A self-supervised learning optimized wireless indoor positioning method

Abstract

The present application relates to the field of indoor positioning technology, and particularly relates to a wireless indoor positioning method based on self-supervised learning optimization, comprising: S1, collecting RSS data of fingerprint points in a positioning area offline to establish a fingerprint database; S2, obtaining RSS data of positioning points online; S3, selecting features to compare with the fingerprint database to obtain a first positioning result; S4, reducing dimension of a plurality of sets of positioning point RSS data by a feature minus one combination and comparing with the fingerprint database respectively to obtain a plurality of second positioning results; S5, determining whether the two positioning results are consistent; if yes, outputting the first positioning result; otherwise, taking the plurality of sets of reduced dimension positioning point RSS data as new selected features to repeat S3-S5 until the two positioning results are consistent. The present application uses the two positioning results to determine and verify the accuracy of data positioning, and improves the effectiveness of positioning data without increasing workload and node device conditions by continuously reducing dimension and feedback eliminating invalid data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of indoor positioning technology, and more particularly to a wireless indoor positioning method optimized by self-supervised learning. BACKGROUND

[0002] With the continuous development of computer technology and network technology, the demand for location-based services has been increasingly researched and applied. Especially in large shopping malls, exhibition halls and other areas with large areas, a complete indoor positioning system is needed for navigation and positioning services. Current GPS technology can meet the needs of outdoor positioning, but it cannot be well applied to indoor environments due to the shielding of indoor walls.

[0003] Current indoor positioning technologies mainly include WLAN, Bluetooth, RFID, ZigBee, etc., and positioning is completed through methods such as measuring distance and matching fingerprint information. The WLAN technology and fingerprint information matching method are widely used due to low device cost and high positioning accuracy. The traditional method sets multiple fingerprint points in a specified area to collect feature data and construct an offline fingerprint database. The RSS data of the positioning point collected online is compared with the offline fingerprint database for positioning, but the positioning is usually one-time positioning, and the result lacks verification. Some technologies filter out unstable signals through filtering processing of the RSS data of the positioning point collected online, and use the stable wireless RSS data obtained for positioning. However, the positioning lacks verification and ignores factors such as indoor environment layout and transformation, personnel movement, human shielding, device collection direction, and device instability, which can produce some stable data that is not easy to filter out. The error of these stable data is very large, for example, the difference in wireless signal strength caused by human shielding can be more than 10 dBm, and the wireless signal is stable and not easy to filter out. Such data is referred to as invalid data by the present application, and invalid data can cause the positioning result of the wireless feature data to differ greatly from the true value to a certain extent.

[0004] Therefore, how to improve the effectiveness of positioning data based on wireless technology and fingerprint information matching method without increasing workload and node devices, greatly reduce the problem of high workload caused by body shielding and other reasons in the positioning process, and further improve the reliability, engineering practicability and result accuracy of positioning. SUMMARY

[0005] Therefore, the present application provides a wireless indoor positioning method optimized by self-supervised learning, which solves the problem of only paying attention to solving the influence of signal time variation on positioning accuracy in the prior art, and ignores the problem of needing to extract effective signals to improve positioning accuracy even for stable wireless signals.

[0006] To achieve the above purpose, the present application adopts the following technical solutions:

[0007] S1, the offline stage uses a device to collect positioning area fingerprint point RSS data, and establishes a fingerprint database;

[0008] S2, the online stage uses a device to obtain positioning point RSS data;

[0009] S3, select features from the positioning point RSS data and compare them with the fingerprint database to obtain a primary positioning result;

[0010] S4, reduce the selected features by one, combine multiple sets of reduced dimension positioning point RSS data, and compare them with the fingerprint database respectively to obtain multiple secondary positioning results;

[0011] S5, compare the multiple secondary positioning results with the primary positioning result to determine whether the results are consistent;

[0012] If the results are consistent, output the primary positioning result; otherwise, use the multiple sets of reduced dimension positioning point RSS data in S4 as new selected features, and repeat S3-S5 until the secondary positioning result is consistent with the primary positioning result.

[0013] Further, S1 includes:

[0014] S11, select M suitable fingerprint points in the indoor environment in combination with the layout of the indoor environment;

[0015] S12, use a device to collect wireless signal strength received on N AP nodes around each fingerprint point;

[0016] S13, construct all collected fingerprint points and all AP nodes received wireless signal strength into a fingerprint database of the offline stage of the positioning area;

[0017] Wherein, the fingerprint database is denoted by D, and its expression is:

[0018]

[0019] Wherein, M is the total number of suitable fingerprint points, N is the total number of wireless signal AP nodes, {r i1 ,r i2 ,...,r iN} is a set of N RSS data collected at the i-th fingerprint point, and r ij represents the RSS data of the i-th fingerprint point on the j-th AP node.

[0020] Further, the positioning point RSS data obtained in S2 has N, denoted as {s1, s2,..., s N}.

[0021] Furthermore, in S3, all the data obtained in S2 is selected as feature data, compared with the fingerprint database, and a single location result is obtained, which is represented as L1 = (l 1x ,l 1y ).

[0022] Furthermore, in S4, the features of the N location point RSS data selected in S3 are reduced by one, resulting in N sets of location point RSS data with reduced dimension, represented as {s1, s2, ..., s...} N-1},…,{s2,s3,…,s N Each of these is compared with the fingerprint database to obtain N secondary localization results, denoted as P1, P2, ..., P. N .

[0023] Furthermore, in S5, the step of comparing the N secondary positioning results obtained in S4 with the primary positioning results and determining whether the two positioning results are consistent is as follows:

[0024] S51. A preset distance threshold is set, wherein the distance threshold is the distance between two nearest-neighbor fingerprint points in the grid method. times;

[0025] S52. Divide the N secondary positioning results with the primary positioning results respectively;

[0026] S53. Select the maximum value from N position differences and compare it with the set distance threshold;

[0027] S54. If the following formula is satisfied, it means that the two positioning results of the data before and after dimensionality reduction are consistent;

[0028] The formula is:

[0029] d≥max{d1,d2,...,d N}

[0030] Where d is a pre-set distance threshold, d i Let be the position difference between the two positions of the i-th time, and let i be the value of [1, N].

[0031] Furthermore, S5 also includes the following steps:

[0032] S55. If the above formula is not satisfied, it means that the two positioning results of the data before and after dimensionality reduction are consistent. It is necessary to return to S3 and select one set of feature data from the N sets of dimensionality reduction positioning point RSS data, and repeat S3 to S5.

[0033] If the two positioning results of the N sets of one-dimensional reduced positioning point RSS data are inconsistent after repeating S3-S5, S3 continues to select one set of two-dimensional reduced positioning point RSS data as characteristic data, and repeats S3-S5.

[0034] After multiple dimension reduction and self-supervised learning optimization, S3 selects one set of k-dimensional reduced positioning point RSS data from (N-k+1) groups as characteristic data, and outputs the one positioning result of the set of data after repeating S3-S5.

[0035] Further, S5 further comprises the following steps:

[0036] S55, if the above formula is not satisfied, it means that the two positioning results of the data before and after dimension reduction are consistent, and it is necessary to return to S3 and select one set of one-dimensional reduced positioning point RSS data as characteristic data, and repeat S3-S5;

[0037] If the two positioning results of the N sets of one-dimensional reduced positioning point RSS data are inconsistent after repeating S3-S5, S3 continues to select one set of two-dimensional reduced positioning point RSS data as characteristic data, and repeats S3-S5.

[0038] After multiple dimension reduction and self-supervised learning optimization, S3 selects one set of k-dimensional reduced positioning point RSS data from (N-k+1) groups as characteristic data, and outputs the one positioning result of the set of data after repeating S3-S5.

[0039] Further, the device used in S1 and S2 is a smart phone or a tablet, and the wireless signal collected is a WiFi or Bluetooth signal.

[0040] Further, the one positioning result and the two positioning results in S3 and S4 are obtained by using the wknn algorithm.

[0041] According to the above technical solution, compared with the prior art, the present application has the following beneficial effects:

[0042] (1) The present application obtains multiple secondary positioning results by dimension reduction, combination and secondary positioning, which are used to compare with the one positioning result, and can verify the positioning effect of the multiple wireless signal node signal strength characteristic data of a positioning point collected in the online stage. This method does not increase the workload and node device, and is very simple, practical and feasible.

[0043] (2) The present application uses dimension reduction, combination, secondary positioning, feature discrimination and feedback to form a self-supervised learning optimization process of indoor wireless positioning data, which can gradually eliminate invalid data in the selected data, and finally output the one positioning result which passes the consistency verification.

[0044] (3) The application can solve the problem that the wireless signal strength collected on different AP nodes in the online stage is stable and different due to indoor environment changes and personnel shielding, that is, invalid data is removed from the collected stable wireless signal feature data to avoid its impact on indoor wireless positioning accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.

[0046] Figure 1 A flowchart of a wireless indoor positioning method optimized by self-supervised learning provided by the present application;

[0047] Figure 2 The secondary positioning result graphs with and without invalid data when six data are selected for one and two dimensional reduction in the embodiments. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be described clearly and completely in the embodiments of the present application with reference to the drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0049] The embodiments of the present application disclose a wireless indoor positioning method optimized by self-supervised learning, comprising the following steps:

[0050] S1, in the offline stage, RSS data of fingerprint points in the positioning area is collected by the device to establish a fingerprint database;

[0051] S2, in the online stage, RSS data of the positioning point is obtained by the device;

[0052] S3, the selected features are compared with the fingerprint database to obtain a primary positioning result;

[0053] S4, one is subtracted from the selected features to combine multiple sets of reduced positioning point RSS data, which are compared with the fingerprint database respectively to obtain multiple secondary positioning results;

[0054] S5, the multiple secondary positioning results are compared with the primary positioning result to determine whether the results are consistent;

[0055] If the results are consistent, output the first positioning result; otherwise, take the positioning point RSS data of the multiple sets of dimension reduction in S4 as new selected features, repeat S3-S5 until the second positioning result is consistent with the first positioning result.

[0056] In this embodiment, S1 includes:

[0057] S11, select M suitable fingerprint points in the indoor environment in combination with the layout of the indoor environment;

[0058] S12, collect wireless signal strength received on N AP nodes around each fingerprint point by the device;

[0059] S13, construct all collected fingerprint points and wireless signal strength received on all AP nodes into a fingerprint database of the offline stage of the positioning area;

[0060] Wherein, the fingerprint database is denoted as D, and its expression is:

[0061]

[0062] Wherein, M is the total number of suitable fingerprint points, N is the total number of wireless signal AP nodes, {r i1 ,r i2 ,...,r iN} is a set of N RSS data collected at the i-th fingerprint point, r ij represents the RSS data of the i-th fingerprint point on the j-th AP node.

[0063] In this embodiment, the positioning point RSS data obtained in S2 has N, denoted as {s1,s2,...,s N}.

[0064] In this embodiment, in S3, all data obtained in S2 are selected as feature data, compared with the fingerprint database, and a first positioning result is obtained, denoted as L1=(l 1x ,l 1y ).

[0065] In this embodiment, in S4, the N positioning point RSS data selected in S3 are reduced by one in feature, combined into N sets of positioning point RSS data reduced by one dimension, respectively denoted as

[0066] {s1,s2,...,s N-1},…,{s2,s3,…,s N}; each of which is compared with the fingerprint database and N second positioning results are obtained, denoted as P1,P2,…,P N .

[0067] In this embodiment, step S5, which compares the N secondary positioning results obtained in S4 with the primary positioning results and determines whether the two positioning results are consistent, is as follows:

[0068] S51. A preset distance threshold is set, wherein the distance threshold is the distance between two nearest-neighbor fingerprint points in the grid method. times;

[0069] S52. Divide the N secondary positioning results with the primary positioning results respectively;

[0070] S53. Select the maximum value from N position differences and compare it with the set distance threshold;

[0071] S54. If the following formula is satisfied, it means that the two positioning results of the data before and after dimensionality reduction are consistent;

[0072] The formula is:

[0073] d≥max{d1,d2,…,d N}

[0074] Where d is a pre-set distance threshold, d i Let be the position difference between the two positions of the i-th time, and let i be the value of [1, N].

[0075] In this embodiment, S5 further includes the following steps:

[0076] S55. If the above formula is not satisfied, it means that the two positioning results of the data before and after dimensionality reduction are consistent. It is necessary to return to S3 and select one set of feature data from the N sets of dimensionality reduction positioning point RSS data, and repeat S3 to S5.

[0077] If the two positioning results of the N sets of reduced one-dimensional positioning point RSS data are inconsistent after repeating S3 to S5, then S3 continues to select one set as feature data from the (N-1) sets of reduced two-dimensional positioning point RSS data, and repeats S3 to S5.

[0078] After multiple dimensionality reduction and self-supervised learning optimizations, S3 selects a set of feature data from the (N-k+1) groups of k-dimensionality-reduced RSS data of the localization points. If the localization results are consistent after repeating S3 to S5, then the localization result of this set of data is output.

[0079] In this embodiment, S5 further includes the following steps:

[0080] S55. If the above formula is not satisfied, it means that the two positioning results of the data before and after dimensionality reduction are consistent. It is necessary to return to S3 and select one set of feature data from the N sets of dimensionality reduction positioning point RSS data, and repeat S3 to S5.

[0081] If the two positioning results of the N groups of reduced one-dimensional positioning point RSS data after repeating S3-S5 are inconsistent, S3 continues to select one group of reduced two-dimensional positioning point RSS data from the (N-1) groups of reduced two-dimensional positioning point RSS data as the feature data, and repeats S3-S5.

[0082] After multiple dimension reduction and self-supervised learning optimization, S3 selects one group of feature data from the (N-k+1) groups of reduced k-dimensional positioning point RSS data, and repeats S3-S5 until the two positioning results are consistent, and then outputs the one positioning result of the group of data.

[0083] In this embodiment, the device used in S1 and S2 is a smart phone or a tablet, and the wireless signal collected is a WiFi or Bluetooth signal.

[0084] In this embodiment, the one positioning result and the two positioning results in S3 and S4 are obtained by using the wknn algorithm.

[0085] Ideally, the initial positioning point RSS data collected in the online stage is directly obtained after one-dimensional reduction combination and twice positioning result discrimination, and the discrimination result is consistent, and the one positioning result is output; however, in actual use, the initial positioning point RSS data collected in the online stage may be invalid due to changes in indoor environment, signal collection direction and personnel shielding, and the like, so that one or more invalid data exist in the collected positioning point RSS data. The positioning result obtained by one positioning lacks reliable verification and cannot avoid the influence of the collected error and stable data on the indoor wireless positioning accuracy, so the above specific solutions are proposed to realize the self-supervision process of the feature data in indoor positioning.

[0086] The two positioning results of the present application can verify the accuracy of the one positioning result, which is better than the traditional method of positioning only by filtering, because the positioning accuracy can be verified.

[0087] In order to more conveniently understand the wireless positioning method of the present application, it is assumed that 6 feature data are selected in S3 and compared with the fingerprint database to obtain the one positioning result; 5 feature data are randomly selected from the 6 feature data, i.e. the combination of 6 selected 5, which can be combined into 6 groups of reduced one-dimensional positioning point RSS data containing 5 feature data for secondary positioning, and 6 secondary positioning results are obtained.

[0088] If the 6 feature data are all valid data, the error between the 6 secondary positioning results and the one positioning result will not be too large, in other words, the 6 secondary positioning positions obtained by using 5 feature data are similar to the one positioning position obtained by using 6 feature data.

[0089] If there is one stable but large error invalid data in the 6 feature data, the 6 feature data after feature reduction by one will combine 6 groups of 5 feature data positioning point RSS data, i.e. reduced one-dimensional positioning point feature data, which are compared with the fingerprint database respectively to obtain 6 secondary positioning results;

[0090] Since 5 groups of the reduced one-dimensional positioning point feature data contain stable but large error invalid data, the positioning results of the 5 groups of the reduced one-dimensional positioning point feature data have large errors, which will exceed the set position threshold and be inconsistent with the primary positioning result before dimension reduction;

[0091] Similarly, continuing to reduce the dimension in the above manner, 5 groups of 4 feature data positioning point RSS data, i.e. reduced two-dimensional positioning point feature data, are obtained, which are compared with the fingerprint database respectively to obtain 5 secondary positioning results;

[0092] Since one group of the 6 groups of reduced one-dimensional positioning point feature data has removed the invalid feature data, when continuing to reduce the dimension, combine and determine the two positioning results, the two positioning results will be consistent.

[0093] Figure 2 In the figures, Figure 1 is the secondary positioning result of 5 feature data in the positioning point RSS data containing large error feature data, and Figure 2 is the secondary positioning result of 4 feature data in the positioning point RSS data not containing large error feature data;

[0094] From the two figures, it is not difficult to see that the secondary positioning results not containing large error feature data are close in position, and vice versa.

[0095] Through many experiments similar to the above, it is found that when the positioning point RSS data contains large error feature data, the average positioning accuracy of the positioning result can be improved by about 20% after using the wireless positioning method of the application.

[0096] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant part can be referred to the method part.

[0097] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the appended claims are intended to cover all such modifications that do not depart from the true spirit and scope of the application. Therefore, the application is not limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A self-supervised learning-optimized wireless indoor positioning method, characterized in that, The main steps include: S1. In the offline stage, the device collects RSS data of fingerprint points in the positioning area to establish a fingerprint database. S2. During the online phase, the device is used to acquire the RSS data of the positioning point; S3. Select features from the location point RSS data and compare them with the fingerprint database to obtain a location result; S4. Reduce the selected features by one, combine them to generate multiple sets of dimensionality-reduced location point RSS data, compare them with the fingerprint database respectively, and obtain multiple secondary localization results. S5. Compare multiple secondary positioning results with the primary positioning result to determine whether the results are consistent; If the results are consistent, output the first localization result; otherwise, use the RSS data of the multiple sets of dimensionality-reduced localization points in S4 as the newly selected features, and repeat S3 to S5 until the second localization result is consistent with the first localization result. In S5, the steps to compare the N secondary positioning results obtained in S4 with the primary positioning results and determine whether the two positioning results are consistent are as follows: S51. A preset distance threshold is set, wherein the distance threshold is the distance between two nearest-neighbor fingerprint points in the grid method. times; S52. Divide the N secondary positioning results with the primary positioning results respectively; S53. Select the maximum value from N position differences and compare it with the set distance threshold; S54. If the following formula is satisfied, it means that the two positioning results of the data before and after dimensionality reduction are consistent; The formula is: d≥max{d1,d2,...,d N } Where d is a pre-set distance threshold, d i Let i be the position difference between the two positioning operations for the i-th time, and let i be the value in the range [1, N]. S5 also includes the following steps: S55. If the above formula is not satisfied, it means that the two positioning results of the data before and after dimensionality reduction are inconsistent. It is necessary to return to S3 and select one set of feature data from the N sets of dimensionality reduction positioning point RSS data, and repeat S3 to S5. If the two positioning results of the N sets of reduced one-dimensional positioning point RSS data are inconsistent after repeating S3 to S5, then S3 continues to select one set as feature data from the (N-1) sets of reduced two-dimensional positioning point RSS data, and repeats S3 to S5. After multiple dimensionality reduction and self-supervised learning optimizations, S3 selects a set of feature data from the (N-k+1) sets of k-dimensional reduced location point RSS data. If the two location results are consistent after repeating S3 to S5, then the location result of this set of data is output. Also includes: S6. If all the data obtained in S2 cannot achieve consistency between the two positioning results even after multiple dimensionality reduction and self-supervised learning optimizations, then return to S2 to obtain a new set of positioning point RSS data and continue to execute S3 to S5.

2. The self-supervised learning-optimized wireless indoor positioning method according to claim 1, characterized in that, S1 includes: S11. Select M suitable fingerprint points indoors, taking into account the layout of the indoor environment. S12. Use the device to collect the wireless signal strength received on N AP nodes around each fingerprint point; S13. Construct a fingerprint database for the offline phase of the positioning area by collecting all fingerprint points and the wireless signal strength received on all AP nodes. The fingerprint database is denoted by D, and its expression is: ； Where M is the number of suitable fingerprint points, N is the number of wireless signal AP nodes, and {r i1 ,r i2 ,...,r iN Let r be the set of N RSS data collected at the i-th fingerprint point. ij This represents the RSS data of the i-th fingerprint point on the j-th AP node.

3. The self-supervised learning-optimized wireless indoor positioning method according to claim 2, characterized in that, There are N RSS data points for the location points obtained in S2, represented as {s1, s2, ..., s...} N } 4. The self-supervised learning-optimized wireless indoor positioning method according to claim 3, characterized in that, In S3, all the data obtained in S2 is selected as feature data, compared with the fingerprint database, and a location result is obtained.

5. The self-supervised learning-optimized wireless indoor positioning method according to claim 4, characterized in that, In S4, the features of the N location point RSS data selected in S3 are reduced by one, resulting in N sets of location point RSS data with reduced dimension, denoted as {s1, s2, ..., s...} N-1 } ,...,{s2,s3,...,s N Each of these is compared with the fingerprint database to obtain N secondary localization results, denoted as P1, P2, ..., P... N .

6. The self-supervised learning-optimized wireless indoor positioning method according to claim 1, characterized in that, The devices used in S1 and S2 are smartphones or tablets, and the wireless signals collected are WiFi or Bluetooth signals.

7. The self-supervised learning-optimized wireless indoor positioning method according to claim 1, characterized in that, The primary and secondary localization results in S3 and S4 are obtained using the WKNN algorithm.

Images