Connect public, paid and private patent data with Google Patents Public Datasets

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

Download PDF

Info

Publication number
CN103648106A
CN103648106A CN 201310750528 CN201310750528A CN103648106A CN 103648106 A CN103648106 A CN 103648106A CN 201310750528 CN201310750528 CN 201310750528 CN 201310750528 A CN201310750528 A CN 201310750528A CN 103648106 A CN103648106 A CN 103648106A
Authority
CN
Grant status
Application
Patent type
Prior art keywords
positioning
radio
method
indoor
wifi
Prior art date
Application number
CN 201310750528
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
Original Assignee
哈尔滨工业大学
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

Links

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

—种基于类别匹配的半监督流形学习的WiFi室内定位方法 - kind of WiFi indoor positioning semi-supervised manifold learning based on matching categories

技术领域 FIELD

[0001] 本发明涉及一种室内定位方法,具体涉及一种基于类别匹配的半监督流形学习的WiFi室内定位方法。 [0001] The present invention relates to a method for indoor positioning, particularly relates to a WiFi indoor positioning method based on the category matching semi-supervised learning manifold.

背景技术 Background technique

[0002] 随着无线局域网络在世界范围的飞速发展和移动终端设备的广泛普及,近年来出现了许多室内定位相关的技术和应用。 [0002] With the widespread popularity of wireless local area networks in the world and the rapid development of mobile terminal equipment, in recent years there have been many indoor positioning technology and related applications. 由于多径效应、信号衰减及室内定位环境的复杂性,基于传统的信号传播模型的室内定位方法难以达到高精度的室内定位要求。 Due to multipath effects, attenuation, and the complexity of indoor positioning environment, the conventional method of indoor positioning based on a signal propagation model is difficult to achieve high accuracy indoor positioning requirements. 基于到达时间(Time of Arrival),到达时间差(Time Difference of Arrival)和到达角度(Angles ofArrival)等定位方法虽然可以基本满足定位精度需求,然而都需要定位终端有额外的硬件设备支持,具有较大局限性,从而导致基于上述几类定位方法的室内定位系统没有得到普及。 Based on the arrival time (Time of Arrival), a time difference (Time Difference of Arrival) and angle of arrival (Angles ofArrival) and other positioning method, although can basically meet the positioning accuracy requirements reach, however, it requires positioning terminal has additional hardware support, with a larger limitations, resulting in not been universal indoor location system based on the above-mentioned types of positioning method.

[0003] 目前,基于WLAN位置指纹(Finger Print)的WiFi室内定位方法得到了广泛应用。 [0003] Currently, WLAN location fingerprint (Finger Print) of WiFi-based indoor positioning method has been widely used. 该方法的网络构建方法成本低廉,其使用2.4GHz ISMdndustrial Science Medicine)公共频段且无需在现有设施之上添加定位测量专用硬件。 The method of constructing the network cost method, using 2.4GHz ISMdndustrial Science Medicine) common frequency band and without adding special hardware positioning measurements on existing facilities. 只需要通过移动终端的无线网卡及相应软件测量接收到的接入点(Access Point, AP)的信号强度(Received SignalStrength, RSS),由此来构建网络信号覆盖图(Radio Map),进而通过匹配算法来预测移动用户所处位置的坐标,或相对位置。 Only received through the wireless network card of the mobile terminal and an access point corresponding to the measurement software (Access Point, AP) signal strength (Received SignalStrength, RSS), thereby constructing a network coverage map (Radio Map), and further by matching algorithm to predict the coordinates of the mobile user's location, or relative position.

[0004] 然而通过该方式建立的Radio Map包含有庞大的数据信息,且随着定位区域扩大,Radio Map可能(依据定位匹配方式及算法选择)呈指数形势增长。 [0004] However, the manner established by Radio Map contains huge data information, and with the location area to expand, Radio Map may grow exponentially situation (based positioning methods and matching algorithm selection). 获得尽可能多的相关数据特征信息对于整个系统来说会提升定位精度,但是处理大量的特征信息增加算法开销,定位算法无法在处理能力有限的移动终端上有效运行,同时某些特征信息可能是对于定位没有作用甚至有负面作用,致使匹配效率降低,从而导致匹配定位算法的实现变得更加复杂,并且定位精度下降。 Get as much relevant data feature information for the whole system will improve location accuracy, but a lot of the feature information processing algorithms to increase overhead, positioning algorithm can not operate effectively on a limited processing capability of the mobile terminal, while some feature information may be for positioning does not work even have a negative effect, resulting in matching efficiency is reduced, resulting in matching to achieve localization algorithm becomes more complex, and positioning accuracy.

[0005] 当AP的数目增加及定位的参考点(Reference Point)增加时,Radio Map的数据信息增加。 [0005] When the number of the AP and the positioning of the reference point (Reference Point) increases, the data information of the Radio Map. 此时,Radio Map中代表的AP数目的信息表示了数据的维数。 At this time, information representing the number of AP Radio Map showing the dimension of the data. 因此,当AP数目增加,Radio Map就变成了高维数据。 Thus, when increasing the number of AP, Radio Map becomes a high-dimensional data. 为减轻处理高维数据的负担,降维算法是有效的解决方法之一。 To alleviate the burden of processing high-dimensional data, dimension reduction algorithm is an effective solution. 高维数据可能包含很多特征,这些特征都在描述同一个事物,这些特征一定程度上是紧密相连的。 High-dimensional data may contain many features that are describing the same thing, to some extent, these characteristics are closely linked. 如当从各个角度对同一个物体同时拍照时,得到的数据就含有重叠的信息。 Simultaneously such as when photographing the same object from various angles, the resulting data to contain overlapping information. 如果能得到这些数据的一些简化的不重叠的表达,将会极大地提高数据处理运行的效率并一定程度上提高准确度。 Some simplifications can be obtained if the data do not overlap with the expression of these will greatly improve the efficiency of data processing operation and improve the accuracy of a certain extent. 降维算法的目的也正是在于提高高维数据的处理效率。 Dimensionality reduction algorithms aim is also to improve the efficiency of processing high-dimensional data.

[0006] 除了可以简化数据使其能够高效处理外,降维方法还可以实现数据可视化。 [0006] In addition it is possible to simplify the data processing efficiency, the method may further dimension reduction data visualization. 由于很多统计学的和机器学习算法对于最优解的准确性很差,降维的可视化应用可以令用户能够实际看到高维数据的空间结构和算法输出的能力,具有很强的应用价值。 Since many statistical and machine learning algorithms for the optimal solution accuracy is poor, lower-dimensional visualization applications can make the user the ability to be able to actually see the spatial structure and algorithm output of high-dimensional data, it has a strong practical value.

[0007]目前有很多基于不同目的的降维算法,包括有线性与非线性降维算法。 [0007] There are many different purposes dimensionality reduction algorithms, including linear and nonlinear dimensionality reduction algorithm. 其中PCA(Principal Component Analysis)和LDA(Linear Discriminant Analysis)是典型的线性降维算法。 Wherein the PCA (Principal Component Analysis) and LDA (Linear Discriminant Analysis) is a typical linear dimension reduction algorithm. 这一类算法对于具有线性结构的高维数据有着良好的处理结果,但对于非线性结构的高维数据没有好的结果。 This class of algorithms have good processing results, but no good result for the non-linear high-dimensional data structure for high-dimensional data having a linear structure. 典型的非线性降维算法以流形学习(ManifoldLearning)算法。 Typical nonlinear dimensionality reduction manifold learning algorithm (ManifoldLearning) algorithm. 2000年Science杂志上同一期发表了3篇有关于流形学习算法中提出了2 种经典的流形学习算法:LLE (Local Linear Embedding)及ISOMAP (Isometric Mapping)。 In 2000 on the same issue of Science magazine published three on manifold learning algorithms have been proposed two kinds of classic manifold learning algorithms: LLE (Local Linear Embedding) and ISOMAP (Isometric Mapping). 由此,各种基于不同的准则的流形学习算法被提出并有一部分流形学习算法应用于图像处理方面。 Accordingly, various manifold learning algorithms based on different criteria have been proposed and a portion of manifold learning algorithms applied to the image processing. LDE (Local Discriminant Embedding)算法是流形学习算法中较晚提出的,它是一种的典型的基于特征提取的流形学习算法,而不只基于可视化目标。 LDE (Local Discriminant Embedding) algorithm are manifold learning algorithm proposed later, which is a typical feature extraction manifold learning algorithms based on, not only based on the visual target.

[0008] 对于上述的降维算法并不能对在线获得的RSS数据或者新增的RSS来提高降维的精度。 [0008] For the above dimension reduction algorithm and can not get online RSS data or new RSS to improve the accuracy of dimensionality reduction. 由于室内定位环境的变化,在不同时刻,特别是在长时间后,RSS数据之间的相关性就会降低。 Indoor positioning due to changes in the environment, at different times, especially after a long time, the correlation between the RSS data is reduced. 在已有算法中,有一类算法可以将在不同时间的RSS数据同时加入相应的已有数据中,从而增强不同数据之间的相关性,提高降维精度。 In the existing algorithm, there is a corresponding class of algorithms can be added to existing data in the RSS data simultaneously at different times, so as to enhance the correlation between the different data and improve the accuracy of dimension reduction. 这一类算法通常称之半监督算法。 This type of algorithm usually termed semi-supervised algorithm. 根据半监督算法的特点,提出半监督鉴别嵌入(Sem1-supervised DiscriminantEmbedding, SDE )算法。 According to the characteristics of semi-supervised algorithm on semi-supervised embedded identification (Sem1-supervised DiscriminantEmbedding, SDE) algorithm.

[0009] 根据SDE算法的特点,采用类别匹配的方式将在线新得到的RSS加入已有RadioMap,然后进行局部鉴别嵌入降维,从而提出基于类别匹配的半监督局部鉴别嵌入算法(Classification Matching Based Sem1-supervised Discriminant Embedding,CM-SDE)。 [0009] According to the characteristics SDE algorithm, using the category matching manner RSS new line has been added RadioMap, and partial identification embedding dimension reduction, thereby embedding algorithm proposed (Classification Matching Based Sem1 category matching local authentication based on semi-supervised -supervised Discriminant Embedding, CM-SDE). 采用CM-SDE算法对Radio Map进行降维,得出降维后的Radio Map,将对降维后的RadioMap室内定位,从而提出基于CM-SDE算法的WiFi室内定位算法。 Using CM-SDE Radio Map algorithm to reduce the dimension after dimension reduction derived Radio Map, will drop RadioMap indoor positioning dimensionality, which made WiFi indoor positioning algorithm based CM-SDE algorithms.

发明内容 SUMMARY

[0010] 本发明是要解决现有WiFi室内定位方法中存在的Radio Map数据库大,以及由于在线定位阶段计算复杂度高而引起的难以应用在线阶段获得RSS数据、难于在移动终端实现以及难于保证定位的实时性要求等问题,而提供了一种基于类别匹配的半监督流形学习的WiFi室内定位方法。 [0010] The present invention is to solve the Radio Map database is large, and difficult to apply because the online phase line positioning stage high computational complexity caused by the conventional WiFi indoor positioning method exists RSS data obtained, it is difficult to achieve and difficult to ensure that the mobile terminal positioning real-time requirements and other issues, and provides a WiFi indoor positioning method for matching categories based on semi-supervised manifold learning.

[0011] 基于类别匹配的半监督流形学习的WiFi室内定位方法离线阶段定位过程按以下步骤实现: [0011] WiFi indoor positioning method of positioning the offline phase match the category of semi-supervised manifold learning based on the realization of the following steps:

[0012] 一、对待定位的室内区域布置AP,使无线信号覆盖待定位的室内区域,完成WiFi网络构建; [0012] First, the treatment chamber is disposed targeting the AP, so that an indoor wireless coverage area to be positioned to complete WiFi network construction;

[0013] 在待定位的室内区域规则选取并记录参考点的相应坐标,测量并依次记录参考点接收到的所有AP的RSS信号作为位置特征信息,构建Radio Map,并存储Radio Map ; [0013] and the corresponding coordinates of the reference point selected recording rules to be positioned in the interior region of, and sequentially record all measurement reference point AP received RSS information signal as the feature position, constructed Radio Map, and stores the Radio Map;

[0014] 二、采用GMST本征维数估计算法对步骤一中构建的Radio Map的本征维数进行分析,得到的本征维数作为CM-SDE算法的输入参数之一,决定Radio Map降维后的维数; [0014] Second, the use GMST intrinsic dimension algorithm estimates intrinsic dimension constructed in step a Radio Map which were analyzed, one of the intrinsic dimensionality as CM-SDE algorithm input parameters obtained, determined drop Radio Map the dimension after dimension;

[0015] 三、采用KFCM算法对Radio Map进行聚类分析,实现建立的Radio Map的类别标记,并作为CM-SDE的输入参数之一,并且提供相应的初始聚类中心及类别标记; [0015] Third, the use of the Radio Map algorithm KFCM cluster analysis, to achieve category tags Radio Map established as one of the input parameters of the CM-SDE, and provides a corresponding category marker and initial cluster centers;

[0016] 四、步骤二中的本征维数与步骤三中的类别标记作为输入参数,采用CM-SDE算法对步骤一中构建的Radio Map降维,得出相应的降维后的RadioMap'RadioMap*作为在匹配定位数据库用于在线定位阶段; [0016] IV in two steps in the intrinsic dimensionality Step three categories labeled as an input parameter, using the Radio Map CM-SDE dimension reduction algorithm constructed in one step, the dimensionality reduction obtained RadioMap corresponding ' RadioMap * as matching the positioning database for on-line positioning stage;

[0017] 五、将不同用户在线定位阶段测试得到的未标记RSS,采用类别匹配的方式加入至已有Radio Map中,得到相应的包含未标记信号覆盖图RadioMapul,通过类别匹配更新的聚类中心作为CM-SDE算法中新的类别输入参数; [0017] Five different online users unlabelled RSS positioning phase of the test obtained using the category matching methods have been added to the Radio Map, to give the corresponding cluster center unlabeled coverage map comprises RadioMapul, by matching the updated category as a new category of input parameters CM-SDE algorithm;

[0018] 六、步骤五中的更新的聚类中心作为输入参数,采用CM-SDE对RadioMapul降维得到特征变换矩阵V' ,Ψ与RadioMap*共同构成在线匹配定位数据库,用于在线阶段定位;其中,所述线阶段定位具体为: [0018] VI, step 5 of the updated cluster center as an input parameter, using CM-SDE RadioMapul of dimension reduction transformation matrix obtained wherein V ', Ψ together with RadioMap * configuration matching online location database for locating the online phase; wherein the wire positioning stage is specifically:

[0019] 六(一)、在线测试RSS ; [0019] VI (a), online test RSS;

[0020] 六(二)、采用V将RSS降维为RSS* ; [0020] VI (B), using the RSS V of dimensionality reduction RSS *;

[0021] 六(三)、采用KNN算法进行匹配定位输出定位结果; [0021] VI (c), using the KNN algorithm for outputting a positioning result Targeting;

[0022] 六(四)、用户定位终端定位数据库更新; [0022] hexa (tetra), the positioning terminal location database updates the user;

[0023] 即完成了一种基于类别匹配的半监督流形学习的WiFi室内定位方法的离线阶段实现方式。 [0023] That is done offline phase WiFi indoor positioning method based on the category matching semi-supervised learning implementations manifold.

[0024] 基于类别匹配的半监督流形学习的WiFi室内定位方法在线阶段定位过程通过下述步骤实现: [0024] WiFi indoor positioning method of positioning the online phase semi-supervised manifold matching categories based learning achieved by the following steps:

[0025] 一、在线测试RSS; [0025] First, the online test RSS;

[0026] 二、将得到的待定位点的RSS采用特征变换矩阵降维变换得到RSS* ; [0026] Second, the anchor point will be obtained using the RSS feature dimensionality reduction transformation matrix transform RSS *;

[0027] 三、采用KNN算法对RSS*与Radio Map*匹配位,对待定位点的具体位置坐标进行预测并进行在线数据的更新,其实现过程为: [0027] Third, the use of KNN algorithm with specific location coordinates RSS * Radio Map * Match bit treat setpoint is predicted and updated online data, the implementation process is:

[0028] (I)在线阶段,测试点处接收的RSS= [AP1, AP2,…,APn],与特征变换矩阵V'相乘,从而得出降维后的RSS' =[AP1; AP2,…APd],其中d表示本征维数; [0028] (I) online phase, at the test point the received RSS = [AP1, AP2, ..., APn], wherein the transform matrix V 'is multiplied, so as to arrive after the dimensionality reduction RSS' = [AP1; AP2, ... APd], where d represents the intrinsic dimension;

[0029] (2)采用KNN算法实现RSS^与Radio Map*的匹配,采用与RSS^最近的K个参考点的坐标的平均值作为测试点(X' ,Y'),其表达式为: [0029] (2) with the KNN algorithm RSS ^ Radio Map * match, using the average value of the coordinates of the nearest reference point K RSS ^ as a test point (X ', Y'), which is expressed as:

[0030] [0030]

Figure CN103648106AD00061

[0031] 式中,(X' , y')为测试点预测的坐标,(Xi, y)为第i个近邻点的坐标,K为KNN算法中近邻的数目; [0031] In the formula, (X ', y') coordinates of the test point predicted, (Xi, y) coordinates of the i-th neighboring point, K is the number of neighbor KNN algorithm;

[0032] 四、用户定位终端定位数据库更新,即完成了基于类别匹配的半监督流形学习的WiFi在线阶段室内定位方法。 [0032] Fourth, locate the user terminal location database updates, complete the online phase WiFi indoor positioning methods that match the category based on semi-supervised manifold learning.

[0033] 发明效果: [0033] Effect of the Invention:

[0034] 针对本发明提出的CM-SDE算法及背景需要求,通过对哈尔滨工业大学科学园2A栋12层走廊构成的室内定位区域进行定位。 [0034] for the CM-SDE algorithms and background provided by the present invention, on request, by the indoor positioning region of Harbin Institute of Technology Science Park, Building 12 layer 2A corridor composed of positioning. 采用联想V450笔记本电脑并结合NetStumbler软件测试所有参考点处的RSS,构成Radio Map,并采用CM-SDE算法对Radio Map降维并采用KNN算法实现室内定位。 Lenovo V450 notebook computers using combined NetStumbler software testing RSS at all reference points, constitute Radio Map, and the use of CM-SDE Radio Map dimensionality reduction algorithm and KNN algorithm using indoor location. 在附图4的仿真中,降维后的Radio Map的维数为原始RadioMap维数的三分之一。 In the simulation of FIG. 4, the dimension is reduced to one-third of the original number of Radio Map RadioMap dimensional dimensionality. 从附图4的仿真结果来看,采用CM-SDE算法与初始的KNN的算法的定位性能可比拟,但CM-SDE的定位复杂度仅为三分之一,且CM-SDE算法可以有效地应用实时得到的新的RSS来Radio Map的密度,从而有效的提高定位精度。 From the point of view of the simulation results in Figure 4, the positioning performance of the algorithm using CM-SDE KNN algorithm and will be comparable to the original, but the positioning of the CM-SDE complexity is only one third, and the CM-SDE algorithm can real-time to get a new RSS Radio Map of density, so as to effectively improve the positioning accuracy.

附图说明 BRIEF DESCRIPTION

[0035] 图1是本发明中离线数据库定位实施流程图;实线箭头表示步骤之间的数据传输;[0036] 图2是本发明中在线数据库定位实施流程图;实线箭头表示步骤之间的数据传输; [0035] FIG. 1 is positioned according to the present invention the offline database flowchart; solid arrow indicates a data transmission between the steps; [0036] FIG 2 is positioned according to the present invention, a flowchart of the online database; solid arrow indicates the step between data transfer;

[0037] 图3是基于WiFi的室内定位网络的构建及实验环境示意图; [0037] FIG. 3 is a schematic view of experimental and build environmental chamber a WiFi-based positioning network;

[0038] 图4是采样网格图; [0038] FIG 4 is a sampling grid;

[0039] 图5是采用CM-SDE与KNN算法定位性能对比图;其中,表示CM-SDE,表示 [0039] FIG. 5 is the use of CM-SDE with the KNN algorithm performance comparison chart positioned; wherein represents a CM-SDE, represents

KNN。 KNN.

具体实施方式 detailed description

[0040] 具体实施方式一:本实施方式的基于类别匹配的半监督流形学习的WiFi室内定位方法离线阶段定位过程按以下步骤实现: [0040] DETAILED DESCRIPTION a: Manifold semi-supervised learning based on matching categories WiFi indoor positioning method of positioning the offline phase process embodiment according to the present embodiment implemented by the following steps:

[0041] 一、对待定位的室内区域布置AP,使无线信号覆盖待定位的室内区域,完成WiFi网络构建; [0041] First, the treatment chamber is disposed targeting the AP, so that an indoor wireless coverage area to be positioned to complete WiFi network construction;

[0042] 在待定位的室内区域规则选取并记录参考点的相应坐标,测量并依次记录参考点接收到的所有AP的RSS信号作为位置特征信息,构建Radio Map,并存储Radio Map ; [0042] and the corresponding coordinates of the reference point selected recording rules to be positioned in the interior region of, and sequentially record all measurement reference point AP received RSS information signal as the feature position, constructed Radio Map, and stores the Radio Map;

[0043] 二、采用GMST本征维数估计算法对步骤一中构建的Radio Map的本征维数进行分析,得到的本征维数作为CM-SDE算法的输入参数之一,决定Radio Map降维后的维数; [0043] Second, the use GMST intrinsic dimension algorithm estimates intrinsic dimension constructed in step a Radio Map which were analyzed, one of the intrinsic dimensionality as CM-SDE algorithm input parameters obtained, determined drop Radio Map the dimension after dimension;

[0044] 三、采用KFCM算法对Radio Map进行聚类分析,实现建立的Radio Map的类别标记,并作为CM-SDE的输入参数之一,并且提供相应的初始聚类中心及类别标记; [0044] Third, the use of the Radio Map algorithm KFCM cluster analysis, to achieve category tags Radio Map established as one of the input parameters of the CM-SDE, and provides a corresponding category marker and initial cluster centers;

[0045] 四、步骤二中的本征维数与步骤三中的类别标记作为输入参数,采用CM-SDE算法对步骤一中构建的Radio Map降维,得出相应的降维后的RadioMap'RadioMap*作为在匹配定位数据库用于在线定位阶段; [0045] IV in two steps in the intrinsic dimensionality Step three categories labeled as an input parameter, using the Radio Map CM-SDE dimension reduction algorithm constructed in one step, the dimensionality reduction obtained RadioMap corresponding ' RadioMap * as matching the positioning database for on-line positioning stage;

[0046] 五、将不同用户在线定位阶段测试得到的未标记RSS,采用类别匹配的方式加入至已有Radio Map中,得到相应的包含未标记信号覆盖图RadioMapul,通过类别匹配更新的聚类中心作为CM-SDE算法中新的类别输入参数; [0046] Five different online users unlabelled RSS positioning phase of the test obtained using the category matching methods have been added to the Radio Map, to give the corresponding cluster center unlabeled coverage map comprises RadioMapul, by matching the updated category as a new category of input parameters CM-SDE algorithm;

[0047] 六、步骤五中的更新的聚类中心作为输入参数,采用CM-SDE对RadioMapul降维得到特征变换矩阵V' ,Ψ与RadioMap*共同构成在线匹配定位数据库,用于在线阶段定位;其中,所述线阶段定位具体为: [0047] VI, step 5 of the updated cluster center as an input parameter, using CM-SDE RadioMapul of dimension reduction transformation matrix obtained wherein V ', Ψ together with RadioMap * configuration matching online location database for locating the online phase; wherein the wire positioning stage is specifically:

[0048] 六(一)、在线测试RSS ; [0048] VI (a), online test RSS;

[0049] 六(二)、采用V将RSS降维为RSS* ; [0049] VI (B), using the RSS V of dimensionality reduction RSS *;

[0050] 六(三)、采用KNN算法进行匹配定位输出定位结果; [0050] VI (c), using the KNN algorithm for outputting a positioning result Targeting;

[0051] 六(四)、用户定位终端定位数据库更新; [0051] hexa (tetra), the positioning terminal location database updates the user;

[0052] 即完成了一种基于类别匹配的半监督流形学习的WiFi室内定位方法的离线阶段实现方式。 [0052] That is done offline phase WiFi indoor positioning method based on the category matching semi-supervised learning implementations manifold.

[0053] 所述离线阶段与在线阶段均在定位终端完成。 [0053] The online phase and the offline phase are completed in the positioning terminal.

[0054] 具体实施方式二:本实施方式与具体实施方式一不同的是:步骤二中采用GMST本征维数估计算法对步骤一中构建的Radio Map的本征维数进行分析,其计算公式为: [0054] DETAILED Embodiment 2: This embodiment and the exemplary first embodiment except that: in step two using GMST intrinsic dimension estimates intrinsic dimension algorithm to step a constructed Radio Map which was analyzed, calculated for:

[0055] d dim =+,测地距最小生成树算法 [0055] d dim = +, from the minimum spanning tree algorithm geodesic

1-α[0056] 上式中1 中的a表示最小生成树的线性拟合表达式y=ax+b的斜率。 1-α [0056] in the above formula represents a linear minimum spanning tree 1 a fitting expression y = ax + b is the slope.

1-α 1-α

[0057] 其它步骤及参数与具体实施方式一相同。 [0057] Other steps and parameters same as a specific embodiment.

[0058] 具体实施方式三:本实施方式与具体实施方式一或二不同的是:步骤四中生成新的降维后的信号覆盖图RadioMap*与步骤六中的V'表达式为: [0058] DETAILED Embodiment 3: Embodiment of the present embodiment and the exemplary embodiment one or two exceptions: Step four new signals generated overlay RadioMap * and six step after the dimensionality reduction V 'expression:

[0059] Radio Map*=V/.X [0059] Radio Map * = V / .X

[0060] X是需要降维的Radio Map。 [0060] X is a need to reduce the dimension Radio Map.

[0061 ] 其它步骤及参数与具体实施方式一或二相同。 [0061] The other steps and parameters DETAILED embodiment one or two.

[0062] 具体实施方式四:本实施方式与具体实施方式一至三之一不同的是:步骤五中类别匹配的实现,其核心流程分两步完成: [0062] DETAILED DESCRIPTION IV: with one embodiment of the present embodiment DETAILED one to three different embodiments are: to achieve category matching step 5, which core processes two steps:

[0063] 第一步,寻找未标记RSS的类别属性,由下式完成类别属性标记: [0063] The first step to find the unmarked category RSS property, marked by the completion of the following formula category properties:

Figure CN103648106AD00081

[0065] 第二步:对RSS进行门限检测:通过计算并判定广义符号值与门限值的关系,从而实现Radio Map及类别标记数据的更新,广义符号值及聚类中心的更新分别下两式完成: [0065] Step Two: an RSS threshold detection: the relationship is determined by calculation and generalized symbol values ​​with a threshold value, thereby achieving updating Radio Map classes and marker data, updates generalized symbol values ​​and at two cluster centers are complete formula:

Figure CN103648106AD00082

[0068] 其中,所述门限值VT=0.9Ν。 [0068] wherein, the threshold value VT = 0.9Ν.

[0069] 其它步骤及参数与具体实施方式一至三之一相同。 [0069] The other steps and parameters DETAILED one embodiment one to three.

[0070] 具体实施方式五:基于类别匹配的半监督流形学习的WiFi在线阶段室内定位方法通过下述步骤实现: [0070] DETAILED DESCRIPTION five: a WiFi indoor positioning online phase semi-supervised manifold matching categories based learning achieved by the following steps:

[0071] 一、在线测试RSS; [0071] First, the online test RSS;

[0072] 二、将得到的待定位点的RSS采用特征变换矩阵降维变换得到RSS* ; [0072] Second, the anchor point will be obtained using the RSS feature dimensionality reduction transformation matrix transform RSS *;

[0073] 三、采用KNN算法对RSS*与Radio Map*匹配位,对待定位点的具体位置坐标进行预测并进行在线数据的更新,其实现过程为: [0073] Third, the use of KNN algorithm with specific location coordinates RSS * Radio Map * Match bit treat setpoint is predicted and updated online data, the implementation process is:

[0074] (I)在线阶段,测试点处接收的RSS= [AP1, AP2,…,APn],与特征变换矩阵V'相乘,从而得出降维后的RSS' =[AP1; AP2,…APd],其中d表示本征维数; [0074] (I) online phase, at the test point the received RSS = [AP1, AP2, ..., APn], wherein the transform matrix V 'is multiplied, so as to arrive after the dimensionality reduction RSS' = [AP1; AP2, ... APd], where d represents the intrinsic dimension;

[0075] (2)采用KNN算法实现RSS^与Radio Map*的匹配,采用与RSS^最近的K个参考点的坐标的平均值作为测试点(X' ,1'),其表达式为: [0075] (2) with the KNN algorithm RSS ^ Radio Map * match, using the average value of the coordinates of the nearest reference point K RSS ^ as a test point (X ', 1'), which is expressed as:

Figure CN103648106AD00083

[0077] 式中,(X' , y')为测试点预测的坐标,(Xi, y)为第i个近邻点的坐标,K为KNN算法中近邻的数目; [0077] In the formula, (X ', y') coordinates of the test point predicted, (Xi, y) coordinates of the i-th neighboring point, K is the number of neighbor KNN algorithm;

[0078] 四、用户定位终端定位数据库更新,即完成了基于类别匹配的半监督流形学习的WiFi在线阶段室内定位方法; [0078] Fourth, the user terminal location database updates positioning, to complete the online phase WiFi indoor positioning method for semi-supervised learning manifold based matches category;

[0079] 所述离线阶段在服务器上完成;在线阶段在定位终端上完成。 [0079] The offline phase performed on the server; online phase is completed on the positioning terminal. [0080] 具体实施方式六:本实施方式与具体实施方式五不同的是:步骤三中对待定位点的具体位置坐标进行预测并进行在线数据的更新由权利要求1中所述的离线数据库和在线数据库实现: [0080] DETAILED DESCRIPTION six: five different embodiments of the present embodiment and the exemplary embodiment is: to treat a specific anchor point predicted position coordinates and three step data is updated online and off-line database 1. The claims database implementation:

[0081] 离线定位数据库的实现方式为:将定位用户本次定位测得的RSS作为未标记数据采用类别匹配方式加入到Radio Map中,并在移动终端上实现对本地在线匹配定位数据库的更新,实现动态的更新本地的数据,从而实现离线数据库定位方式; [0081] The implementation of the offline location database is: positioning a user that the positioning of the measured RSS untagged data using the category matching mode is added to the Radio Map in and implement local matching online update location databases on the mobile terminal, achieve dynamic update the local data, thereby achieving targeting offline database;

[0082] 在线数据库的实现方式为:用户在线定位完成后,将用户本次在线测得的RSS值上传至在线定位数据库所在的服务器,并在服务器端将在线定位数据库进行更新将将在线定位数据传回上传RSS数据的定位终端。 Implementation [0082] The online database is: users online positioning Upon completion, the upload user that the line measured RSS value to an online location server database is located, and the online location database is updated to the line position data in the server returns data uploaded RSS positioning terminal.

[0083] 其它步骤及参数与具体实施方式一至五之一相同。 [0083] The other steps and parameters DETAILED one embodiment one to five.

[0084] 仿真实验: [0084] simulation:

[0085] 一、结合附图3对本仿真实验做出详细说明:图示为哈尔滨工业大学科学园2A栋12层的平面图示意,基于WiFi的室内定位系统就是基于该实验环境下建立。 [0085] First, in conjunction with the accompanying drawings 3 to make detailed description of the simulation experiment: Science illustrated as Harbin Institute of Technology Park, Building 12 layer 2A a schematic plan view, WiFi-based positioning system is the indoor environment based on this establishment. 在实验环境中,总共布置27个AP,AP布置的位置为蓝色无线发射信号形状标记所在处。 In the experimental environment, a total of 27 are arranged AP, AP shape disposed at a location where the blue marker radio transmit signals. AP离房间地面高度为2米。 AP from the floor of the room height of 2 meters. 在离线阶段,在联想V450笔记本上安装NetStumbler软件,在所有参考点的四个不同的方位上连续采样记录AP的100个RSS值,以及AP的相关信息。 In the offline stage, NetStumbler software installed on the Lenovo V450 notebook, continuous sampling records information about the AP 100 RSS values, as well as the AP in four different directions all reference points. 将所有的采样点的物理坐标及相应的物理坐标及RSS值存储为定位过程所调用的数据,建立Radio Map。 The physical coordinates all the sample points and the coordinates and the corresponding physical RSS value is stored as the positioning data of the called procedure, the establishment of Radio Map. 在实验环境共有900个参考点,其采样密度为0.5米X0.5米,如附图4所示。 In a total of 900 experimental environment point of reference, the sampling density is 0.5 m X0.5 m, as illustrated in Figure 4. Radio Map作为CM-SDE算法的输入参数及本征维数估计算法的输入数据。 Radio Map which algorithm input data as an input parameter estimation algorithm and CM-SDE present intrinsic dimensionality. [0086] 二、Radio Map的本征维数的获取通过下述步骤实现: [0086] Second, the intrinsic dimension Radio Map which acquisition achieved by the following steps:

[0087] 本征维数是对于高维数据进行本征空间维数及空间重建所需最小的独立变量的个数。 [0087] The intrinsic dimension is the eigenspace dimensions and spatial reconstruction of the required minimum number of independent variables with respect to high-dimensional data. 在具体实际计算中,由于高维数据的本征并不明显,通常不是寻求得到确切的本征维数,而是寻求估计本征维数的可信取值。 In particular the actual calculation, due to the intrinsic high-dimensional data is not obvious, it is not usually seek to get the exact dimensions of the intrinsic, but seek credible estimate of intrinsic value dimension. 具体的说,给定一个来自高维空间的样本,本征维数估计算法的中心任务和重要内容就是通过这些样本数据来确定这个高维结构的本征维数。 Specifically, given a sample from a high-dimensional space, the intrinsic dimension estimate the central task and an important part of the algorithm is to determine the intrinsic dimension of the high-dimensional structure of these sample data.

[0088] Radio Map的本征维数的估计是CM-SDE算法的重要输入参数,这关系到降维的结果是否能够代表Radio Map的高维空间的特征,因此准确有效的本征维数的估计至关重要。 [0088] The estimate of Radio Map of the intrinsic dimension is an important input parameter CM-SDE algorithm, which is related to whether dimensionality reduction results can represent the characteristics of high-dimensional space of the Radio Map, so accurate and effective number of intrinsic dimension estimate is essential. 目前,常用本征维数估计算法分为两类:局部估计与全局估计。 At present, the common intrinsic dimension estimation algorithm is divided into two categories: partial estimation and global. 采用全局算法估计对RadioMap的本征维数进行估计,并作为CM-SDE算法的输入变量。 Using the estimated global logarithmic algorithm RadioMap intrinsic dimension is estimated, as input variables and CM-SDE algorithm. 本实验中采用测地线最小生成树算法(Geodesic Minimum Spanning Tree, GMST)对Radio Map 的本征维数进行估计。 Intrinsic number of the Radio Map dimensions used in the experiment is estimated geodesic minimum spanning tree algorithm (Geodesic Minimum Spanning Tree, GMST).

[0089] 下面对GMST算法的理论进行分析。 [0089] The following theoretical GMST algorithm is analyzed.

[0090] 测地线最小生成树(GMST)估计是基于测地线最小生成树的长度函数依赖于本征维数d。 [0090] geodesic minimum spanning tree (GMST) estimated to be the minimum spanning tree of the geodesic length function depends on the intrinsic dimension d based. GMST是指定义在数据集X上的近邻曲线的最小生成树。 Refers to the minimum spanning tree is defined GMST neighbor graph on the X data set. GMST的长度函数L(X)是在测地线最小生成树中所有边缘对应的欧氏距离之和。 GMST the length function L (X) is a geodesic minimum spanning tree corresponding to all of the edges and the Euclidean distance.

[0091] GMST估计在数据集X上构造一条近邻曲线G,其中,在X内每一个数据点Xi都和它的k个近邻"相连接。测地线最小生成树T定义为X上的最小曲线,它具有长度 [0091] GMST configured to estimate a curve G in the neighborhood data set X, wherein X in each of the data points Xi are the k and its neighbors "are connected. Geodesic minimum spanning tree T is defined as the smallest X curve, having a length

[0092] [0092]

Figure CN103648106AD00091

[0093] 其中,T是曲线G的所有子树集合,e是树T的一个边缘,ge是边缘e对应的欧氏距离,其计算公式为式(2)所示。 [0093] where, T is the curve G of all subtrees set, e is an edge of the tree T, GE edge corresponding to the Euclidean distance e, which is calculated as shown in formula (2).

[0094] [0094]

Figure CN103648106AD00101

[0095] 在GMST估计中,一些子集Jc=X由各种大小m组成,并且子集A的GMST的长度 Length [0095] In GMST estimation, some subset of the Jc = X m by the composition of various sizes, and a subset A of GMST

L(A)也需要计算。 L (A) can be calculated. 理论上,f;是线性的,从而可以由y=ax+b这种形式的函数来估计, Theoretically, F; is linear, which can be estimated by y = ax + b of this function of the form,

维数的估计。 Estimates of the number of dimensions. 由GMST算法给出本征维数d的表达式为式(3)所示。 GMST algorithm is given by the intrinsic dimension d expression represented by the formula (3). 本征维数d是CM-SDE算法的另一个重要的输入参数。 Intrinsic dimension d is another CM-SDE algorithm important input parameter.

Figure CN103648106AD00102

[0097] 三、CM-SDE是一种半监督流形学习算法,在CM-SDE算法实现的过程中需要对所有的参考点进行类标记。 [0097] III, CM-SDE manifold is a semi-supervised learning algorithms, class labels required for all reference points during the CM-SDE algorithm implementation. 考虑到目前的WiFi室内定位环境中,参考点数目将近1000点,并没有人为地对所有的参考点进行类标记,而是采用一定的分类算法对参考点的类别进行标记。 Given the current WiFi indoor positioning environment, the number of reference points nearly 1000, and performed without human class labels of all reference points, instead of using a certain category classification algorithm for marking reference points.

[0098] 聚类的目标是将数据集X= Ix1, X2,…,xj划分为c类且各类数据之间互不相关。 [0098] The goal is to cluster the data set X = Ix1, X2, ..., xj and divided into class c uncorrelated between different data. 基本的聚类算法按如下步骤实现: The basic clustering algorithm as follows:

[0099] (I)生成c个聚类中心,记为Vi, i=l,2,…,C。 [0099] (I) c generated cluster centers, referred to as Vi, i = l, 2, ..., C.

[0100] (2)将数据集X=U1, X2,…,xn}的每个元素归类,采用最近邻(Nearest Neighbor)算法判定元素的归属关系,其等价表达式为: [0100] (2) The data set X = U1, X2, ..., xn} of each collation element, the nearest neighbor (Nearest Neighbor) method determines the belongingness of the element, which expression is equivalent to:

[0101 ] [0101]

Figure CN103648106AD00103

[0102] 式(4)中,Xi为第i个数据点,Gi为第i类构成的邻接关系图。 In [0102] the formula (4), Xi is the i-th data point, Gi is the i-type configuration of FIG abutting relationship. D(Xi,Vj)表示计算Xi与'之间的欧式距离。 D (Xi, Vj) represents the Euclidean distance between the calculated and Xi & apos.

[0103] (3)聚类中心的更新,对于第i类的聚类中心更新为: [0103] (3) updating the cluster centers for the cluster centers class i is updated to:

[0104] [0104]

Figure CN103648106AD00104

[0105] 式(5)中,II表示计算某类中元素的个数。 In [0105] formula (5), II represents a certain number of computing elements.

[0106] (4)收敛性校验及迭代 [0106] (4) Iteration convergence and verification

[0107] 若满足以下四种情况下的收敛性条件之一,则迭代停止,否则重复执行(2)~(3),直到迭代收敛或者达到最大执行次数。 [0107] If the convergence conditions in one of the following four conditions is satisfied, the iteration is stopped, otherwise repeatedly performed (2) to (3), converges or until the maximum number of iterations performed. 四种收敛性判定条件为: Four kinds of convergence determination conditions are:

[0108] 条件一:聚类中心不变; [0108] Condition a: same cluster centers;

[0109] 条件二:每个聚类的元素不变; [0109] The second condition: the same element of each cluster;

[0110] 条件三:聚类中心变化收敛于半径ε内; [0110] Three conditions: cluster center converge on changes in the radius of the [epsilon];

[0111] 条件四:聚类元素变化收敛于半径ε内。 [0111] Condition IV: element changes fall within the cluster radius ε.

[0112] 总的来说,上述收敛性判定条件可以表述为下式: [0112] In general, the above-described convergence determination condition can be expressed as the formula:

[0113] [0113]

Figure CN103648106AD00105

[0114] 其中,Vi'表示更新后的第i类的聚类中心。 [0114] wherein, Vi 'represents the center of the cluster of class i updated.

[0115] 从上述聚类算法的基本实现方式分析,用于判定元素的归属关系的算法构成了聚类的核心。 [0115] from the basic implementation described above clustering analysis algorithm for determination of attribution of the elements constituting the core cluster. 不同类型的聚类算法提出不同的归类指标,一般将该函数称为损耗函数(LossFunction)。 Different types of clustering algorithms proposed classification of different indicators, this function is generally referred to as the loss function (LossFunction). 基本的聚类方法中采用欧式距离作为损耗函数。 The basic clustering approach using Euclidean distance as the loss function. 本专利采用KFCM算法对RadioMap进行类别分析得出聚类中心和Radio Map的类别标记。 This patent uses KFCM algorithm RadioMap category analysis results for labeled category cluster centers and the Radio Map. KFCM算法的理论分析如下: Theory KFCM algorithm is as follows:

[0116]引入核函数的模糊c均值聚类的目标是将原始的数据集所在空间变换至无穷维的希尔伯特空间(Hilbert Space),再对变换后的空间作相应的聚类分析。 [0116] Fuzzy c-means clustering objective function is introduced into the core space of the original data set where conversion to infinite dimensional Hilbert space (Hilbert Space), and then the transformed space for the corresponding cluster analysis. 通过核函数的变换,将原始数据之间的类别特征进一步变换后更易于表述和区分。 By the transformation kernel function, the class distinction between the original data and further transformed representation is easier to distinguish. 基于核函数的模糊c均值聚类算法的目标函数为: Based on the objective function fuzzy c-means clustering algorithm kernel function is:

[0117] [0117]

Figure CN103648106AD00111

[0118] 式(7)中,Φ (xk)馮分别表示在希尔伯特空间下的数据集及相应的聚类中心。 In [0118] formula (7), Φ (xk) Feng represent the data set in the Hilbert space and corresponding cluster centers. 通过推导可以得出KFCM算法的解表述为: KFCM algorithm can be drawn through the derivation of the solution expressed as:

[0119] [0119]

Figure CN103648106AD00112

[0120] KFCM的解的关键在于计算希尔伯特空间的损耗函数或者相似度函数。 [0120] Solutions KFCM critical loss function is to calculate the Hilbert space or similarity function. 在本文中考虑引入高斯核函数(Gaussian Kernel Function)的FCM (Fuzzy C-Means)算法的理论分析及其实现。 Theoretical Analysis and Realization Algorithm incorporated Gaussian kernel (Gaussian Kernel Function) The FCM (Fuzzy C-Means) contemplated herein. 高斯核函数如式(9)所示。 Gaussian kernel equation (9).

[0121] [0121]

Figure CN103648106AD00113

[0122] 希尔伯特空间中,由式表述其相应的损耗函数,该式进一步表述为式(10)。 [0122] Hilbert space, expressed by the formula and the corresponding loss of its function, the formula is further expressed by Equation (10).

[0123] [0123]

Figure CN103648106AD00114

[0124] 式(10)中,〈.,.>表示计算相应式的核函数值。 It shows a calculation formula corresponding to the kernel function value of [0124] Formula (10), <.,.>. 而实际上,无穷空间的变换不存在,因此,对式(10)进一步简化为式(11)所示。 In fact, transformation infinite space does not exist, therefore, of the formula (10) is further simplified as shown in formula (11).

[0125] [0125]

Figure CN103648106AD00115

[0126] 式(11)的全展开式为式(12)所示。 [0126] Formula (11) Expansions of Formula (12) shown in FIG.

Figure CN103648106AD00121

[0128] 式(12)中,<Φ (xk), Φ (Xj) >由高斯核函数计算,即: [0128] Formula (12), <Φ (xk), Φ (Xj)> calculated by the Gaussian kernel, namely:

[0129] [0129]

Figure CN103648106AD00122

[0130] 在算法实现中,不是随机生成聚类中心,而是从数据集X=Ix1, X2,…,χη}中随机选择c个元素作为聚类中心,构成集合Y= Iy1,…,y。 [0130] In the algorithm, the cluster centers are not randomly generated, but from the data set X = Ix1, X2, ..., χη} randomly selected as a cluster center c elements, the configuration set Y = Iy1, ..., y . }。 }. 因此,初始化的损耗函数值计算如式所示。 Thus, the loss function value is calculated as shown in Formula initialized.

[0131] [0131]

Figure CN103648106AD00123

[0132] 四、运用CM-SDE算法实现对Radio Map进行降维并获取特征权值矩阵过程通过下述步骤实现: [0132] Fourth, the use of CM-SDE Radio Map algorithm to reduce the dimension and weight acquisition value matrix process characterized by the following steps implemented:

[0133] CM-SDE算法是基于标记数据与未标记数据的类间散度及类内散度最大化的一种流形学习算法。 [0133] CM-SDE algorithm is based on the divergence between class labels and untagged data and maximizing a class scatter manifold learning algorithms. 在对CM-SDE算法进行理论分析之前对CM-SDE算法给定的输入数据做如下 CM-SDE as follows for a given algorithm input data prior to CM-SDE algorithm theory analysis

说明:输入高维数据点匕}= e Rn,数据点Xi的类标记为Ji e {I, 2,…,P},其中P表示将高 Explanation: high-dimensional data points dagger} = e Rn, the data points marked with a class of Xi Ji e {I, 2, ..., P}, where P represents a high

维数据划分为P个子流形,即将输入的高维数据分成P类,记P类的聚类中心为V= Iv1, V2,…,VpI。 Dimensional data is divided into P sub-manifold, forthcoming high-dimensional input data into classes P, P type referred cluster center is V = Iv1, V2, ..., VpI. 将输入的高维数据表示成矩阵的形式:X=[Xl,X2,…,xj e Rnxm0从矩阵表示的形式来看,矩阵中的列代表一个高维数据点。 High-dimensional input data represented as a matrix: X = [Xl, X2, ..., xj e Rnxm0 from the point of view represented in the form of a matrix, the matrix columns represent a high-dimensional data points.

[0134] 对于包含未标记数据的RadioMapul,其中所有的未标记数据Xu= [xul, xu2,…,xj e Rnxk进行类别匹配,同时已标记数据记为X1=[Xll,X12,…,X1J GRn'对于Xu中的所有数据进行有序类别匹配。 [0134] For RadioMapul contains unmarked data in which all of the unlabeled data Xu = [xul, xu2, ..., xj e Rnxk undergo class matches, while marked data referred to as X1 = [Xll, X12, ..., X1J GRn 'for all the data in an orderly Xu matching category. 有序的含义是当分配某一个未标记数据后,会影响相应类的聚类中心,因此会对下一未标记数据的类判别会有影响。 When ordered meaning assigned one of unlabeled data, the cluster centers can affect the appropriate class, so will the next data type is determined will be affected unlabeled. 本专利中主要的考虑信号的采集的时间顺序。 Time-sequentially acquired signals major consideration in this patent. 假定Xu是按时间顺序排列。 Xu is assumed in chronological order. 采用式(15)计算类的归属Xul,并采用式(5)更新相应的聚类中心。 Using the formula (15) is calculated based home Xul, and the use of formula (5) update the corresponding cluster centers. 然后依次将所有的未标记数据进行类别匹配 Followed by all the data categories matching unlabeled

[0135] [0135]

Figure CN103648106AD00124

[0136] CM-SDE算法的目标函数为: [0136] The objective function CM-SDE algorithm is as follows:

[0137] [0137]

Figure CN103648106AD00125

[0138] 式(16)中Sw、Sb、St分别表示类内散度、类间散度及总散度可以由式(17)计算:[0139] [0138] Formula (16) Sw, Sb, St within-class scatter respectively, and the total divergence between classes divergence may be calculated from the formula (17): [0139]

Figure CN103648106AD00131

[0140] 式(17)中, In [0140] formula (17),

Figure CN103648106AD00132

为第i类的均值,Ii为第i类的釆样点的数目; Are mean of class i, Ii is the number of samples preclude the class i;

Figure CN103648106AD00133

为全体釆样点的均值,N为釆样点的数目。 Are mean preclude all samples, N is the number of samples Bian.

[0141] 对于式(16)所示的目标函数同样可以表示局部鉴别嵌入流形学习算法的目标函数形式,其表达式为: [0141] For the formula (16) the objective function may also be represented by a partial embedding object discrimination function forms the manifold learning algorithm, and its expression is:

[0142] [0142]

Figure CN103648106AD00134

[0143] 式(18)中,Wij表示同类数据间的权重分配,Wi/表示不同类数据间的权重分配,分别表示为Wnxn和Wnxn'。 In [0143] formula (18), Wij denotes a weight between similar data reallocation, Wi / represents the weight between the different types of data reallocation, and are represented Wnxn Wnxn '. 权重计算过程由两步完成。 The weight calculation process is done in two steps. 第一步:构造邻域图。 Step: Construction of the neighborhood graph. 根据高维数据点的类标记信息及其近邻关系构造无方向图G及G,。 The marker information and the neighbor relationship type high-dimensional data structure points undirected graph G, and G ,. 其中近邻关系是采用KNN算法给出的准则,即选择数据点最近的K个点作为其邻居,G表示当Xi与xj的类标记信息yi=yj时且X1、χj互为K近邻关系;G'示当Xi与χj的类标记信息Yi≠yj时且X1、Xj互为K近邻关系。 Wherein the neighborhood relation is given using KNN algorithm criterion, i.e., to select the nearest data point as its neighbor points K, G represents a class when Xi and xj flag information when yi = yj and X1, χj mutual neighbor relation K; G 'illustrates when Xi and Yi χj class mark information and X1, Xj mutual neighbor relation when K ≠ yj. 第二步:计算权值矩阵。 Step two: calculate the weight matrix. 根据第一步构造的邻接图采用类高斯函数进行权值矩阵的计算。 It is calculated according to the weight matrix constructed adjacent to the first step of FIG using Gaussian functions. 其表达式(19)、(20)为所示。 Its expression (19), (20) as shown in FIG. 公式中表示近邻点Xi与xj之间的权值,||X1-Xj ||2为近邻点.与xj之间的距离,采用矩阵方式计算距离,t为权值归一化参数,U、L分别表示未标记和已标记的采样点的数目。 Represents the formula weight between neighbor points Xi and xj, || X1-Xj || 2 is the neighbor point and the distance between xj, matrix calculation using a distance, t values ​​were normalized to the right parameter, U, L represent the number of labeled and unlabeled sample points. 根据分析可以知道,Wnxn和Wnxn'可以由三部分构成,分别是:已标记数据与已标记数据之间的权重、已标记数据与未标记数据之间的权重及未标 The analysis can know, Wnxn and Wnxn 'may be constituted by three parts, namely: the data marked with marked weight between the weight data, marked weight between the weight data and the unlabeled data not labeled and

记数据与未标记数据之间的权重,分别表示为 Right between the data referred to unlabeled data weight are expressed as

Figure CN103648106AD00135

[0144] [0144]

Figure CN103648106AD00136
Figure CN103648106AD00141

[0146]由上述计算公式及矩阵的性质可以得:WHz、Wlu=Wuxll、 [0146] can be obtained from the above formula and properties of matrix: WHz, Wlu = Wuxll,

Wll =1 = 。 Wll = 1 =. 由此可以推导出Wnxn和Wnxn'表示为分块矩阵的形式,如式(21)所示。 It can be deduced Wnxn and Wnxn 'block represented as a matrix form, equation (21) shown in FIG.

Figure CN103648106AD00142

[0148] 根据矩阵的计算式: [0148] The matrix calculation formula:

Figure CN103648106AD00143

,计算式表示为矩阵A的矩阵的计算方法,计算式 Calculating method for calculating the matrix equation represents a matrix A, calculation formula

给出的方法与矩阵的迹的计算式一致,即:| |A| |2=tr(AAT)。 The method of calculation formula given trace of the matrix is ​​consistent, that is: | | A | | 2 = tr (AAT). 由此式(18)可以表示为矩阵的迹的计算方式: Thus formula (18) can be expressed as a trace of a matrix is ​​calculated:

[0149] [0149]

Figure CN103648106AD00144

[0150] 式(22)可以简化为: [0150] Formula (22) can be simplified to:

[0151] [0151]

Figure CN103648106AD00145

[0152]由矩阵迹的计算的标量性质及权值元素均为实数,可以将式(23)简化为: [0152] scalar nature and weight matrix calculation trace elements are real numbers, may be of formula (23) simplifies to:

[0153] [0153]

Figure CN103648106AD00146

[0154] 根据简单的数学关系,可以将式(24)简化为: [0154] According to a simple mathematical relationship can be simplified to formula (24) as:

[0155] J(V)=2tr{VT[X(D/ -Ψ NXN)XT]V} (25) [0155] J (V) = 2tr {VT [X (D / -Ψ NXN) XT] V} (25)

[0156] 式(25)中:X为输入数据,λ和V为特征值与特征向量,W和Wi分别为G及Gi对应的权值矩阵,D及D'均为对角阵,其对角元素可以由式(26)表示。 [0156] Formula (25): X is input data, and V [lambda] is the eigenvalue and eigenvector, W, and Wi is the weight G and Gi, respectively, corresponding to the values ​​of the matrix, D and D 'are diagonal, its diagonal elements may be represented by the formula (26).

[0157] [0157]

Figure CN103648106AD00147

[0158] 根据式(25)的推导方式,同理可以将式(18)中的约束条件写成如式(25)相似的形式,由此,可以将(18)表示为如下形式: [0158] The derivation of Formula (25), empathy may be constraints in the formula (18) written in a similar form as the formula (25), whereby, (18) can be expressed as follows:

[0159] [0159]

Figure CN103648106AD00151

[0160] 对式(27)应用拉格朗日(Lagrange)乘数法,可以得出式(28)所示: [0160] The formula (27) Lagrange (LAGRANGE) multiplier method, can be obtained by formula (28):

[0161] [0161]

Figure CN103648106AD00152

[0162] 对式(28)进行广义特征值分解,得出其特征值分解的特征值及特征向量,表示为:A=LX1, λ 2,…,λ η]τ,其对应的特征向量为:V= [V1, V2,…,Vn] T。 [0162] The formula (28) is generalized eigenvalue decomposition, eigenvalues ​​and eigenvectors obtained which eigenvalue decomposition, is expressed as: A = LX1, λ 2, ..., λ η] τ, feature vectors corresponding to : V = [V1, V2, ..., Vn] T. 取前d个最大的特征值对应的特征向量构成变换矩阵V= [Vl,V2,…,vd]。 D take the first largest eigenvalue of the eigenvector corresponding to the configuration of the transform matrix V = [Vl, V2, ..., vd]. 由CM-SDE算法的输出数据变换方法可以得出,降维后数据为: It can be derived from the output of the data conversion method of CM-SDE algorithm, the data dimension reduction:

[0163] [0163]

Figure CN103648106AD00153

[0164] 式(29)中,Zi表示输入高维数据点Xi变换后的低维输出数据。 In [0164] formula (29), Zi represents the input of the low-dimensional output data points Xi transform high-dimensional data. 从本专利中发明内容给出的离线阶段的实施步骤为:第一步先采用CM-SDE算法对所有参考点Radio Map进行降维处理,得到相应的参考点的降维后的Radio Map,即作为在线阶段的匹配定位数据库(RadioMap*)ο第二步对添加未标记数据的Radio Map,即RadioMapul进行降维处理,得到特征变换矩阵V'。 Step embodiment of the invention the content from the offline phase in this patent are given: the first step using CM-SDE dimensionality reduction algorithm for all the reference points Radio Map, Radio obtained after the dimension reduction of the corresponding reference point Map, i.e., online database as a matching positioning stage (RadioMap *) ο a second step of adding unlabeled data Radio Map, i.e. reduce the dimension RadioMapul, wherein the transformation matrix to obtain V '. 由此可以建立离线阶段所需要数据库:RadioMap*和V'。 Whereby the database can be established offline phase required: RadioMap * and V '.

[0165] 五、由不同的用户在线阶段定位阶段获得的RSS是未标记类别属性的,其加入Radio Map,并构成Radio Mapul的过程称为类别。 [0165] V., obtained from different stages online user positioning stage RSS unlabeled category attribute, added Radio Map, and constitute a process referred to as Radio Mapul category. 其实现方法如下所述: This is accomplished as follows:

[0166] 将不同用户在线阶段定位阶段测试得到的未标记RSS,采用类别匹配的方式加入至已有Radio Map中,得到相应的包含未标记信号覆盖图RadioMapul ;通过类别匹配方法增加Radio Map的数据量,进而提高Radio Map的密度,为CM-SDE算法提供新的降维数据,同时可以更新聚类中心,为CM-SDE算法提供新的类别数据。 [0166] Different users online stage of the positioning phase of the test to obtain unlabeled RSS, using matching methods category added to an existing Radio Map give the corresponding unlabelled coverage map comprises RadioMapul; increase in data category by the Radio Map matching method amount, thereby increasing the density of Radio Map, providing a new dimension reduction data for the CM-SDE algorithm, and can update the cluster centers, provide a new category of data as CM-SDE algorithms. 类别匹配方法分为两步,其实现过程如下所述: Category matching method in two steps, the realization process is as follows:

[0167] 第一步,寻找未标记RSS的类别属性。 [0167] The first step to find the unmarked category attribute of RSS. 记一组未标记RSS为RSSi,与步骤三中的聚类中心进行匹配,由式(4)完成RSSi的类别标记。 Referred to a group of unlabeled RSS matching cluster center in three RSSI, and the step, by the formula (4) to complete the RSSI category tags.

[0168] 第二步:对RSS进行门限检测。 [0168] The second step: to RSS threshold detection. 对于聚类中心vi表示为Vi=Gil, vi2,…,viN) ,N为室内定位系统中AP的个数。 For cluster centers vi is expressed as Vi = Gil, vi2, ..., viN), N is the number of the AP indoor positioning system. RSSi表示为RSSiKRSS^ RSSi2,…,RSSiN)。 RSSi expressed as RSSiKRSS ^ RSSi2, ..., RSSiN). 计算下式所定义的广义符号值: Symbol value calculated by the following generalized formula defined:

[0169] [0169]

Figure CN103648106AD00154

[0170] 其中,sgn(.)定义为: (.) [0170] where, sgn is defined as:

[0171] [0171]

Figure CN103648106AD00155

[0172] 当Si大于设定门限值时,则将RSSi加入Radio Map中,并更新聚类中心,否则舍弃RSSi,不加入Radio Map中。 [0172] When Si is greater than the set threshold, then the RSSi join Radio Map and update cluster centers, otherwise reject RSSi, Radio Map not join in. 在本专利中,门限值VT=0.9N。 In this patent, the threshold value VT = 0.9N. 聚类中心的更新由式(5)完成: Updated cluster center is completed by formula (5):

[0173] 六、基于CM-SDE算法的WiFi室内定位方法的离线数据库实现方式: [0173] Six offline database WiFi indoor positioning method based CM-SDE algorithms are implemented:

[0174] 离线数据库方式由三部分构成。 [0174] offline database embodiment consists of three parts. 第一,所有参考点的Radio Map的建立,并采用CM-SDE算法得到RadioMap'第二,再随机采样U点未标记数据并添加入原有的Radio Map中,并用CM-SDE算法得到V',并将形成的定位数据库下载(存储)到定位的移动终端。 First, all the Radio Map establish reference points, and the use of CM-SDE algorithm RadioMap 'second, then U randomly sampled data points and added to the unlabeled Radio Map which the original, and V obtained by CM-SDE algorithm' , download location database (memory) and the resulting positioning of the mobile terminal. 第三,在线定位实现及Radio Map更新。 Third, implement and Radio Map locating online update. 第三部分的具体实现如下所述: DETAILED said third portion to achieve the following:

[0175] 在线阶段,测试点处接收的RSS= [AP1, AP2,…,APn],η表示室内定位系统布置的AP的数目。 [0175] online phase, at the test point the received RSS = [AP1, AP2, ..., APn], η represents the number of indoor location system arrangement of an AP. 将RSS与特征变换矩阵V'相乘,从而得出降维后的RSS' = [AP1, AP2,…,APJ,其中d表示本征维数。 The RSS and wherein transform matrix V 'is multiplied to obtain the RSS dimensionality reduction' = [AP1, AP2, ..., APJ, wherein d represents the intrinsic dimension. 再采用KNN算法实现RSS ^与RadioMap*的匹配。 Then using KNN algorithm RSS ^ matches the RadioMap *. 采用与RSS^最近的K个参考点的坐标的平均值作为测试点(X',y'),其表达式为: RSS uses the latest average value K ^ coordinate point as the reference test point (X ', y'), which is expressed as:

Figure CN103648106AD00161

[0177] 将定位用户本次定位测得的RSS作为未标记数据采用类别匹配方式加入到RadioMap中,并在移动终端上实现对本地在线匹配定位数据库的更新,实现动态的更新本地的数据,从而实现离线数据库定位方式,附图1所示于类别匹配的半监督流形学习的WiFi室内定位方法在用户定位终端实现; [0177] The positioning of the user that the positioning measured RSS untagged data using the category matching mode added to RadioMap in, and to implement local matching online update location databases on the mobile terminal, dynamically update the local data, thereby offline database targeting methods, the drawings WiFi indoor positioning method of semi-supervised learning manifold to match the category shown implemented in the user terminal 1 is positioned;

[0178] 七、基于CM-SDE算法的WiFi室内定位方法的在线数据库实现方式: [0178] Seven, an online database of WiFi indoor positioning method based on the CM-SDE algorithms are implemented:

[0179] 在线数据库方式由四部分构成。 [0179] Database Online embodiment consists of four parts. 第一,所有参考点的Radio Map的建立,并采用CM-SDE算法得到RadioMap'第二,再随机采样U点未标记数据并添加入原有的Radio Map中,并用CM-SDE算法得到V',并将形成的定位数据库下载(存储)到定位的移动终端。 First, all the Radio Map establish reference points, and the use of CM-SDE algorithm RadioMap 'second, then U randomly sampled data points and added to the unlabeled Radio Map which the original, and V obtained by CM-SDE algorithm' , download location database (memory) and the resulting positioning of the mobile terminal. 第三,在线定位实现及Radio Map更新。 Third, implement and Radio Map locating online update. 第三部分的具体实现如下所述: DETAILED said third portion to achieve the following:

[0180] 在线阶段,测试点处接收的RSS= [AP1, AP2,…,APn],η表示室内定位系统布置的AP的数目。 [0180] online phase, at the test point the received RSS = [AP1, AP2, ..., APn], η represents the number of indoor location system arrangement of an AP. 将RSS与特征变换矩阵V'相乘,从而得出降维后的RSS' = [AP1, AP2,…,APJ,其中d表示本征维数。 The RSS and wherein transform matrix V 'is multiplied to obtain the RSS dimensionality reduction' = [AP1, AP2, ..., APJ, wherein d represents the intrinsic dimension. 再采用KNN算法实现RSS ^与RadioMap*的匹配。 Then using KNN algorithm RSS ^ matches the RadioMap *. 采用与RSS^最近的K个参考点的坐标的平均值作为测试点(X',y'),其表达式为: RSS uses the latest average value K ^ coordinate point as the reference test point (X ', y'), which is expressed as:

Figure CN103648106AD00162

[0182] 第四部分:用户在线定位完成后,将用户本次在线测得的RSS值上传至在线定位数据库所在的服务器,并在服务器端将在线定位数据库进行更新将将在线定位数据传回上传RSS数据的定位终端,即附图2所示的离线阶段在在线定位数据库所在服务器上完成,而在线阶段在定位终端完成。 [0182] Part IV: After the positioning is completed online users, users upload this online measured RSS value to an online location server database resides, and the location database is updated online at the server side return line positioning data upload positioning terminal RSS data, that the drawings shown in the offline phase is completed on-line server where the database is located, and is positioned at the terminal to complete the online phase.

Claims (6)

1.一种基于类别匹配的半监督流形学习的WiFi室内定位方法,其特征在于基于类别匹配的半监督流形学习的WiFi室内定位方法离线阶段定位过程按以下步骤实现: 一、对待定位的室内区域布置AP,使无线信号覆盖待定位的室内区域,完成WiFi网络构建; 在待定位的室内区域规则选取并记录参考点的相应坐标,测量并依次记录参考点接收到的所有AP的RSS信号作为位置特征信息,构建Radio Map,并存储Radio Map ; 二、采用GMST本征维数估计算法对步骤一中构建的Radio Map的本征维数进行分析,得到的本征维数作为CM-SDE算法的输入参数之一,决定Radio Map降维后的维数; 三、采用KFCM算法对Radio Map进行聚类分析,实现建立的Radio Map的类别标记,并作为CM-SDE的输入参数之一,并且提供相应的初始聚类中心及类别标记; 四、步骤二中的本征维数与步骤三中的类别标记作为输入参 CLAIMS 1. A method of locating WiFi indoor semi-supervised learning manifold based matches category, wherein WiFi indoor positioning method of positioning the offline phase matching categories based semi-supervised learning Manifold achieved by the following steps: a treat positioned interior region disposed AP, so that the wireless signal coverage interior area to be located, to complete WiFi network construction; select and corresponding coordinates of the reference points recorded in the interior region of the rule to be located is measured and sequentially recorded all the AP the reference point of the received RSS signal as the position of the feature information constructed Radio Map, and stores the Radio Map; Second, the use GMST intrinsic dimension estimates intrinsic dimension algorithm to step a constructed Radio Map which analyzed the dimension of an intrinsic obtained as CM-SDE one of the input parameters of the algorithm to determine the dimension of the Radio Map dimensionality reduction; Third, the use of Radio Map KFCM algorithm clustering analysis, realize Radio Map category tags established as one of the input parameters of the CM-SDE, and provides a corresponding category marker and initial cluster centers; Fourth, in step two of the intrinsic dimensionality of step three categories labeled as an input variable ,采用CM-SDE算法对步骤一中构建的Radio Map降维,得出相应的降维后的RadioMap'RadioMap*作为在匹配定位数据库用于在线定位阶段; 五、将不同用户在线定位阶段测试得到的未标记RSS,采用类别匹配的方式加入至已有Radio Map中,得到相应的包含未标记信号覆盖图RadioMapul,通过类别匹配更新的聚类中心作为CM-SDE算法中新的类别输入参数; 六、步骤五中的更新的聚类中心作为输入参数,采用CM-SDE对RadioMapul降维得到特征变换矩阵V' , Y'与RadioMap*共同构成在线匹配定位数据库,用于在线阶段定位;其中,所述线阶段定位具体为: 六(一)、在线测试RSS ; 六(二)、采用V将RSS降维为RSS* ; 六(三)、采用KNN算法进行匹配定位输出定位结果; 六(四)、用户定位终端定位数据库更新; 即完成了一种基于类别匹配的半监督流形学习的WiFi室内定位方法的离线阶段实现方式。 Using Radio Map CM-SDE dimension reduction algorithm constructed in step a, after obtained the corresponding dimension reduction RadioMap'RadioMap * as a database for locating the matching line positioning stage; five, different users online positioning stage test result unlabeled RSS, using matching methods category added to an existing Radio Map to obtain the corresponding unlabeled coverage map comprises RadioMapul, by updating the category matching the category of the new cluster center as input parameters CM-SDE algorithm; six , Fifth step of updating the cluster centers as an input parameter, using CM-SDE RadioMapul of dimension reduction transformation matrix obtained wherein V ', Y' and RadioMap * together constitute the matching online location database for locating the online phase; wherein the stage of the positioning of said lines in particular: VI (a), line testing RSS; VI (b), with V the RSS dimension reduction for the RSS *; six (iii), using the KNN algorithm targeting outputting a positioning result; hexa (tetra) , positioning the user terminal location database updates; to complete the offline phase WiFi indoor positioning method based on the category matching semi-supervised learning implementations manifold.
2.根据权利要求1所述的一种基于类别匹配的半监督流形学习的WiFi室内定位方法,其特征在于步骤二中采用GMST本征维数估计算法对步骤一中构建的Radio Map的本征维数进行分析,其计算公式为: ^ntrinsicdim 测地距最小生成树算法1-α 上式中+中的a表示最小生成树的线性拟合表达式y=ax+b的斜率。 The WiFi indoor positioning method of claim 1 manifolds semi-supervised learning based on the category matching claim, wherein the step of using two GMST intrinsic dimension estimation step of the present algorithm constructed in a Radio Map which is Zheng dimension analysis, which is calculated as: ^ ntrinsicdim geodesic distance on the minimum spanning tree algorithm in the formula 1-α + is a minimum spanning tree represents a linear fit of the expression y = ax + b is the slope. 1-α 1-α
3.根据权利要求1所述的一种基于类别匹配的半监督流形学习的WiFi室内定位方法,其特征在于步骤四中生成新的降维后的信号覆盖图RadioMap*与步骤六中的V'表达式为: Radio Map*=V1.XX是需要降维的Radio Map。 The WiFi indoor positioning method of claim 1 manifolds semi-supervised learning based on the category matching claim, wherein the coverage map generated in step four new dimension reduction RadioMap * V and six in step 'expression is: Radio Map * = v1.XX is the need to reduce the dimension Radio Map.
4.根据权利要求1所述的一种基于类别匹配的半监督流形学习的WiFi室内定位方法,其特征在于步骤五中类别匹配的实现,其核心流程分两步完成:第一步,寻找未标记RSS的类别属性,由下式完成类别属性标记: The WiFi indoor positioning method of claim 1 manifolds semi-supervised learning based on the category matching claim, wherein the category matching step 5 to achieve, which core processes two steps: first, to find category attribute of unlabeled RSS, designated by the category attribute complete formula:
Figure CN103648106AC00031
第二步:对RSS进行门限检测:通过计算并判定广义符号值与门限值的关系,从而实现Radio Map及类别标记数据的更新,广义符号值及聚类中心的更新分别下两式完成: Step Two: RSS threshold detection of: determining by calculation and the relationship between generalized symbol values ​​with a threshold value, thereby achieving updating Radio Map classes and marker data, updates generalized symbol values ​​and the cluster centers are two complete formula:
Figure CN103648106AC00032
其中,所述门限值VT=0.9N。 Wherein the threshold value VT = 0.9N.
5.一种基于类别匹配的半监督流形学习的WiFi室内定位方法,其特征在于基于类别匹配的半监督流形学习的WiFi室内定位方法在线阶段定位过程通过下述步骤实现: 一、在线测试RSS ; 二、将得到的待定位点的RSS采用特征变换矩阵降维变换得到RSS* ; 三、采用KNN算法对RSS*与Radio Map*匹配位,对待定位点的具体位置坐标进行预测并进行在线数据的更新,其实现过程为: (1)在线阶段,测试点处接收的RSS= [AP1, AP2,…,APn],与特征变换矩阵V'相乘,从而得出降维后的RSS' =[AP1; AP2,…APd],其中d表示本征维数; (2)采用KNN算法实现RSS^与Radio Map*的匹配,采用与RSS^最近的K个参考点的坐标的平均值作为测试点,1'),其表达式为: A method of indoor positioning WiFi manifold category matching semi-supervised learning based, characterized in that the online phase WiFi indoor positioning method of positioning manifold semi-supervised learning based on the category matching achieved by the following steps: a, online test RSS; two anchor points to be obtained RSS using feature transform matrix dimensionality reduction transform RSS *; Third, using KNN algorithm specific location coordinates RSS * and Radio Map * match bit treat setpoint is predicted, and online update data, which process is implemented as follows: (1) online phase, at the test point the received RSS = [AP1, AP2, ..., APn], wherein the transform matrix V 'is multiplied, so as to arrive after the dimension reduction RSS' = [AP1; AP2, ... APd], where d represents the intrinsic dimension; (2) with the KNN algorithm RSS ^ Radio Map * matches using the average RSS ^ K latest coordinates of reference points as test point, 1 '), which is expressed as:
Figure CN103648106AC00033
式中,U',Y')为测试点预测的坐标,(Xi, Yi)为第i个近邻点的坐标,K为KNN算法中近邻的数目; 四、用户定位终端定位数据库更新,即完成了基于类别匹配的半监督流形学习的WiFi在线阶段室内定位方法。 Wherein, U ', Y') for the test point predicted coordinate, (Xi, Yi) is the i-th coordinate of neighboring points, K is the number of KNN algorithm neighbor; Fourth, the user positioning terminal location database is updated, i.e., complete a WiFi indoor positioning methods online phase matching categories based on semi-supervised manifold learning.
6.根据权利要求5所述的一种基于类别匹配的半监督流形学习的WiFi室内定位方法,其特征在于步骤三中对待定位点的具体位置坐标进行预测并进行在线数据的更新由权利要求I中所述的离线数据库和在线数据库实现: 离线定位数据库的实现方式为:将定位用户本次定位测得的RSS作为未标记数据采用类别匹配方式加入到Radio Map中,并在移动终端上实现对本地在线匹配定位数据库的更新,实现动态的更新本地的数据,从而实现离线数据库定位方式; 在线数据库的实现方式为:用户在线定位完成后,将用户本次在线测得的RSS值上传至在线定位数据库所在的服务器,并在服务器端将在线定位数据库进行更新将将在线定位数据传回上传RSS数据的定位终端。 The WiFi indoor positioning method of claim 5 Manifold semi-supervised learning based on the category matching claim, wherein the step of treating the setpoint position coordinates for three specific prediction data and updated online by the claims I in the offline databases and online databases implemented: the offline location database implementation of: positioning a user that the positioning of the measured RSS untagged data were added to the category matching mode in the Radio Map, and implemented on the mobile terminal local matching online update location databases, dynamic update local data, thereby realizing the offline database targeting; implementation of online databases is: users online positioning Upon completion, the upload user that the line measured RSS values ​​to the online positioning server database resides, and upload positioning terminal server RSS data in the online database is updated to locate the positioning data back online.
CN 201310750528 2013-12-31 2013-12-31 WiFi indoor positioning method for matching categories based on semi-supervised manifold learning CN103648106B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 201310750528 CN103648106B (en) 2013-12-31 2013-12-31 WiFi indoor positioning method for matching categories based on semi-supervised manifold learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 201310750528 CN103648106B (en) 2013-12-31 2013-12-31 WiFi indoor positioning method for matching categories based on semi-supervised manifold learning

Publications (2)

Publication Number Publication Date
CN103648106A true 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
CN 201310750528 CN103648106B (en) 2013-12-31 2013-12-31 WiFi indoor positioning method for matching categories based on semi-supervised manifold learning

Country Status (1)

Country Link
CN (1) CN103648106B (en)

Cited By (6)

* 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

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 (7)

* 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
CN104185275B (en) * 2014-09-10 2017-11-17 北京航空航天大学 A kind of interior locating method based 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

Also Published As

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

Similar Documents

Publication Publication Date Title
Nguyen et al. A kernel-based learning approach to ad hoc sensor network localization
Pan et al. Multidimensional vector regression for accurate and low-cost location estimation in pervasive computing
Ma et al. Semi-supervised clustering algorithm for community structure detection in complex networks
Slama et al. Accurate 3D action recognition using learning on the Grassmann manifold
Dong et al. A calibration-free localization solution for handling signal strength variance
CN101551809A (en) Search method of SAR images classified based on Gauss hybrid model
France et al. Two-way multidimensional scaling: A review
CN101980250A (en) Method for identifying target based on dimension reduction local feature descriptor and hidden conditional random field
CN103278170A (en) Mobile robot cascading map building method based on remarkable scenic spot detection
Zhang et al. Endmember extraction of hyperspectral remote sensing images based on the ant colony optimization (ACO) algorithm
Wang et al. Load profiling and its application to demand response: A review
CN101930537A (en) Method and system for identifying three-dimensional face based on bending invariant related features
CN103139907A (en) Indoor wireless positioning method by utilizing fingerprint technique
Chen et al. Relevance metric learning for person re-identification by exploiting listwise similarities
CN102802260A (en) WLAN indoor positioning method based on matrix correlation
Kokiopoulou et al. Graph-based classification of multiple observation sets
Sun et al. Adaptive localization through transfer learning in indoor wi-fi environment
CN103152823A (en) Wireless indoor positioning method
Scholz Validation of nonlinear PCA
CN103476118A (en) WLAN indoor location fingerprint positioning method used for real-time monitoring
Pan et al. Accurate and low-cost location estimation using kernels
CN103020654A (en) Synthetic aperture radar (SAR) image bionic recognition method based on sample generation and nuclear local feature fusion
Guney et al. Resonant frequency calculation for circular microstrip antennas with a dielectric cover using adaptive network-based fuzzy inference system optimized by various algorithms
Liu et al. DP-FACT: Towards topological mapping and scene recognition with color for omnidirectional camera
Ouyang et al. Multi-robot active sensing of non-stationary Gaussian process-based environmental phenomena

Legal Events

Date Code Title Description
C10 Entry into substantive examination
C14 Grant of patent or utility model