CN102427603A - Positioning method of WLAN (Wireless Local Area Network) indoor mobile user based on positioning error estimation - Google Patents

Abstract

The invention provides a positioning method of a WLAN (Wireless Local Area Network) indoor mobile user based on positioning error estimation and relates to a positioning method of the WLAN indoor mobile user. In order to overcome the defects of traditional distance depending type and mode matching type fingerprint matching positioning methods and the problem of few RSS (Received Signal Strength) samples used for positioning by a mobile user, a nonlinear mapping function between the RSS samples and a positioning error, which is trained in an off-line stage is used to establish a relation between the RSS samples and the positioning error. In an on-line stage, the received RSS samples serve as the input of the nonlinear mapping function to estimate the positioning error, and the estimated positioning error is utilized to modify a positioning result of the distance depending type fingerprint matching method, thereby eliminating the influence of the positioning error and greatly promoting the positioning precision of the mobile user. When a position fingerprint database is not updated, the positioning precision of the positioning method provided by the invention is higher than that of the distance depending type fingerprint matching method. When the position fingerprint database is updated and the nonlinear mapping relation needs to be retrained, the distance depending type fingerprint matching method still can be utilized, thereby achieving higher practical application value.

Description

The indoor mobile subscriber's localization method of WLAN based on the position error estimation

Technical field

The present invention relates to WLAN indoor orientation method based on the position error estimation.

Background technology

In recent years, people increase day by day the position demand for services and open.But the positioning accuracy that receives covering of building and cellular system owing to satellite-signal under indoor environment is difficult to satisfy the requirement that indoor positioning is served.Therefore, for providing accurate location-based service, indoor user received many research organizations pay much.Along with extensively being arranged in the indoor environment, WLAN communication service is provided for the user; With respect to other indoor positioning technology; Like ultra broadband, ultrasonic wave, infrared ray, bluetooth etc., be considered to a kind of promising solution because of its economy based on the WLAN indoor locating system of fingerprint matching algorithm.This method generally is divided into two stages, that is: off-line phase and online stage.In off-line phase, at first select some certain location, promptly reference point (Reference Point, RP).On each reference point, gather enough RSS (Received Signal Strength) samples then and set up a location fingerprint database by RSS sample and reference point coordinate.In the online stage, being complementary with the RSS database of having set up through the measured RSS sample of user side wireless network card to estimate user's position coordinates.At present; Correlative study for fingerprint matching algorithm launches; Many algorithms and technology also are used as fingerprint matching algorithm; As based on the artificial neural net method of pattern matching, SVMs method, Adaptive Neuro-fuzzy Inference method and the nearest neighbor method that relies on based on distance, k nearest neighbor algorithm (K Nearest Neighbors, KNN), weighting k nearest neighbor method (Weighted KNN, WKNN).

Fingerprint matching algorithm based on pattern matching need be in off-line phase to model training, to set up the non-linear relation between RSS sample and the coordinate.If training result is well then can reach very high positioning accuracy, otherwise, if the indoor radio environmental change; The location fingerprint database update; Then can cause original Nonlinear Mapping relation to lose efficacy, need the plenty of time to train again, this was difficult to accomplish in the real-time positioning stage.By contrast; The advantage of the fingerprint matching algorithm that relies on based on distance be do not need the off-line phase training, calculate simple, insensitive to the indoor radio environmental change, but shortcoming to be precision limited and be subject to the selection of reference point locations and the influence of neighbour's reference point number.On the other hand, when the mobile subscriber was located, it is bigger that difficulty is compared fixed-line subscriber.The signal intensity samples number that user's wireless network card reads is less, can't reflect the radio characteristic of user present position comprehensively, so position error usually can be bigger, therefore need utilize out of Memory to improve good location accuracy.Current research only concentrates on how to calculate accurate localization result more, seldom can start with from the angle of calculation of position errors.If can estimate position error accurately, just can utilize the position error of estimation to eliminate the influence of error in the positioning result, obtain accurate localization result more.

Summary of the invention

The present invention is in order to solve the low problem of the indoor mobile subscriber's positioning accuracy of WLAN of existing method, thereby a kind of indoor mobile subscriber's localization method of estimating based on position error of WLAN is provided.

Based on the indoor mobile subscriber's localization method of WLAN that position error is estimated, it is realized by following steps:

Step 1, in indoor target localization environment, arrange N access point AP, guarantee that the optional position in the indoor target localization environment all can collect the signal strength signal intensity from all access point AP; M reference point is set in said target environment; N, M are positive integer;

Step 2, in off-line phase, write down M reference point place coordinate set, gather W RSS sample RSS in each reference point _Re, and according to all RSS sample RSS of M reference point place coordinate and each reference point _ReSet up the location fingerprint database; W is a positive integer;

Step 3, in the target localization environment, select the training points of Z location point,, set up RSS sample set RSS at V RSS sample of each training points collection as the Nonlinear Mapping function _TrWith position error (δ x _Tr, δ y _Tr) between Nonlinear Mapping relation; Z, V are positive integer;

Step 4, in the online stage, when the mobile subscriber receives RSS sample set RSS in test point _TeAfter, utilize the fingerprint matching method compute location result who relies on based on distance

Simultaneously also with input calculation of position errors (the δ x of the RSS sample of receiving as the Nonlinear Mapping function _Te, δ y _Te); Utilize formula then:

$\left\{\begin{array}{c}{x}_P=x_{\mathrm{Te}}^{\mathrm{DD}}+{\mathrm{\δx}}_{\mathrm{Te}}\\ y_p=y_{\mathrm{Te}}^{\mathrm{DD}}+{\mathrm{\δy}}_{\mathrm{Te}}\end{array}\right.$

Obtain final mobile subscriber's positioning result (x _P, y _P).

Z location point of selection in the target localization environment described in the step 3 gathered V RSS sample as the training points of Nonlinear Mapping function in each training points, sets up RSS sample set RSS _TrWith position error (δ x _Tr, δ y _Tr) between the concrete grammar of Nonlinear Mapping relation be:

At first, the coordinate (x of Z location point of record _Tr, y _Tr) and the RSS sample set RSS that collects in each position _TrThe fingerprint matching method that utilization relies on based on distance is calculated the positioning result of these training points

Utilize formula then:

$\left\{\begin{array}{c}{\mathrm{\δx}}_{\mathrm{Tr}}=x_{\mathrm{Tr}}-x_{\mathrm{Tr}}^{\mathrm{DD}}\\ {\mathrm{\δy}}_{\mathrm{Tr}}=y_{\mathrm{Tr}}-y_{\mathrm{Tr}}^{\mathrm{DD}}\end{array}\right.$

Calculation of position errors (δ x _Tr, δ y _Tr);

With RSS sample set RSS _TrAnd position error (δ x _Tr, δ y _Tr) respectively as the input and output of training Nonlinear Mapping function, train in off-line phase, obtain RSS sample set RSS _TrAnd position error (δ x _Tr, δ y _Tr) between Nonlinear Mapping relation.

Beneficial effect: method of the present invention at first utilizes the fingerprint matching method that relies on based on distance to draw positioning result; Be utilized in the Nonlinear Mapping Function Estimation position error of off-line phase training then; The positioning result of corrected range dependent form fingerprint matching method is eliminated affect positioning.This method was not when both fingerprint database had upgraded in the position; Relative distance dependent form fingerprint matching method increases substantially positioning accuracy; Again can be in the position fingerprint database when upgrading; Normal mode matching type fingerprint matching method can't be used under the situation that need train the Nonlinear Mapping relation again, utilizes apart from dependent form fingerprint matching method and accomplishes the location.By this method, can only receive under the situation of minority RSS sample, eliminate the influence of error in the positioning result to a certain extent, obtain accurate localization result more the mobile subscriber.

Description of drawings

Fig. 1 is the schematic flow sheet of localization method of the present invention.(x among the figure _Re, y _Re) be coordinates of reference points and RSS _ReBe the RSS of reference point place sample; (x _Tr, y _Tr) be the coordinate and the RSS of training points _TrRSS sample for the collection of training points place;

For the training points place utilizes apart from dependent form fingerprint matching method according to RSS _TrThe elements of a fix that calculate; (δ x _Tr, δ y _Tr) be the position error at training points place; RSS _TeWith

Be respectively the RSS sample that receives and the same positioning result at test point place of online stage apart from the dependent form fingerprint matching algorithm; (δ x _Te, δ y _Te) be to utilize the Nonlinear Mapping function of off-line phase training according to RSS sample RSS _TeThe position error of estimating; (x _P, y _P) be final positioning result.Fig. 2 is the experimental situation sketch map of embodiment two.Fig. 3 is the positioning result and the neural net (Back Propagation Neural Network, BPNN) the emulation sketch map of positioning result after the estimation error and test point that adopt back-propagation algorithm that utilizes the KNN algorithm.Fig. 4 is positioning result that utilizes the KNN algorithm and the neural net that adopts back-propagation algorithm (Back Propagation Neural Network, BPNN) the emulation sketch map as a result of the cumulative probability after the estimation error.

Embodiment

Embodiment one: combine Fig. 1 that this execution mode is described, the step of this execution mode is following:

Step 1, set up two-dimensional coordinate system in off-line phase; And gather from access points (Access Point at the reference point place of known coordinate; AP) reception signal strength signal intensity RSS (Received Signal Strength) sample, and set up the location fingerprint database;

Step 2, calculate in online stage; Utilize position fingerprint matching in reception signal strength signal intensity RSS that KNN that the present invention selects receives user's wireless network card apart from dependent form location fingerprint matching algorithm and the radio signal coverage diagram, calculate one at the beginning of the elements of a fix;

If the quantity of RP and AP is respectively m and n.S _IjBe defined as in the database RSS sample average, s from j AP of i reference point _jIt is the RSS sample average that the user measures from n AP.Euclidean distance d _iBy computes

$d_i=\sqrt{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{|s_j-S_{\mathrm{Ij}}|}^2},i=\mathrm{1,2},\·\·\·,m$ Formula 1

Institute's calculated distance is arranged with ascending order then.Selecting has k minimum d in i the distance _iReference point, and get mean value calculation user's the position coordinates of their coordinate.As shown in the formula

$(\stackrel{\‾}{x},\stackrel{\‾}{y})=\frac{1}{k}\underset{i=1}{\overset{k}{\mathrm{\Σ}}}(x_i,y_i)$ Formula 2

Wherein, (x _i, y _i) be the position coordinates of selected RP,

It is the positioning result of KNN algorithm.

Step 3, after setting up the location fingerprint database, in experimental situation, select some location points as training points, set up the Nonlinear Mapping relation between RSS sample and the position error.At first, write down the coordinate (x of these points _Tr, y _Tr) and the RSS sample RSS that collects at this place _TrUtilization is based on the positioning result of these training points of fingerprint matching method KNN algorithm computation of distance dependence

Utilize formula 3 calculation of position errors (δ x then _Tr, δ y _Tr).

$\left\{\begin{array}{c}{\mathrm{\δ\; x}}_{\mathrm{Tr}}=x_{\mathrm{Tr}}-x_{\mathrm{Tr}}^{\mathrm{DD}}\\ {\mathrm{\δ\; y}}_{\mathrm{Tr}}=y_{\mathrm{Tr}}-y_{\mathrm{Tr}}^{\mathrm{DD}}\end{array}\right.$ Formula 3

With RSS sample RSS _TrAnd position error (δ x _Tr, δ y _Tr) respectively as the input and output of training based on the neural net of back-propagation algorithm, train in off-line phase, obtain RSS sample RSS _TrAnd position error (δ x _Tr, δ y _Tr) between Nonlinear Mapping relation.

Step 4: in the online stage, when the user receives RSS sample RSS in test point _TeAfter, utilize the fingerprint matching method compute location result identical with step 3 Simultaneously also with the RSS sample of receiving as input calculation of position errors (the δ x that trains based on the neural net of back-propagation algorithm _Te, δ y _Te).Utilize formula 4 to obtain final positioning result (x afterwards _P, y _P).

$\left\{\begin{array}{c}{x}_P=x_{\mathrm{Te}}^{\mathrm{DD}}+{\mathrm{\δ\; x}}_{\mathrm{Te}}\\ y_p=y_{\mathrm{Te}}^{\mathrm{DD}}+{\mathrm{\δ\; y}}_{\mathrm{Te}}\end{array}\right.$ Formula 4

By this method, can only receive under the situation of minority RSS sample, eliminate the influence of error in the positioning result to a certain extent, obtain accurate localization result more the mobile subscriber.

Embodiment two: combine Fig. 1 to Fig. 4 that this execution mode is described; This execution mode is the process according to embodiment one with embodiment one difference; The validity of method of testing under the experimental situation of Fig. 2; Wherein, 9 Linksys WAP54G AP are arranged in the indoor environment of 24.9m * 66.4m.The experiment path be in 3 meters wide corridors from A to D.In the test, utilize Asus's notebook computer image data.It is equipped with Intel PRO/Wireless 3945ABG wireless network card and RSS sample collection software NetStumbler, 2 RSS samples of sampling rate per second.In off-line phase, select the reference point of 182 spacing 1m altogether along the corridor, each RP goes up and collects 150 seconds totally 300 RSS samples.And, have 960 training RSS samples along the experiment path and be used for training BPNN by collection.In the online stage, choose 96 test points altogether along the experiment path, spacing is 0.6m, 2 RSS samples of the collection of each test point.This paper chooses the KNN algorithm as the fingerprint matching method, and neighbour's parameter k of KNN algorithm is made as 7.The present invention selects one three layers BPNN.It is made up of an input layer, a hiding layer and an output layer.In the RSS sample of input layer input from different AP, the position error of output layer output X axle and Y axle.Hide layer neuron number and cycle of training number be respectively 20 and 10000 times, the performance of KNN algorithm and KNN/BPNN method proposed by the invention is more as shown in table 1.

Table 1 algorithm performance relatively

Experimental result shows that after the adding estimation error, the positioning accuracy of system significantly improves.Compare the mean error of KNN algorithm and have only 2.54m, this paper proposition can reach 1.33m based on the mean error of the localization method KNN/BPNN of estimation error.As shown in Figure 3, KNN is very undesirable at the positioning result at iron gate place, and this is because in this local reflection owing to iron gate, radio signal more complicated.But, after the estimation error of application based on BPNN, eliminated these negative effects fully according to estimated position error.

Claims (2)

1. the indoor mobile subscriber's localization method of estimating based on position error of WLAN, it is characterized in that: it is realized by following steps:

Step 1, in indoor target localization environment, arrange N access point AP, guarantee that the optional position in the indoor target localization environment all can collect the signal strength signal intensity from all access point AP; M reference point is set in said target environment; N, M are positive integer;

Step 2, in off-line phase, write down M reference point place coordinate set, gather W RSS sample RSS in each reference point _Re, and according to all RSS sample RSS of M reference point place coordinate and each reference point _ReSet up the location fingerprint database; W is a positive integer;

Step 3, in the target localization environment, select the training points of Z location point,, set up RSS sample set RSS at V RSS sample of each training points collection as the Nonlinear Mapping function _TrWith position error (δ x _Tr, δ y _Tr) between Nonlinear Mapping relation; Z, V are positive integer;

Step 4, in the online stage, when the mobile subscriber receives RSS sample set RSS in test point _TeAfter, utilize the fingerprint matching method compute location result who relies on based on distance

Simultaneously also with input calculation of position errors (the δ x of the RSS sample of receiving as the Nonlinear Mapping function _Te, δ y _Te); Utilize formula then:

$\left\{\begin{array}{c}{x}_P=x_{\mathrm{Te}}^{\mathrm{DD}}+{\mathrm{\δx}}_{\mathrm{Te}}\\ y_p=y_{\mathrm{Te}}^{\mathrm{DD}}+{\mathrm{\δy}}_{\mathrm{Te}}\end{array}\right.$

Obtain final mobile subscriber's positioning result (x _P, y _P).

2. the indoor mobile subscriber's localization method of estimating based on position error of WLAN according to claim 1; It is characterized in that Z the location point of selection in the target localization environment described in the step 3 is as the training points of Nonlinear Mapping function; Gather V RSS sample in each training points, set up RSS sample set RSS _TrWith position error (δ x _Tr, δ y _Tr) between the concrete grammar of Nonlinear Mapping relation be:

At first, the coordinate (x of Z location point of record _Tr, y _Tr) and the RSS sample set RSS that collects in each position _TrThe fingerprint matching method that utilization relies on based on distance is calculated the positioning result of these training points

Utilize formula then:

$\left\{\begin{array}{c}{\mathrm{\δx}}_{\mathrm{Tr}}=x_{\mathrm{Tr}}-x_{\mathrm{Tr}}^{\mathrm{DD}}\\ {\mathrm{\δy}}_{\mathrm{Tr}}=y_{\mathrm{Tr}}-y_{\mathrm{Tr}}^{\mathrm{DD}}\end{array}\right.$

Calculation of position errors (δ x _Tr, δ y _Tr);

With RSS sample set RSS _TrAnd position error (δ x _Tr, δ y _Tr) respectively as the input and output of training Nonlinear Mapping function, train in off-line phase, obtain RSS sample set RSS _TrAnd position error (δ x _Tr, δ y _Tr) between Nonlinear Mapping relation.

Images