CN110933596B - Fingerprint positioning method based on metric learning - Google Patents

Fingerprint positioning method based on metric learning Download PDF

Info

Publication number
CN110933596B
CN110933596B CN201911228508.6A CN201911228508A CN110933596B CN 110933596 B CN110933596 B CN 110933596B CN 201911228508 A CN201911228508 A CN 201911228508A CN 110933596 B CN110933596 B CN 110933596B
Authority
CN
China
Prior art keywords
data
rss
fingerprint
matrix
rss data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
CN201911228508.6A
Other languages
Chinese (zh)
Other versions
CN110933596A (en
Inventor
马琳
张永亮
王彬
谭学治
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Harbin Institute of Technology
Original Assignee
Harbin Institute of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Harbin Institute of Technology filed Critical Harbin Institute of Technology
Priority to CN201911228508.6A priority Critical patent/CN110933596B/en
Publication of CN110933596A publication Critical patent/CN110933596A/en
Application granted granted Critical
Publication of CN110933596B publication Critical patent/CN110933596B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/021—Services related to particular areas, e.g. point of interest [POI] services, venue services or geofences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/023—Services making use of location information using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30—Services specially adapted for particular environments, situations or purposes
    • H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Collating Specific Patterns (AREA)
  • Length Measuring Devices With Unspecified Measuring Means (AREA)

Abstract

The invention relates to a fingerprint positioning method based on metric learning. The method comprises the following steps: setting a plurality of reference points in a target indoor environment, establishing a Cartesian coordinate system, and generating a two-dimensional Euclidean space corresponding to an indoor physical space; collecting RSS data from n estimation Access Points (AP) on m reference points; adding a position tag to RSS data acquired by each pre-deployed reference point RP to generate original data; preprocessing the original data to obtain a similarity measurement matrix, and establishing a fingerprint map library according to RSS data; and loading the similarity measurement matrix and the fingerprint map library into a KNN algorithm to obtain the estimated position of the user. The invention fully utilizes the mapping information from the physical space to the signal space, reduces the reduction of positioning precision caused by the complex indoor building structure and further improves the precision of indoor positioning.

Description

Fingerprint positioning method based on metric learning
Technical Field
The invention relates to the technical field of signal positioning, in particular to a fingerprint positioning method based on metric learning.
Background
In recent years, the application of Location Based Services (LBS) has led to extensive research on wireless Location technology. Among all positioning technologies, radio fingerprinting technology has proven to be the most effective one. The method has the characteristics of simple and quick calculation, convenient hardware arrangement, stable system and the like, and is widely applied to various scenes. The indoor environment is one of the main scenes of LBS application, and an accurate real-time indoor positioning scheme is urgently needed for people in daily life.
To our knowledge, indoor positioning methods are mainly divided into two main categories: distance-based positioning methods and fingerprint-based positioning methods. The main idea of the distance-based approach is the distance between an Access Point (AP) and a terminal device. Therefore, the propagation model, the arrival angle, the arrival time difference and the like are common means for estimating the terminal position. However, the variety of indoor environments and the problems caused by the complicated internal structure will undoubtedly hinder the practicability of the distance measuring method. The fingerprint-based method is a positioning information retrieval method that positions a terminal by comparing the similarity of the online phase and the offline phase of Received Signal Strength (RSS). Particularly, in the off-line stage, the system collects RSS from the access points at each pre-deployed Reference Point (RP) and constructs a fingerprint database to map RSS distribution and physical spatial layout. In the online stage, the position of the terminal can be estimated by matching the RSS of the unknown position coordinates received in real time with the database. As can be seen from the above, the accuracy of indoor positioning depends critically on the similarity calculation method between different RSS received from the online and offline phases. Therefore, improving the existing fingerprint similarity calculation method becomes one of the important means for improving the fingerprint positioning precision.
Most of the current algorithms related to fingerprint matching do not consider the problem of collecting RSS statistical distributions from different APs at different RP locations. In practical situations, matching algorithms such as KNN or WKNN simply use the euclidean distance to evaluate the similarity between two sets of data. One of the bases for euclidean distance establishment is to require different dimensions of data to be statistically subject to the same distribution. In addition, correlation exists among RSS from different APs in a real environment, so that an actually acquired RSS matrix has great redundancy, and the improvement of fingerprint positioning accuracy is not beneficial. In summary, the conventional location matching method cannot reflect the complex situation from the signal space to the physical space in the actual situation.
Disclosure of Invention
The invention provides a fingerprint positioning method based on metric learning, aiming at solving the problems that the traditional fingerprint positioning algorithm ignores that RSS data from different APs are distributed differently and correlation exists between the data is not considered, and the invention provides the following technical scheme:
a fingerprint positioning method based on metric learning comprises the following steps:
step 1: setting a plurality of reference points RP in a target indoor environment, establishing a Cartesian coordinate system, and generating a two-dimensional Euclidean space corresponding to an indoor physical space; respectively collecting RSS data from n Access Points (AP) on m Reference Points (RP);
step 2: adding a position tag to RSS data acquired by each pre-deployed reference point RP to generate original data;
and step 3: preprocessing the original data to obtain a similarity measurement matrix, and establishing a fingerprint map library according to RSS data;
and 4, step 4: and loading the similarity measurement matrix and the fingerprint map library into a KNN algorithm to obtain the estimated position of the user.
Preferably, the step 1 specifically comprises:
step 1.1: setting a plurality of reference points RP according to actual requirements in a target indoor environment, establishing a Cartesian coordinate system, and generating a two-dimensional Euclidean space corresponding to an indoor physical space; expressing the set reference point in Euclidean space, reference point RPiIs expressed as RPi＝(xi,yi)；
Step 1.2: respectively acquiring RSS data from n Access Points (AP) on m Reference Points (RP) by adopting signal acquisition equipment pre-installed with WIFE signal acquisition software, and setting the acquisition frequency of the signal acquisition equipment pre-installed with the WIFE signal acquisition software to be 2Hz and the acquisition time to be 2 minutes;
preferably, the reference point spacing is set to one meter per RP spacing.
Preferably, the step 2 specifically comprises:
step 2.1: calculating all AP sets in the indoor signal, and setting all AP sets in the ith data as { AP }iThe total number of data is m, and the total AP set is represented by the following formula:
Figure BDA0002302897610000021
step 2.2: sequencing each piece of the measured RSS data to generate an RSS data set, and expressing the generated RSS data set by the following formula:
Figure BDA0002302897610000022
wherein the content of the first and second substances,
Figure BDA0002302897610000023
for RSS data sets, RSSnIs the nth RSS data;
combining all RSSs according to a reference point RP sequence deployed in advance to generate an RSS data matrix, wherein the RSS data matrix is represented by the following formula:
Figure BDA0002302897610000031
RSS(n，m)is an RSS data matrix;
step 2.3: the method comprises the following steps of performing non-demand data cleaning on collected original data column by column, deleting field data which are not needed, and performing non-demand data cleaning according to the following formula:
Figure BDA0002302897610000032
wherein the celliA yes judgment field; c is a valid field set; d is a reserved field;
step 2.4: performing missing value completion on the processed RSS data, classifying the data missing values of the same-class data column by column, judging the reasons causing data missing, performing different processing on the missing values according to different reasons, and performing classification judgment according to the following formula:
Figure BDA0002302897610000033
wherein the RSSmajorThe RSS value with the highest frequency of occurrence in a certain column of data; NaN is a deletion value; n is the total number of data, thd is the set threshold, thd ∈ (0),1)；
And eliminating data which do not conform to the actual situation.
Preferably, the preprocessing is performed on the original data in the step 3, and the obtained similarity measurement matrix specifically includes:
step 3.1: the vector similarity is measured, and the similarity of the two matrixes is represented by the following formula:
Figure BDA0002302897610000034
wherein the content of the first and second substances,
Figure BDA0002302897610000035
is a function of distance;
Figure BDA0002302897610000036
and
Figure BDA0002302897610000037
the ith and jth data vectors; m is an initial value of the measurement matrix and is set as a unit matrix I;
step 3.2: determining a trained objective function, the trained objective function being represented by:
Figure BDA0002302897610000038
wherein i, j → i is the same type of vector i and vector j, and belongs to the category of vector i; mu is a threshold value; xiijlIs a relaxation factor, which is defined as follows:
Figure BDA0002302897610000041
wherein the content of the first and second substances,
Figure BDA0002302897610000042
is a scale scaling function; xiijlIs a relaxation factor.
Step 3.3: the goal of defining the scaling function is to constrain the domain of the objective function to be within a convex range. It is defined as:
Figure BDA0002302897610000043
wherein (R)min,Rmax) Is the scaled convex domain range; λ is the scaling factor.
Step 3.4: determining an extrapolation function and an inner pull function, the inner pull function and the extrapolation function being represented by:
Figure BDA0002302897610000044
Figure BDA0002302897610000045
wherein epsilonpull(M) is a pull-in function,
Figure BDA0002302897610000046
is an extrapolation function;
step 3.5: all constraint functions are determined and are expressed by:
Figure BDA0002302897610000047
Figure BDA0002302897610000048
ξijl≥0
M≥0
Figure BDA0002302897610000049
step 3.6: and solving the objective function by adopting a random gradient descent method to obtain an optimal similarity measurement matrix M.
Preferably, the step 3 of establishing the fingerprint map library according to the RSS data specifically includes:
averaging RSS data from the same RP to eliminate the effect of noise on the data; RSS data from the same RP are averaged by:
Figure BDA00023028976100000410
wherein, numiIs RPiThe number of data pieces of (a);
and averaging all the data, and finally compressing the RSS data collected by m RPs and from n APs collected by each RP into a fingerprint map library matrix of m multiplied by n elements.
Preferably, the step 4 specifically includes:
step 4.1: inputting the similarity measurement matrix M and the fingerprint database into a KNN matching algorithm, and expressing the similarity measurement matrix M and the fingerprint database input KNN matching algorithm process by the following formula:
Figure BDA0002302897610000051
wherein d isiVector distance; deltaDIs a set of distances; phikIs from deltaDK element sets extracted from the data; r is a coordinate set;
Figure BDA0002302897610000052
to estimate the position coordinates;
step 4.2: inputting the measured signal vector to obtain a plurality of measured estimated coordinates, and finally carrying out centroid solving on the estimated coordinates to obtain the coordinates of the position where the user is located.
The invention has the following beneficial effects:
the invention fully utilizes the mapping information from the physical space to the signal space, reduces the reduction of positioning precision caused by the complex indoor building structure and further improves the precision of indoor positioning.
Drawings
FIG. 1 is a schematic diagram of RP reference point establishment;
FIG. 2 is an RSS fingerprint database establishment process;
FIG. 3 is an RSS data set input to an LMNNP model for metric learning;
FIG. 4 is a schematic diagram of an algorithm for LMNNP;
fig. 5 is a schematic flowchart of a fingerprint location algorithm based on metric learning.
Detailed Description
The present invention will be described in detail with reference to specific examples.
The first embodiment is as follows:
as shown in fig. 5, the present invention provides a fingerprint positioning method based on metric learning, which includes the following steps:
step 1: setting a plurality of reference points in a target indoor environment, establishing a Cartesian coordinate system, and generating a two-dimensional Euclidean space corresponding to an indoor physical space; respectively collecting RSS data from n Access Points (AP) on m Reference Points (RP);
a plurality of reference points are set in a target indoor environment according to actual requirements, a proper position is selected as a coordinate origin to establish a Cartesian coordinate system, and a two-dimensional Euclidean space corresponding to an indoor physical space is generated. At the same time, the set reference point is expressed in Euclidean space, e.g. reference point RPiCan be expressed as RPi＝(xi,yi). The reference point spacing is set herein as one meter per RP spacing as is practical, as shown in fig. 1.
And respectively collecting RSS signals from the n APs on preset m reference points. The acquisition equipment adopts signal acquisition equipment pre-installed with WIFI signal acquisition software. In consideration of the related algorithm processing later, it is not desirable that the RSS signal collected at each RP point is too small. Herein, the RSS acquisition frequency of the device is set to 2Hz and the acquisition time is set to 2 minutes.
Step 2: adding a position tag to RSS data acquired by each pre-deployed reference point RP to generate original data;
and adding a position label, namely a two-dimensional coordinate label of the RP point, to the RSS data acquired by each RP. And then the generated original data is subjected to data preprocessing. The raw data is first subjected to non-demand data cleansing, i.e., values of irrelevant fields are deleted. And secondly, missing value completion is carried out on the data, undetected elements are filled, and meanwhile, missing detected elements are subjected to data completion. Then, the abnormal values in the data are removed, and the data which do not meet the requirements of common sense and obvious error data are removed and supplemented. Finally, the data is standardized to make it more suitable for subsequent data applications. And finally, finishing the establishment of the database.
The set of all APs in the indoor signal is computed. Let all APs in the ith data set as { AP }iThe total number of data is m, and the total AP set is:
Figure BDA0002302897610000061
for each piece of RSS data measured, the RSS data belongs to the set { AP }iThe RSS data of (1) are arranged in sequence, and the format of the generated RSS data is as follows:
Figure BDA0002302897610000062
then all RSSs are combined according to the RP sequence, and the format of the generated RSS data matrix is as follows:
Figure BDA0002302897610000063
and (4) performing non-required data cleaning on the acquired original data column by column, namely deleting the field data which is not required. It is represented as follows:
Figure BDA0002302897610000064
wherein the RSS(n，m)Is an RSS data matrix; celliA yes judgment field; c is a valid field set; d is a reserved field.
And 5, performing missing value completion on the RSS data processed in the step one. Firstly, classifying data missing values column by column for the same-class data, judging reasons causing data missing, and carrying out different processing on the missing values according to different reasons. The criteria are as follows:
Figure BDA0002302897610000071
wherein the RSSmajorThe RSS value with the highest frequency of occurrence in a certain column of data; NaN is a deletion value; n is the total number of such data, thd is the set threshold, thd ∈ (0, 1).
And eliminating data obviously not conforming to the actual situation. And traversing the whole data set, and removing and filling the obviously data which do not meet the requirements until all the data meet the actual requirements. As shown in fig. 3.
And step 3: preprocessing the original data to obtain a similarity measurement matrix, and establishing a fingerprint map library according to RSS data;
and the similarity measurement matrix generation process based on measurement learning comprises the steps of calculating original parameters in the LMNNP model according to the original data and substituting the original parameters into the LMNNP model.
And classifying the preprocessed RSS data sets, wherein the RSS data from the same RP is regarded as one type, and the data from different RPs belong to different types. And inputting all data into a designed LMNNP model for training, and finally obtaining a similarity measurement matrix.
The application process of the fingerprint positioning algorithm based on metric learning comprises the following steps: and establishing a fingerprint map library. RSS data from the same RP are averaged to eliminate the effect of noise on the data. This operation is done on all the data, eventually compressing RSS data from n APs collected from m RPs, t pieces per RP, into an mxn element fingerprint map matrix.
And 4, step 4: loading the similarity measurement matrix and the fingerprint map library into a KNN algorithm to obtain an estimated position of the user;
and inputting the fingerprint database of the generated similarity metric matrix M sum into a KNN matching algorithm, wherein the KNN matching algorithm is as follows.
Figure BDA0002302897610000072
Wherein d isiVector distance; deltaDIs a set of distances; phikIs from deltaDK element sets extracted from the data; r is a coordinate set;
Figure BDA0002302897610000073
to estimate the position coordinates.
Step two: and respectively inputting the actually measured signal vectors according to the mode of the step one to obtain the estimated coordinates of the actual measurement for multiple times. And finally, carrying out centroid solution on the estimated coordinates to finally obtain the coordinates of the position where the user is located.
Specific example 2:
the similarity measurement matrix generation process based on measurement learning comprises the following steps:
the method comprises the following steps: vector similarity is measured. The degree of similarity of the two matrices can be defined by the following equation:
Figure BDA0002302897610000081
wherein the content of the first and second substances,
Figure BDA0002302897610000082
is a function of distance;
Figure BDA0002302897610000083
and
Figure BDA0002302897610000084
the ith and jth data vectors; m is a measurement matrix, and the initial value of the measurement matrix is set as an identity matrix I.
Step two: an objective function of the training is determined.
Figure BDA0002302897610000085
Wherein i, j → i indicates that the vector i and the vector j are the same and belong to the category of the vector i; mu is a threshold value; xiijlIs a relaxation factor, which is defined as follows:
Figure BDA0002302897610000086
wherein the content of the first and second substances,
Figure BDA0002302897610000087
is a scale scaling function; xiijlIs a relaxation factor.
Step three: the goal of defining the scaling function is to constrain the domain of the objective function to be within a convex range. It is defined as:
Figure BDA0002302897610000088
wherein (R)min,Rmax) Is the scaled convex domain range; λ is the scaling factor.
Step four: an extrapolation function and an inner pull function are defined. The pull-in function is defined as:
Figure BDA0002302897610000089
the extrapolation function is defined as:
Figure BDA00023028976100000810
in summary, the physical meaning of the objective function is: it is desirable to make the RSS data of the same category as close as possible and make the RSS data of different categories conditionally forward according to the difference of the similarity by continuously updating the similarity metric matrix.
Step five: all constraint functions are defined.
Figure BDA00023028976100000811
Figure BDA0002302897610000091
ξijl≥0
M≥0
Figure BDA0002302897610000092
Step six: and solving the objective function by adopting a random gradient descent method. The final result is the optimal similarity metric matrix M. As shown in fig. 4.
Specific example 3:
the application process of the fingerprint positioning algorithm based on metric learning comprises the following steps:
and establishing a fingerprint map library. RSS data from the same RP are averaged to eliminate the effect of noise on the data. This operation is done on all the data, eventually compressing RSS data from n APs collected from m RPs, t pieces per RP, into an mxn element fingerprint map matrix.
And carrying out homogeneous averaging on the generated data set. Averaging data belonging to the same class:
Figure BDA0002302897610000093
wherein, numiIs RPiThe number of data pieces.
Figure BDA0002302897610000094
Is a data vector, as shown in fig. 2.
And loading the generated similarity metric matrix into the KNN algorithm. The modified LMNNP-KNN algorithm is then applied to the positioning phase.
In the positioning phase, data cleansing is performed on RSS data from unknown locations. And then, similarity calculation is carried out on the data and RSS fingerprint data in the fingerprint map library, and the algorithm adopts the LMNNP-KNN algorithm generated in the second step. And finally, calculating to obtain the estimated position of the user.
The method has the significance of fully utilizing the mapping information from the physical space to the signal space, reducing the reduction of the positioning precision caused by the complex indoor building structure and further improving the precision of indoor positioning.
The above description is only a preferred embodiment of the fingerprint positioning method based on metric learning, and the protection scope of the fingerprint positioning method based on metric learning is not limited to the above embodiments, and all technical solutions belonging to the idea belong to the protection scope of the present invention. It should be noted that modifications and variations which do not depart from the gist of the invention will be those skilled in the art to which the invention pertains and which are intended to be within the scope of the invention.

Claims (6)

1. A fingerprint positioning method based on metric learning is characterized in that: the method comprises the following steps:
step 1: setting a plurality of reference points RP in a target indoor environment, establishing a Cartesian coordinate system, and generating a two-dimensional Euclidean space corresponding to an indoor physical space; respectively collecting RSS data from n Access Points (AP) on m Reference Points (RP);
step 2: adding a position tag to RSS data acquired by each pre-deployed reference point RP to generate original data;
and step 3: preprocessing the original data to obtain a similarity measurement matrix, and establishing a fingerprint map library according to RSS data;
and 4, step 4: loading the similarity measurement matrix and the fingerprint map library into a KNN algorithm to obtain an estimated position of the user;
the preprocessing is performed on the original data in the step 3, and the obtained similarity measurement matrix specifically includes:
step 3.1: the vector similarity is measured, and the similarity of the two matrixes is represented by the following formula:
Figure FDA0002832404010000011
wherein the content of the first and second substances,
Figure FDA0002832404010000012
is a function of distance;
Figure FDA0002832404010000013
and
Figure FDA0002832404010000014
the ith and jth data vectors; m is the initial value of the measurement matrix, and I is the unit matrix;
step 3.2: determining a trained objective function, the trained objective function being represented by:
Figure FDA0002832404010000015
wherein i, j → i is the same type of vector i and vector j, and belongs to the category of vector i; mu is a threshold value; xiijlIs a relaxation factor, which is defined as follows:
Figure FDA0002832404010000016
wherein the content of the first and second substances,
Figure FDA0002832404010000017
is a scale scaling function; xiijlIs a relaxation factor;
step 3.3: the goal of defining the scaling function is to constrain the domain of the objective function to be within a convex range, which is defined as:
Figure FDA0002832404010000018
wherein (R)min,Rmax) Is the scaled convex domain range; λ is a scaling factor;
step 3.4: determining an extrapolation function and an inner pull function, the inner pull function and the extrapolation function being represented by:
Figure FDA0002832404010000019
Figure FDA00028324040100000110
wherein epsilonpull(M) is a pull-in function,
Figure FDA0002832404010000021
is an extrapolation function;
step 3.5: all constraint functions are determined and are expressed by:
Figure FDA0002832404010000022
Figure FDA0002832404010000023
ξijl≥0
M≥0
Figure FDA0002832404010000024
step 3.6: and solving the objective function by adopting a random gradient descent method to obtain an optimal similarity measurement matrix M.
2. The fingerprint positioning method based on metric learning as claimed in claim 1, wherein: the step 1 specifically comprises the following steps:
step 1.1: setting a plurality of reference points RP according to actual requirements in a target indoor environment, establishing a Cartesian coordinate system, and generating a two-dimensional Euclidean space corresponding to an indoor physical space; expressing the set reference point in Euclidean space, reference point RPiIs expressed as RPi＝(xi,yi)；
Step 1.2: the method comprises the steps of adopting signal acquisition equipment pre-installed with WIFE signal acquisition software to respectively acquire RSS data from n Access Points (AP) on m Reference Points (RP), and setting the acquisition frequency of the signal acquisition equipment pre-installed with the WIFE signal acquisition software to be 2Hz and the acquisition time to be 2 minutes.
3. The fingerprint positioning method based on metric learning as claimed in claim 2, wherein: the reference point spacing is set to one meter per RP spacing.
4. The fingerprint positioning method based on metric learning as claimed in claim 1, wherein: the step 2 specifically comprises the following steps:
step 2.1: calculating all AP sets in the indoor signal, and setting all AP sets in the ith data as { AP }iThe total number of data is m, and the total AP set is represented by the following formula:
Figure FDA0002832404010000025
step 2.2: sequencing each piece of the measured RSS data to generate an RSS data set, and expressing the generated RSS data set by the following formula:
Figure FDA0002832404010000031
wherein the content of the first and second substances,
Figure FDA0002832404010000032
for RSS data sets, RSSnIs the nth RSS data;
combining all RSSs according to a reference point RP sequence deployed in advance to generate an RSS data matrix, wherein the RSS data matrix is represented by the following formula:
Figure FDA0002832404010000033
RSS(n，m)is an RSS data matrix;
step 2.3: the method comprises the following steps of performing non-demand data cleaning on collected original data column by column, deleting field data which are not needed, and performing non-demand data cleaning according to the following formula:
Figure FDA0002832404010000034
wherein the celliA yes judgment field; c is a valid field set; d is a reserved field;
step 2.4: performing missing value completion on the processed RSS data, classifying the data missing values of the same-class data column by column, judging the reasons causing data missing, performing different processing on the missing values according to different reasons, and performing classification judgment according to the following formula:
Figure FDA0002832404010000035
wherein the RSSmajorThe RSS value with the highest frequency of occurrence in a certain column of data; NaN is a deletion value; n is the total number of data, thd is a set threshold, and thd belongs to (0, 1);
and eliminating data which do not conform to the actual situation.
5. The fingerprint positioning method based on metric learning as claimed in claim 1, wherein: the step 3 of establishing the fingerprint map library according to the RSS data specifically comprises the following steps:
averaging RSS data from the same RP to eliminate the effect of noise on the data; RSS data from the same RP are averaged by:
Figure FDA0002832404010000036
wherein, numiIs RPiThe number of data pieces of (a);
and averaging all the data, and finally compressing the RSS data collected by m RPs and from n APs collected by each RP into a fingerprint map library matrix of m multiplied by n elements.
6. The fingerprint positioning method based on metric learning as claimed in claim 1, wherein: the step 4 specifically comprises the following steps:
step 4.1: inputting the similarity measurement matrix M and the fingerprint database into a KNN matching algorithm, and expressing the similarity measurement matrix M and the fingerprint database input KNN matching algorithm process by the following formula:
Figure FDA0002832404010000041
wherein d isiVector distance; deltaDIs a set of distances; phikIs from deltaDK element sets extracted from the data; r is a coordinate set;
Figure FDA0002832404010000042
f is the RSS fingerprint vector for estimating the position coordinates;
step 4.2: inputting the measured signal vector to obtain a plurality of measured estimated coordinates, and finally carrying out centroid solving on the estimated coordinates to obtain the coordinates of the position where the user is located.
CN201911228508.6A 2019-12-04 2019-12-04 Fingerprint positioning method based on metric learning Expired - Fee Related CN110933596B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911228508.6A CN110933596B (en) 2019-12-04 2019-12-04 Fingerprint positioning method based on metric learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911228508.6A CN110933596B (en) 2019-12-04 2019-12-04 Fingerprint positioning method based on metric learning

Publications (2)

Publication Number Publication Date
CN110933596A CN110933596A (en) 2020-03-27
CN110933596B true CN110933596B (en) 2021-04-13

Family

ID=69856726

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911228508.6A Expired - Fee Related CN110933596B (en) 2019-12-04 2019-12-04 Fingerprint positioning method based on metric learning

Country Status (1)

Country Link
CN (1) CN110933596B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114449435B (en) * 2020-10-19 2023-12-15 中国移动通信集团重庆有限公司 Fingerprint positioning method based on measurement learning of large-interval neighbor
CN113055814B (en) * 2021-02-05 2023-03-21 江苏大学 Indoor positioning method for improving WKNN
CN113518424A (en) * 2021-08-04 2021-10-19 国网浙江省电力有限公司嘉兴供电公司 Transformer substation operation robot and accurate positioning method thereof
CN113723234B (en) * 2021-08-17 2023-07-07 中铁第四勘察设计院集团有限公司 Fingerprint passive perception positioning method, device and storage medium
CN114286282B (en) * 2021-11-18 2023-12-26 中国科学院空天信息创新研究院 Fingerprint positioning method based on WiFi RSS data and dimension reduction algorithm of mobile phone
CN114745658B (en) * 2022-03-01 2024-06-04 华中科技大学 Distance estimation method and medium based on decision fusion
CN114866971A (en) * 2022-05-06 2022-08-05 中国石油大学(华东) Indoor positioning method and device based on kernel function feature extraction and lasso algorithm
CN115683121B (en) * 2022-12-30 2023-03-17 武汉大学 Method for generating multi-source ubiquitous positioning fingerprint based on measured track

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101651951A (en) * 2009-09-15 2010-02-17 哈尔滨工业大学 Establishing method and positioning method of indoor positioning network of support vector machine based on WLAN
CN101778399A (en) * 2010-01-18 2010-07-14 哈尔滨工业大学 Method for optimizing WLAN (Wireless Local Area Network) indoor ANN (Artificial Neural Network) positioning based on FCM (fuzzy C-mean) and least-squares curve surface fitting methods
CN102131290A (en) * 2011-04-26 2011-07-20 哈尔滨工业大学 WLAN (Wireless Local Area Network) indoor neighbourhood matching positioning method based on autocorrelation filtering

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140315569A1 (en) * 2013-04-21 2014-10-23 International Business Machines Corporation Positioning System in a Wireless Communication Network
CN106093852A (en) * 2016-05-27 2016-11-09 东华大学 A kind of method improving WiFi fingerprint location precision and efficiency
CN107037399A (en) * 2017-05-10 2017-08-11 重庆大学 A kind of Wi Fi indoor orientation methods based on deep learning
CN108834041B (en) * 2018-05-18 2020-08-11 哈尔滨工业大学 Indoor position fingerprint positioning Radio Map establishing method based on tensor reconstruction

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101651951A (en) * 2009-09-15 2010-02-17 哈尔滨工业大学 Establishing method and positioning method of indoor positioning network of support vector machine based on WLAN
CN101778399A (en) * 2010-01-18 2010-07-14 哈尔滨工业大学 Method for optimizing WLAN (Wireless Local Area Network) indoor ANN (Artificial Neural Network) positioning based on FCM (fuzzy C-mean) and least-squares curve surface fitting methods
CN102131290A (en) * 2011-04-26 2011-07-20 哈尔滨工业大学 WLAN (Wireless Local Area Network) indoor neighbourhood matching positioning method based on autocorrelation filtering

Also Published As

Publication number Publication date
CN110933596A (en) 2020-03-27

Similar Documents

Publication Publication Date Title
CN110933596B (en) Fingerprint positioning method based on metric learning
CN107703480B (en) Mixed kernel function indoor positioning method based on machine learning
He et al. Tilejunction: Mitigating signal noise for fingerprint-based indoor localization
CN105657823B (en) WIFI indoor weighted K nearest neighbor positioning algorithm based on kernel function main feature extraction
Abdallah et al. A machine learning approach for localization in cellular environments
Yoo Change detection of RSSI fingerprint pattern for indoor positioning system
Caso et al. Frequentist inference for WiFi fingerprinting 3D indoor positioning
Hoang et al. Parameter estimation and classification of censored Gaussian data with application to WiFi indoor positioning
Ni et al. Fingerprint-MDS based algorithm for indoor wireless localization
Yang et al. Multi-floor indoor localization based on RBF network with initialization, calibration, and update
CN107295538B (en) Positioning reliability calculation method, positioning method using reliability and positioning instrument
CN108680897B (en) Indoor positioning method and device, electronic equipment and storage medium
Subakti et al. Indoor Localization with Fingerprint Feature Extraction
Swangmuang et al. On clustering RSS fingerprints for improving scalability of performance prediction of indoor positioning systems
CN107613466B (en) Indoor positioning method based on fingerprint similarity under ultra-dense wireless network
Madray et al. Relative angle correction for distance estimation using K-nearest neighbors
Abdullah et al. A probability neural network-Jensen-Shannon divergence for a fingerprint based localization
Pandey et al. Residual neural networks for heterogeneous smart device localization in IoT networks
CN108445443B (en) KNN-based fingerprint point clustering method
CN111654808A (en) Method and system for updating fingerprint database and wifi positioning method and system
CN113518308B (en) Optimal AP screening method in indoor positioning
Peng et al. Uck-means: A customized k-means for clustering uncertain measurement data
Chong et al. Metrics for feature-aided track association
Vladislav et al. Implementation of indoor positioning methods: virtual hospital case
US11408964B2 (en) Location estimating apparatus, location estimating method and program storing recording medium, and location estimating system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20210413

Termination date: 20211204