algorithms based on distance restraint
CN105657823B (en) * 2015-12-16 2020-07-14 吉林大学 WIFI indoor weighted K nearest neighbor positioning algorithm based on kernel function main feature extraction
CN105657823A (en) * 2015-12-16 2016-06-08 吉林大学 WIFI indoor weighted K nearest neighbor positioning algorithm based on kernel function main feature extraction
CN106028446A (en) * 2016-07-15 2016-10-12 西华大学 Indoor parking lot location method
CN106028446B (en) * 2016-07-15 2019-04-02 西华大学 Parking garage localization method
WO2018094502A1 (en) * 2016-11-22 2018-05-31 Aerial Technologies Device-free localization methods within smart indoor environments
CN107277773A (en) * 2017-07-10 2017-10-20 广东工业大学 Combine the adaptive location method of a variety of contextual models
CN107277773B (en) * 2017-07-10 2020-04-17 广东工业大学 Adaptive positioning method combining multiple contextual models
CN107862757A (en) * 2017-11-03 2018-03-30 广东广凌信息科技股份有限公司 A kind of movable attendance checking method and system based on Wi Fi fingerprints
CN108519578A (en) * 2018-03-23 2018-09-11 天津大学 A kind of indoor positioning fingerprint base construction method based on intelligent perception
CN108600002A (en) * 2018-04-17 2018-09-28 浙江工业大学 A kind of mobile edge calculations shunting decision-making technique based on semi-supervised learning

Also Published As

Publication number Publication date
CN103648106B (en) 2017-03-22

Similar Documents

Publication Publication Date Title
CN103648106A (en) WiFi indoor positioning method of semi-supervised manifold learning based on category matching
Huckemann Intrinsic inference on the mean geodesic of planar shapes and tree discrimination by leaf growth
CN103079269A (en) LDE (Linear Discriminant Analysis) algorithm-based WiFi (Wireless Fidelity) indoor locating method
CN104540221A (en) WLAN indoor positioning method based on semi-supervised SDE algorithm
Wu et al. Spatio-temporal outlier detection in precipitation data
CN103440512A (en) Identifying method of brain cognitive states based on tensor locality preserving projection
Hombal et al. Bias minimization in Gaussian process surrogate modeling for uncertainty quantification
CN102930533A (en) Semi-supervised hyperspectral image dimension reduction method based on improved K-means clustering
Lai et al. Point Cloud Discretization of Fokker--Planck Operators for Committor Functions
Wang et al. A polygon-based clustering and analysis framework for mining spatial datasets
Ermon et al. Smt-aided combinatorial materials discovery
CN104835174A (en) Robustness model fitting method based on supermap mode search
CN109034179B (en) Rock stratum classification method based on Mahalanobis distance IDTW
CN111090679B (en) Time sequence data representation learning method based on time sequence influence and graph embedding
CN105704676A (en) Method for improving fingerprint indoor positioning precision through employing signal time correlation
CN109889981B (en) Positioning method and system based on binary classification technology
Petelin et al. Tla: Topological landscape analysis for single-objective continuous optimization problem instances
Pulkkinen Ridge-based method for finding curvilinear structures from noisy data
Peng et al. Uck-means: A customized k-means for clustering uncertain measurement data
CN108462940A (en) A kind of indoor orientation method based on Wi-Fi hotspot
Yu et al. Cluster-based best match scanning for large-scale missing data imputation
CN103679201B (en) Calibration method of point set matching for image matching, recognition and retrieval
Hui et al. Power equipment segmentation of 3D point clouds based on geodesic distance with K-means clustering
Saini et al. Locally Interpretable Model Agnostic Explanations using Gaussian Processes
Hastari et al. Students trajectory pattern finding scheme based on RSSI geolocation as a part of smart campus

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20190612

Address after: 150000 Heilongjiang Harbin Dalian economic and Trade Zone, the North Road and Xingkai Road intersection

Patentee after: Harbin University of Technology Robot Group Co., Ltd.

Address before: 150001 No. 92 West straight street, Nangang District, Heilongjiang, Harbin

Patentee before: Harbin Institute of Technology