KR20120048376A - Knn/pfcm hybrid mehod using gath-geva method for indoor location determination in waln - Google Patents

Abstract

PURPOSE: A wireless indoor location measuring method based on a PFCM group using KNN and GG method is provided to estimate a location of a terminal by using an existing AP(Access Point). CONSTITUTION: The first electric wave feature data is stored in a database with a wireless fingerprint method(S110). K reference points are selected by using a KNN method(S120). The k reference points have a first electric wave feature value near a second electric wave feature value. A group of the first electric wave feature data of the k reference points is formed by using a GG method(S130). A group of a minimum distance between a group central vector and the first electric wave feature data is selected(S140). An average of a location coordinate of the reference points belong to the group is produced. A location of a terminal is estimated(S150).

Description

Hybrid wireless indoor positioning method based on PFCC clustering using KEN and KN method {KNN / PFCM HYBRID MEHOD USING GATH-GEVA METHOD FOR INDOOR LOCATION DETERMINATION IN WALN}

The present invention relates to a hybrid wireless indoor positioning method based on PFCM clustering using KNN and GG methods in WLAN. More specifically, the present invention relates to a wireless fingerprint database based on signal noise ratio (SNR) received from an AP. The present invention relates to a hybrid wireless indoor positioning method based on PFCM (Possibilistic Fuzzy C-Means) clustering using k-Nearest Neighbor (KNN) and GG (GATH-GEVA) methods.

In the location-based service, positioning technology is the core / based technology, but GPS is the center of today, but in the shadow area such as indoors, a new type of positioning technology is required, and positioning technology based on wireless communication infrastructure has been studied.

CDMA-based wireless positioning technology in the outdoors, wireless positioning technology using infrastructure such as WLAN, UWB, infrared, ultrasonic, RFID, etc. have been studied and partially commercialized.

Recently, researches on indoor positioning of wireless LAN have been actively conducted in many countries such as the US and Europe. The PlaceLab project, which is typically under way at Intel, basically measures the location based on the WLAN signal value.

In addition, the company is researching more accurate location tracking method by building up the latest network infrastructure such as GSM, Bluetooth, RFID, etc.

The ASK-IT project for European traffic weakness is studying indoor location measurement using the MOTE sensor network, which is a modification of the wireless LAN network when providing home services and indoor guidance.

In addition, research on the propagation model of the radio signal and the RADAR system using the Nearest Neighbor technique that selects the nearest point to the calculated point are actively conducted.

On the other hand, the commercialized and marketed systems include Ekahau Real-Time Location System (RTLS) and AeroScout Visibility System.

Specific positioning methods presented in the above-described studies include triangulation methods, which are also used in GPS, and wireless fingerprinting, which measures the location in units of cells by dividing the area into a grid like a cellular network.

In this case, the wireless fingerprint method is a location estimation method using probabilistic modeling, and uses noise and surrounding environment information as information for location tracking.

In addition, an algorithm of a wireless fingerprint method for determining a location using a wireless LAN is implemented by a K-NN method, a Bayesian method, a neural network, a decision tree method, and the like.

The Fuzzy C-Means (FCM) clustering method is proposed by Bezdek as a data classification algorithm that classifies each data belonging to one cluster according to the degree of belonging to the cluster.

FCM has been successfully applied in many fields, but FCM sometimes yields results that are different from intuitive partitioning by allowing the sum of one point to belong to all clusters.

These constraints also make FCM noise sensitive. In order to solve this problem, Possibilistic C-Mean (PCM) has been proposed that removes the limitation of FCM.

PCM eliminates noise susceptibility by preventing noise or outliers from belonging to any cluster using the probability theory rather than the traditional probabilistic method, but because each cluster does not affect each other, it is sensitive to initialization and sometimes duplicated. There was a problem finding the cluster.

In order to overcome the noise sensitivity of the FCM and the overlap cluster problem of the PCM, a method of using both a probability measure or affiliation measure and likelihood measure or typicality needs to be devised.

In order to solve the problems of the prior art described above, the wireless fingerprint by measuring the signal to noise ratio (SNR) received from a plurality of AP (Access Point) in the training step for a pattern matching algorithm in a wireless environment The purpose of this study is to provide a hybrid wireless indoor positioning method based on PFCM clustering using KNN and GG method that creates and utilizes a (fingerprint) database.

Another object of the present invention is to provide a mixed wireless indoor positioning method based on PFCM clustering using KNN and GG methods for signal noise ratio data using an existing AP without using special equipment for positioning.

The object of the present invention is not limited to the above-mentioned object, and other objects that are not mentioned will be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, a mixed wireless indoor positioning method based on PFCM clustering using the KNN and GG methods according to an aspect of the present invention includes: a first collected at a plurality of reference points (RP) in a wireless environment; And storing the propagation characteristic value data in a database by a wireless fingerprint method and having the second propagation characteristic value data received from a plurality of access points at a test point (TP) where a terminal is located. Searching for using Possibilistic Fuzzy C-Mean (PFCM) mixing technique using k-Nearest Neighbor (KNN) / GG method.

The retrieval of the first propagation value corresponding to the second propagation characteristic value is for estimating the position of the terminal having the second propagation characteristic value. To this end, the present invention provides a first propagation value close to the received second propagation characteristic value. Reference points having propagation characteristic values are selected by the KNN technique, and first propagation characteristic value data of the selected k reference points are clustered by the PFCM technique using the GG method.

The cluster having the smallest GK distance to the cluster center vector and the first propagation characteristic value data is selected from among the plurality of clustered clusters, and an arithmetic mean of the position coordinates of the reference points belonging to the selected cluster is calculated as the position of the terminal.

According to the present invention, in a training step for a pattern matching algorithm in a wireless environment, a propagation characteristic value (for example, a signal noise ratio (SNR)) received from a plurality of APs is measured, which is a database of a wireless fingerprint scheme. Therefore, we propose a hybrid wireless indoor positioning method based on the PFCM clustering method using the KNN method and the GG method that can estimate the location of the terminal using the existing AP without using any special equipment for positioning.

According to the conventional KNN method, since the computational amount increases in a wide and complex indoor environment, it is difficult to find the optimal position of the terminal. In the present invention, to solve this problem, the PFCM mixing algorithm using the KNN / GG method is applied.

According to the present invention to which the PFCM mixing algorithm using the KNN / GG method is applied, the database search efficiency can be increased and the accuracy of the search can be improved.

1 is a flowchart illustrating a mixed wireless indoor positioning method based on PFCM clustering using a KNN and a GG method according to an embodiment of the present invention.
2 is an exemplary view showing an example of a place where a mixed wireless indoor positioning method based on PFCM clustering using the KNN and GG methods according to the present invention is executed.

Advantages and features of the present invention and methods for achieving them will be apparent with reference to the embodiments described below in detail with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but will be implemented in various forms, and only the present embodiments are intended to complete the disclosure of the present invention, and the general knowledge in the art to which the present invention pertains. It is provided to fully convey the scope of the invention to those skilled in the art, and the present invention is defined only by the scope of the claims. It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. In the present specification, the singular form includes plural forms unless otherwise specified in the specification.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. First of all, in adding reference numerals to the elements of each drawing, the same elements are denoted by the same reference numerals as much as possible even though they are shown in other drawings, and in describing the present invention, If the description can obscure the subject matter of the present invention, the detailed description is omitted.

1 and 2, a PFCM clustering-based mixed wireless indoor positioning method using a KNN and a GG method according to an embodiment of the present invention will be described. 1 is a flowchart illustrating a mixed wireless indoor positioning method based on PFCM clustering using the KNN and GG methods according to an embodiment of the present invention, and FIG. 2 is a flowchart illustrating a KNN and GG method according to an embodiment of the present invention. It is an exemplary diagram showing an example of a place where the PFCM clustering-based mixed wireless indoor positioning method is executed.

Wireless fingerprint (Fingerprint) method according to an embodiment of the present invention is a method using a WLAN, the position can be calculated by measuring the signal transmission distance due to signal attenuation by measuring the RF signal strength received by the terminal. .

This is a location estimation method using probabilistic modeling, and uses noise and surrounding information as information for location tracking.

The system for implementing a wireless indoor positioning method according to the present invention comprises, for example, a server that collects indoor WLAN signals, configures a location model (database), and stores them, and a client (or terminal). Can be.

Here, when a client having a WLAN interface function such as a PDA, a PC, a handset, etc. requests a location, the server calculates a location based on the stored location model and returns it to the client.

Hereinafter, a hybrid wireless indoor positioning method based on PFCM clustering using a KNN and a GK method performed by a server in order to return the estimated location of the terminal at the request of the terminal will be described.

In order to perform wireless indoor positioning according to the present invention, first, the first propagation characteristic value data collected at each reference point (RP) in a wireless environment is stored in a database by a wireless fingerprinting method (S110). ).

To this end, as shown in FIG. 2, the space to be positioned is divided into a predetermined range, and the position value of each reference point (RP) is stored in a server database.

Subsequently, the first propagation characteristic value (for example, signal noise ratio: SNR) extracted from the signal strength measurement value reaching the mobile unit (MU) from the access point (AP) together with the position value of each RP is determined by the server. It is stored in the database.

This process is performed repeatedly until the measurement of all the RP of the space to be measured is completed.

In FIG. 2, 63 RP and 40 TP (Test Point) checker boards were designated at 1m intervals in a corridor having a total length of 25m × 4m. Here, the RP is black and the TP points for evaluating the performance of the present invention are displayed in red, and in each TP, the terminal receives signals from two APs.

Next, the server transmits a plurality of second propagation characteristic values (for example, signal noise ratios received from the plurality of APs at a location for transmitting the location request signal) received together with the location request signal of the terminal to the database in step S110. The location value of the terminal having transmitted the location request signal and the most suitable location data information are determined by comparing with the stored first propagation characteristic value, and the location value is provided to the terminal.

To this end, first, the server selects k first propagation characteristic value data close to a second propagation characteristic value (hereinafter, referred to as a signal noise ratio: SNR) received from a plurality of APs using the KNN technique (S120).

Here, the K-Nearest neighbor (KNN) technique is one of memory-based reasoning techniques and is widely used for pattern recognition.

According to the KNN technique, k first propagation characteristic values close to the second propagation characteristic values may be selected from the server database based on the similarity function.

Assuming the numbers of RP and AP are m and n, respectively, the similarity between the received second propagation characteristic value and the first propagation characteristic value data set can be obtained using the distance calculated in Equation 1.

[Equation 1]

는 유사도를 구하기 위하여 산출되는 거리,

는 S110 단계에서 저장된 i번째 RP에서 j번째 AP에 대한 SNR 데이터,

는 복수의 AP 중,

j번째 AP로부터 수신된 SNR 데이터이다.here,

Is the distance computed to find the similarity,

SNR data for the j th AP in the i th RP stored in step S110,

Of the plurality of APs,

SNR data received from the j th AP.

Such KNN technique results in various patterns according to the parameters k and q of Equation 1. The parameter k is the number of RPs for the coordinates to be estimated, and the parameter q means the distance type.

In general, when parameter q is 1, it means Manhattan distance, and when parameter q is 2, it means Euclidean distance. Considering the complexity of the KNN technique and the location prediction, the parameters q generally use 1 and 2.

The k first propagation characteristic value data selected in step S120 are clustered by PFCM (Possibilistic Fuzzy C-Means clustering) technique using the GK method (S130).

Hereinafter, in order to describe the PFCM technique using the GK method in more detail, the FCM technique and the PCM technique will be described first.

The FCM scheme is proposed by Bezdek as a data classification algorithm that classifies each data in one cluster according to the degree of belonging of each data in the cluster.

개의 벡터의 집합을

개의 퍼지 그룹으로 분할하고, 각각의 그룹 안에서 클러스터의 중심(예를 들어, 소속 정도 측정 비용함수가 최소가 되는 지점)을 찾는다.FCM technique

Set of vector dogs

Split into four fuzzy groups and find the center of the cluster within each group (for example, the point at which the membership measurement cost function is minimum).

The FCM technique uses fuzzy partitioning that a given data point can belong to several groups by using a degree of membership (hereinafter, referred to as a membership value) that has a value between 0 and 1 depending on the degree of belonging.

That is, in order to adapt the fuzzy partitioning, the membership function U has elements with values between 0 and 1, and the sum of the membership values for the data set is always 1.

는 각 퍼지 부분집합 즉, 각 클러스터에 소속되는 정도를 가지며, 목적 함수는 수학식 2와 같이 주어진다.Each SNR data point in the FCM

Has a degree belonging to each fuzzy subset, that is, each cluster, and the objective function is given by Equation 2.

[Equation 2]

는 k번째 SNR 데이터 포인트와 i번째 클러스터 중심 사이의 거리 제곱을 나타내며, 수학식 1의 마할라노비스 거리가 그 한 예에 해당한다.In this case, m is a constant representing the degree of fuzziness and is generally set to 2.

Denotes the square of the distance between the k-th SNR data point and the i-th cluster center, and the Mahalanobis distance of Equation 1 corresponds to one example.

Using Lagrange's method, U and are obtained as Equations 3 and 4, respectively, and a local maximum is generally obtained through AO (Alternating Optimization).

&Quot; (3) "

&Quot; (4) "

The PCM technique removes the constraint that the sum of membership values is 1 to improve the non-intuitive membership value of the FCM technique.

The PCM technique uses typicality rather than membership to represent the degree to which each point belongs to a cluster (or cluster). The objective function of the PCM method is shown in Equation 5, and the second term of the right term of Equation 5 is a term added to remove a trivial solution in which the objective function is minimized when all values become zero.

[Equation 5]

here,

=

=

는 0과 1 사이의 수적인 값으로 i번째 클러스터에 속해져 있는

의 k번째 데이터의 전형성 정도를 나타낸다.

는 i번째 클러스터의 중심 벡터이다. here,

Is a numerical value between 0 and 1 that belongs to the i th cluster.

Represents the degree of typicality of the k-th data.

Is the center vector of the i th cluster.

는 각 클러스터의 부피를 나타내는 값으로 클러스터의 크기 추정과 특이점(outlier) 판별에 영향을 미친다.

는 특성 공간상의 변수이다.

Is a value representing the volume of each cluster, which affects the estimation of the cluster size and the outlier discrimination.

Is a variable in the property space.

m is the weight of the exponent indicating the effect on the degree of fuzzyness of the typical function. This value has the same range as and is a parameter that controls the amount of fuzzyness in the classification process. Usually m is set to 2.

값은 수학식 6과 같이 주어지고

값은 수학식 7과 같다.PCM, like FCM, is also commonly used to obtain local maxima through AO method.

The value is given by Equation 6

The value is shown in Equation 7.

&Quot; (6) "

[Equation 7]

Equation 6 considers only the distance between one data point and one cluster center, unlike Equation 3 considering the distance between one data point and all cluster centers.

Meanwhile, PFCM (Possibilistic Fuzzy C-Means clustering) technique using the GG method for clustering k first propagation characteristic value data selected in step S120 according to the present invention is designed to compensate for the shortcomings of FCM and PCM techniques. The objective function by the PFCM technique is expressed by Equation 8 below.

[Equation 8]

는 소속도에서 m과 동일한 역할을 전형성에서 하는 상수이다.Where a and b are weight constants for membership and typicality,

Is a constant that plays the same role as m in membership.

는 각 클러스터의 부피를 나타내는 값으로 수학식 5와 같이 계산되고, 클러스터의 크기 추정과 특이점(outlier) 판별에 영향을 미친다. 수학식 8에서의

는 수학식 3과 동일하게 계산되며

와

는 각각 수학식 9, 수학식 10으로 계산된다.Also

Is a value representing the volume of each cluster, calculated as in Equation 5, and affects the estimation of the size of the cluster and the outlier discrimination. In equation (8)

Is calculated equal to Equation 3

Wow

Are calculated by Equations 9 and 10, respectively.

[Equation 9]

[Equation 10]

Unlike the FCM and PCM techniques, PFCM considers the membership and the typicality at the same time as in Equations 9 and 10 for cluster-centric calculation, thereby alleviating the noise sensitivity of the FCM and solving the clustered problem of the PCM.

As such, the PFCM technique combines the advantages of the FCM technique and the PCM technique, but there are some problems due to the use of Euclidean distance. Euclidean distance is effective when the clusters do not overlap each other and form a sphere and the number of data points belonging to each cluster is similar.

Therefore, problems may arise in the case of non-spherical clusters. The GG method proposed by Gath and Geva (hereinafter, referred to as GG) is applied to the FCM scheme in order to improve this problem and find an elliptical cluster.

The GG method uses an improved method to ensure that the singularity does not occur in variance matrices, and uses distances that are inversely proportional to the Gaussian distribution function.

를 수학식 11과 같이 계산한다.Fuzzy Variance Matrix of Each Cluster by GG Method

Is calculated as in Equation 11.

 [Equation 11]

The computed variance matrix is used in Equation 12 to find the distance between the point and the cluster center.

[Equation 12]

는 각 클러스터의 사전확률로 수학식 13과 같이 정의한다.here,

Is defined as Equation 13 as the prior probability of each cluster.

[Equation 13]

Hereinafter, a method of implementing the FCM clustering algorithm using the GG method, which is a method of distance measurement, will be described.

Step 1. (initialization step)

을 초기화한다. 알고리즘 반복횟수를

로 표시한다.Determine the number of clusters c (2 <c <n) and choose the weight m of the exponent. Initial typical function

Initialize Algorithm iteration count

To be displayed.

을 구한다. Step 2. Center of Fuzzy Cluster Using Equation 10

.

Step 3. Using Equation 12, find the distance between the point and the center of the cluster.

을 계산한다.Step 4. New typical function using Equation 14

.

[Equation 14]

, N

이고,

이다.here,

, N

ego,

to be.

이다.Also,

to be.

Step 5.

이면 r=r+1로 정의하고 Step 2 단계를 반복 수행, 그러지 않으면 종료한다. (

: 임계값)Calculate Equation 15

If so, define r = r + 1 and repeat Step 2, otherwise terminate. (

: Threshold)

[Equation 15]

Next, a cluster having a minimum GG distance with respect to the cluster center vector and the first propagation characteristic value data is selected from the clusters of the plurality of first propagation characteristic value data formed by the PFCM method using the GG method in operation S130 (S140). .

Assuming that the RP belonging to the cluster with the smallest GG distance for the SNR data has the characteristics most similar to the SNR data measured at the TP, the selected RP can be used to improve the performance of indoor location estimation.

After selecting the most suitable cluster with the second propagation characteristic value data measured by the terminal among the clusters selected in step S140, the position of the terminal is estimated by calculating an average of RPs in the cluster (S150).

Here, the average of the RPs may include an arithmetic or geometric mean of the position coordinate values of the respective RPs.

As mentioned above, the configuration of the present invention has been described in detail through preferred embodiments of the present invention, but a person having ordinary knowledge in the art to which the present invention pertains disclosed herein without changing the technical spirit or essential features of the present invention. It will be appreciated that the present invention may be embodied in other specific forms than. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. The protection scope of the present invention is shown by the following claims rather than the above description, and all changes or modifications derived from the claims and their equivalents should be construed as being included in the scope of the present invention.

Claims (2)

Storing the first propagation characteristic value data collected at each reference point (RP) in a wireless environment in a database by a wireless fingerprint method; And
A reference point having the first propagation characteristic value data corresponding to the second propagation characteristic values received from a plurality of APs at a test point (TP) where a terminal is located is referred to as KNN (k-Nearest Neighbor) Search using PFCM (Possibilistic Fuzzy C-Mean) Mixing Technique
PFCM clustering-based mixed wireless indoor positioning method using a KNN and GG method comprising a. The method of claim 1, wherein the searching comprises:
Selecting k reference points having the first propagation characteristic value close to the second propagation characteristic value using the KNN technique;
Forming a cluster of the first propagation characteristic value data of the k selected reference points using a PFCM technique using the GG method;
Selecting a cluster having a minimum GG distance with respect to a cluster center vector and the first propagation characteristic value data among the formed clusters; And
Estimating the position of the terminal by calculating an arithmetic mean of the position coordinates of the reference points belonging to the selected cluster;
Hybrid indoor positioning method based on PFCM clustering using KNN and GG methods.

Images