A Wi-Fi RTT / RSS collaborative positioning implementation method considering scene recognition

Abstract

The present invention discloses a Wi-Fi RTT / RSS collaborative positioning implementation method taking into account scene recognition, including the following steps: collecting Wi-Fi RTT and RSS data in a region and extracting features to construct a training data set; using extreme learning machine network training to obtain a scene recognition model; analyzing the Wi-Fi RTT ranging error law and RSS signal distribution characteristics, and constructing a Wi-Fi RTT ranging error compensation model; according to the coordinates of each calibration point in the target area and the Wi-Fi AP coordinates, calculating the theoretical plane distance between the calibration point and the AP, using the RTT ranging error model to simulate the real-time ranging result, using the least squares algorithm and based on the enhanced genetic algorithm to obtain the optimal Wi-Fi AP layout method; constructing a positioning error map of the positioning method in the target area and an artificial fingerprint database for fingerprint positioning. The present invention can improve the accuracy and stability of Wi-Fi FTM positioning.

Description

Technical Field

[0001] The present invention relates to the field of indoor positioning technology, and in particular to a Wi-Fi RTT / RSS collaborative positioning implementation method taking into account scene recognition. Background Art

[0002] After the Wi-Fi Fine Time Measurement (FTM) technology was supported by the IEEE802.11mc protocol released in 2016, it has the potential to be applied on smartphones after years of development. In particular, the Android P system released by Google in 2019 supports Wi-Fi FTM, allowing users to use smartphones to measure the round-trip time (RTT) of Wi-Fi signals between access points (APs) and mobile phones, thereby realizing smartphone RTT ranging positioning. This technology is currently the mainstream method of Wi-Fi FTM positioning. Compared with various traditional indoor positioning methods, RTT ranging positioning has the advantages of convenience, speed, and no reliance on additional proprietary facilities, but it also has disadvantages such as poor ranging stability and large positioning volatility.

[0003] The main reason for the poor stability of Wi-Fi RTT ranging positioning is that the ranging process is easily affected by the indoor environment, and the poor quality of ranging data leads to reduced positioning accuracy and robustness. How to achieve accurate location estimation based on the measured Wi-Fi FTM data is of great significance for the implementation of Wi-Fi FTM positioning technology. Since WiFi RSS (Received Signal Strength) can be synchronously measured during the RTT ranging process, in theory, the implementation of Wi-Fi FTM positioning is not technically limited to the RTT ranging positioning method. In addition, fingerprint or ranging positioning can be achieved using both Wi-Fi RSS and RTT data, which also provides many fusion solutions for improving positioning accuracy and robustness. Therefore, how to use RSS and RTT data to achieve RTT / RSS high-efficiency collaborative positioning is the key to solving the problem of high-precision Wi-Fi FTM positioning. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method for implementing Wi-Fi RTT / RSS collaborative positioning taking into account scene recognition, by constructing a scene recognition model and a fusion positioning method based on Kalman filtering to improve the accuracy and stability of Wi-Fi FTM positioning.

[0005] Technical solution: A Wi-Fi RTT / RSS collaborative positioning implementation method taking into account scene recognition includes an offline phase and an online positioning phase, wherein the work of the offline phase includes the following steps:

[0006] Step A: Divide the target area and define the scene category, collect Wi-Fi RTT and RSS data in the area and extract features, combine RTT / RSS data and feature data to form a training data set; finally, use extreme learning machine network training to obtain a scene recognition model;

[0007] Step B: Analyze the Wi-Fi RTT ranging error law and RSS signal distribution characteristics, train the Wi-Fi signal path loss model, and build a Wi-Fi RTT ranging error compensation model; calculate the theoretical plane distance between the calibration point and the AP according to the coordinates of each calibration point in the target area and the Wi-Fi AP coordinates, use the RTT ranging error model to simulate the real-time ranging results, use the least squares algorithm to calculate the positioning error, and obtain the optimal Wi-Fi AP layout method based on the enhanced genetic algorithm;

[0008] Step C, constructing a positioning error map of the positioning method in the target area;

[0009] Step D, constructing an artificial fingerprint database for fingerprint positioning in the target area;

[0010] The work in the online positioning stage includes: scene recognition and adaptive selection of positioning methods;

[0011] All work in the offline phase is for online positioning services, and all work in the offline phase must be completed before online positioning begins.

[0012] Furthermore, based on the indoor building structure and wireless signal distribution, the indoor environment is divided into corridors, offices, atriums, laboratories, and staircase areas, and each scene is assigned a classification label. Then, several paths are planned in each scene. The testers walk along the paths at a constant speed to collect Wi-Fi RTT and RSS data. Data is collected for each path no less than 100 times, for a duration of no less than 20 seconds, and the sampling frequency is 5 Hz.

[0013] Furthermore, in step A, the steps to obtain the training data set are as follows:

[0014] S11, assuming the number of Wi-Fi APs is k, the measured RTT data is:

[0015] d={d1,…,d k}

[0016] The RSS data is:

[0017] r={r1,…,r k}

[0018] Assume that the acquisition time of a path is n seconds and the length of synchronously collected data is m, then:

[0019]

[0020] Among them, data represents the distance and RSS data collected synchronously;

[0021] S12, normalize the collected RTT and RSS data, and the normalized data are expressed as follows:

[0022]

[0023] in, is the normalized RTT data, is the normalized RSS data; i=1,…,k;

[0024] After data normalization, it is expressed as:

[0025]

[0026] S13, respectively obtain the mean, median, variance, standard deviation, range, maximum value, minimum value, 50% quantile, 75% quantile, and 90% quantile of the RTT and RSS data sequences as statistical features; combine the normalized RTT / RSS data, statistical features, and scene classification labels to form a data set for extreme learning machine model learning.

[0027] Furthermore, in step B, the steps for obtaining the optimal Wi-Fi AP layout are as follows:

[0028] S21, based on the relationship between RSS and distance, the Wi-Fi signal path loss model is fitted as follows:

[0029] R i =R0+10βlog 10 (D i )+ε

[0030] Among them, R i and R0 represent the distance D i and Wi-Fi RSS value at D0, β is the path loss coefficient, and ε represents the measurement random error;

[0031] S22, gridding the target area at fixed intervals, obtaining the coordinates of the gridded points, calculating the theoretical real distance d between the grid points and the Wi-Fi AP, and simulating the real-time ranging data d according to the RTT ranging error model i :d i =d+δ i , where δ i Represents the ranging error given by the RTT ranging error model; based on the simulation results of each grid point, the positioning error is calculated using the least squares method;

[0032] S23, calculate the average error value according to the positioning error of each grid point; and randomly generate N AP layout methods, initialize the genetic algorithm parameters, the population individuals in the genetic algorithm correspond to one AP layout method, select the calculated positioning error average function as the fitness function, and use the genetic algorithm to find the AP layout corresponding to the minimum positioning error as the optimal AP layout.

[0033] Further, in step C, the positioning method includes Wi-Fi RTT ranging, Wi-Fi RTT fingerprinting, Wi-Fi RSS ranging and Wi-Fi RSS fingerprinting;

[0034] The implementation steps for constructing the positioning error map of various positioning methods in the target area are as follows:

[0035] S31, select a number of calibration points in the target area, requiring that these points are evenly distributed in the target area and cover the entire area; then synchronously measure RTT and RSS data at these points;

[0036] S32, according to the collected RTT and RSS data at the calibration points, respectively use the least square method for RTT ranging and RSS ranging positioning, use the K nearest neighbor method to carry out RTT fingerprint positioning and RSS fingerprint positioning, and obtain the positioning error at each calibration point;

[0037] S33, according to the positioning error at the calibration point, linear interpolation is used to generate error maps of the four positioning methods in the target area, and the error maps are divided according to the scene types defined in the offline stage.

[0038] Further, in step D, the artificial fingerprint database includes a Wi-Fi RTT ranging fingerprint library and a Wi-Fi RSS fingerprint library;

[0039] Suppose there are k APs, then the RTT ranging fingerprint Θ i It is expressed as follows:

[0040] Θ i ={d1,…,d k ,x i ,y i}

[0041] Among them, (x i ,y i ) represents the coordinates of the i-th grid point, d k is the distance of the ith grid point relative to the kth AP,

[0042] (X k ,Y k ) represents the coordinates of the kth AP;

[0043] Then the RSS ranging fingerprint Ξi It is expressed as follows:

[0044] Ξ i ={r1,…,r k ,x i ,y i}

[0045] Among them, (x i ,y i ) represents the coordinates of the i-th grid point, r k is the RSS value of the ith grid point relative to the kth AP.

[0046] Furthermore, in the online positioning stage, the RTT / RSS data features are calculated online to form a data sequence with the same dimension as the training data, and the data sequence is input into the scene recognition model to determine the indoor scene where the user is currently located;

[0047] Four positioning error maps are called according to the identified indoor scene, and the two positioning methods with the smallest errors in the area are selected, and Kalman filtering is used to achieve the final position calculation.

[0048] Compared with the prior art, the present invention has the following significant effects:

[0049] The present invention constructs a Wi-Fi RTT / RSS collaborative positioning model that takes scene recognition into account, systematically organizes ranging and fingerprint positioning methods based on RTT and RSS data, so that Wi-Fi FTM positioning in complex environments is no longer limited to RTT ranging; a scene recognition model is proposed for the optimization of positioning strategies, and a fusion positioning method based on Kalman filtering is designed to improve the accuracy and stability of Wi-Fi FTM positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0051] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0052] Figure 1 A schematic diagram of the workflow of a Wi-Fi RTT / RSS collaborative positioning model taking into account scene recognition according to an embodiment of the present invention is shown.

[0053] Step S1, in the offline stage, the target area is divided and the scene category is defined. The Wi-Fi RTT and RSS data in the area are collected using a smartphone and features are extracted. The RTT / RSS data and feature data are combined to form a training data set. Finally, the scene recognition model is trained using an extreme learning machine network.

[0054] Indoor scene division and definition: The indoor environment is divided into corridors, offices, atriums, laboratories, staircase areas, etc. according to the indoor building structure and wireless signal distribution, and each scene is assigned a classification label. Then, several paths are planned in each scene (including regular paths and irregular paths, which can cover the defined scenes). The tester walks along the path at a constant speed to collect Wi-Fi RTT and RSS data. Data is collected for each path at least 100 times, for a duration of at least 20 seconds, and the sampling frequency is 5 Hz.

[0055] Step 11, collect RTT and RSS data;

[0056] Assuming the number of Wi-Fi APs is k, the measured RTT data is:

[0057] d={d1,…,d k} (1)

[0058] The RSS data is:

[0059] r={r1,…,r k} (2)

[0060] Assuming that the acquisition time of a path is n seconds, the length of the synchronously collected data is m = n × 5, which can be expressed as:

[0061]

[0062] Among them, data represents the distance and RSS data collected synchronously.

[0063] Step 12, extracting RTT / RSS data features;

[0064] like Figure 1 As shown, firstly, the Wi-Fi RTT and RSS data on a large number of paths collected in step 11 are normalized, and the normalized data are expressed as follows:

[0065]

[0066]

[0067] in, is the normalized RTT data, is the normalized RSS data; i=1,…,k. and The synchronously collected data can be expressed as follows after normalization:

[0068]

[0069] The mean, median, variance, standard deviation, range, maximum, minimum, 50% quantile, 75% quantile, 90% quantile and other statistical features of the RTT and RSS data sequences are obtained respectively. Finally, the normalized RTT / RSS data, statistical features and scene classification labels are combined to form a data set for extreme learning machine model learning.

[0070] Step 13: scene recognition model training;

[0071] The data set obtained in step 12 is input into the extreme learning machine model, and the model parameters are output after the training is completed to obtain a scene recognition model based on RTT / RSS data.

[0072] Step S2: In the offline stage, the Wi-Fi RTT ranging error law and RSS signal distribution characteristics are analyzed, the Wi-Fi signal path loss model is trained, and the Wi-Fi RTT ranging error model is constructed; according to the coordinates of each calibration point and the Wi-Fi AP coordinates in the target area, the theoretical plane distance between the calibration point and the AP is calculated, the real-time ranging result is simulated using the Wi-Fi RTT ranging error model, the positioning error is calculated using the least squares algorithm, and the optimal Wi-Fi AP layout method is obtained based on the enhanced genetic algorithm.

[0073] The specific implementation steps are as follows:

[0074] Step 21, constructing a Wi-Fi signal path loss model;

[0075] like Figure 1 As shown in the figure, in the offline stage, the RTT / RSS data error law is analyzed to obtain the Wi-Fi RTT ranging error model, which is used to compensate for the RTT / RSS measurement error of real-time measurement and reduce the large mismatch during positioning; according to the relationship between RSS and distance, the Wi-Fi signal path loss model is fitted as follows:

[0076] R i =R0+10βlog 10 (D i )+ε (7)

[0077] Among them, R i , R0 represent the distance D i and Wi-Fi RSS value at D0, β is the path loss coefficient; ε represents the measurement random error. When fitting the path loss model, R i It should be the RSS value after error compensation.

[0078] Step 22, RTT ranging simulation and positioning error calculation;

[0079] The target area is gridded at fixed intervals (such as 0.2 meters), the coordinates of the gridded points are obtained, and the theoretical real distance d between the grid points and the Wi-Fi AP (i.e., the Euclidean distance between the grid point coordinates and the AP coordinates) is calculated. At the same time, the real-time ranging data (i.e., the simulated distance measurement result) d is simulated according to the Wi-Fi RTT ranging error model. i :

[0080] d i =d+δ i (8)

[0081] Among them, δ i Represents the ranging error given by the RTT ranging error model; based on the simulation results of each grid point, the positioning error is calculated using the least squares method.

[0082] Step 23, enhancing the genetic algorithm optimization;

[0083] The average error is calculated based on the positioning error of each grid point; N AP layout methods are randomly generated, and the genetic algorithm parameters are initialized. The individual chromosomes of the population in the genetic algorithm correspond to an AP layout method. The calculated positioning error average function is selected as the fitness function. The AP layout corresponding to the minimum positioning error is found through genetic algorithm optimization, which is the optimal AP layout.

[0084] Step S3: In the offline phase, positioning error graphs of four positioning methods (Wi-Fi RTT ranging, Wi-Fi RTT fingerprint, Wi-Fi RSS ranging, and Wi-Fi RSS fingerprint) in the target area are constructed for the performance evaluation of the online positioning method.

[0085] The specific implementation steps are as follows:

[0086] Step 31, collecting calibration point data;

[0087] Select several calibration points in the target area, which are evenly distributed in the target area and can cover the entire area; then synchronously measure the RTT and RSS data at these points.

[0088] Step 32, calculating the positioning error;

[0089] According to the RTT and RSS data collected at the calibration points, the least squares method is used for RTT ranging and RSS ranging positioning, and the K nearest neighbor method is used for RTT fingerprint positioning and RSS fingerprint positioning to obtain the positioning error at each calibration point.

[0090] Step 33, generating a positioning error map;

[0091] According to the positioning error at the calibration point, linear interpolation is used to generate error maps of the four positioning methods in the target area, and the error maps are divided according to the scene types defined in the offline stage.

[0092] Step S4, in the offline stage, an artificial fingerprint database for fingerprint positioning in the target area is constructed, including a Wi-Fi RTT ranging fingerprint library and a Wi-Fi RSS fingerprint library, to reduce the workload of large-scale manual collection.

[0093] In the offline phase, the plane distance between the grid point and the Wi-Fi AP is calculated. According to the gridded coordinate data in step 2 and the optimal Wi-Fi AP layout, assuming there are k APs, the RTT ranging fingerprint Θ i It is expressed as follows:

[0094] Θ i ={d1,…,d k ,x i ,y i} (9)

[0095] Among them, (x i ,y i ) represents the coordinates of the i-th grid point, d k is the distance of the ith grid point relative to the kth AP,

[0096] (X k ,Y k ) represents the coordinates of the kth AP.

[0097] The plane distance d between the grid point and the Wi-Fi AP calculated according to the path loss model and the above steps k , calculate the Wi-Fi RSS value of each grid point relative to k APs, then the RSS ranging fingerprint Ξ i It is expressed as follows:

[0098] Ξ i ={r1,…,r k ,x i ,y i} (10)

[0099] Among them, (x i ,y i ) represents the coordinates of the i-th grid point, r k is the RSS value of the ith grid point relative to the kth AP.

[0100] Step S5, the online positioning stage includes scene recognition and adaptive selection of positioning methods. The RTT / RSS data features are calculated online to form a data sequence with the same dimension as the training data. The data sequence is input into the scene recognition model to determine the indoor scene where the user is currently located. Four positioning error maps are called according to the identified indoor scene, and the two positioning methods with the smallest errors in the area are selected. Kalman filtering is used to achieve the final position calculation.

[0101] In the online positioning stage, scene recognition: Based on the real-time measured Wi-Fi RTT / RSS data, extract data features such as mean and variance online, organize the data sequence according to step 1 and input it into the scene recognition model. After obtaining the scene recognition result, determine what type of indoor area the user is in.

[0102] Adaptive selection of positioning method: Figure 1 As shown, four corresponding positioning error maps are retrieved according to the identified indoor scene, the performance of the four positioning methods in the area is evaluated by the regional error map, and the two positioning methods with the smallest regional errors are selected as candidate positioning methods. There are 6 possible combinations: RTT ranging + RTT fingerprint, RTT ranging + RSS ranging, RTT ranging + RSS fingerprint, RTT fingerprint + RSS ranging, RTT fingerprint + RSS fingerprint, and RSS ranging + RSS fingerprint.

[0103] Fusion positioning calculation: Kalman filtering is used to achieve the fusion of the two positioning methods; assuming that RTT ranging and RTT fingerprint positioning are selected in scenario A (where the RTT fingerprint positioning error map is the smallest), when the Kalman filter combines the two positioning methods, RTT fingerprint positioning is used as the system state, and the RTT ranging positioning result is used as the system observation, and then the Kalman filter standard process is executed to achieve the combination of the two positioning results.

[0104] It should be noted that the sequence of the four steps in the offline stage in this embodiment is not strictly executed according to the above process. It can be understood that the steps in the offline stage can be interchanged under appropriate circumstances, but the work in the offline stage is all for online positioning services, and all offline stage tasks need to be completed before online positioning begins.

[0105] The terms "including" and "having" in this embodiment are intended to cover non-exclusive inclusions. For example, the inclusion of interpolation methods and optimization algorithms in the implementation steps does not limit the use of the methods in this embodiment, and other available methods not listed may be optionally included for the implementation steps.

[0106] The above describes an implementation of the present invention, but does not limit the scope of the invention patent. It should be noted that researchers in this field can carry out similar research according to the implementation of the present invention, and can also make a series of improvements or adjustments to the steps, algorithms, etc. in the implementation scheme according to practical problems, but all improvements based on the concept of this application are within the scope of protection of this application.

Claims

1. A method for implementing Wi-Fi RTT / RSS collaborative positioning taking into account scene recognition, characterized in that: It includes an offline phase and an online positioning phase, wherein the work in the offline phase includes the following steps: Step A: Divide the target area and define the scene category, collect Wi-Fi RTT and RSS data in the area and extract features, combine RTT / RSS data and feature data to form a training data set; finally, use extreme learning machine network training to obtain a scene recognition model; Step B: Analyze the Wi-Fi RTT ranging error law and RSS signal distribution characteristics, train the Wi-Fi signal path loss model, and build a Wi-Fi RTT ranging error compensation model; calculate the theoretical plane distance between the calibration point and the AP according to the coordinates of each calibration point in the target area and the Wi-Fi AP coordinates, use the RTT ranging error model to simulate the real-time ranging results, use the least squares algorithm to calculate the positioning error, and obtain the optimal Wi-Fi AP layout method based on the enhanced genetic algorithm; Step C: construct a positioning error map of various positioning methods in the target area according to the positioning methods of Wi-Fi RTT ranging, Wi-Fi RTT fingerprint, Wi-Fi RSS ranging and Wi-Fi RSS fingerprint. The implementation steps are as follows: S31, select a number of calibration points in the target area, requiring that these points are evenly distributed in the target area and cover the entire area; then synchronously measure RTT and RSS data at these points; S32, according to the collected RTT and RSS data at the calibration points, respectively use the least square method for RTT ranging and RSS ranging positioning, use the K nearest neighbor method to carry out RTT fingerprint positioning and RSS fingerprint positioning, and obtain the positioning error at each calibration point; S33, generating error maps of four positioning methods in the target area by linear interpolation according to the positioning errors at the calibration points, and dividing the error maps according to the scene types defined in the offline stage; Step D, constructing an artificial fingerprint database for fingerprint positioning in the target area; The work in the online positioning stage includes: the adaptive combination of scene recognition and positioning methods: RTT ranging + RTT fingerprint, RTT ranging + RSS ranging, RTT ranging + RSS fingerprint, RTT fingerprint + RSS ranging, RTT fingerprint + RSS fingerprint, RSS ranging + RSS fingerprint; The positioning method is adaptively selected. The RTT / RSS data features are calculated online to form a data sequence with the same dimension as the training data. The data sequence is input into the scene recognition model to determine the indoor scene where the user is currently located. Four positioning error maps are called according to the identified indoor scene, and the two positioning methods with the smallest error in the area are selected. The Kalman filter is used to achieve the final position calculation. All work in the offline phase is for online positioning services, and all work in the offline phase must be completed before online positioning begins.

2. The Wi-Fi RTT / RSS collaborative positioning implementation method taking into account scene recognition according to claim 1, characterized in that: According to the indoor building structure and wireless signal distribution, the indoor environment is divided into corridors, offices, atriums, laboratories, and staircases, and each scene is assigned a classification label. Then, several paths are planned in each scene. The tester walks along the path at a constant speed to collect Wi-Fi RTT and RSS data. Data is collected for each path no less than 100 times, for a duration of no less than 20 seconds, and the sampling frequency is 5 Hz.

3. The Wi-Fi RTT / RSS collaborative positioning implementation method taking into account scene recognition according to claim 1, characterized in that: In step A, the steps to obtain the training data set are as follows: S11, assuming the number of Wi-Fi APs is k, the measured RTT data is: d={d1,…,d k } The RSS data is: r＝{r1,…,r k} Assume that the acquisition time of a path is n seconds and the length of synchronously collected data is m, then: Among them, data represents the distance and RSS data collected synchronously; S12, normalize the collected RTT and RSS data, and the normalized data are expressed as follows: in, is the normalized RTT data, is the normalized RSS data; i=1,…,k; After data normalization, it is expressed as: S13, respectively obtain the mean, median, variance, standard deviation, range, maximum value, minimum value, 50% quantile, 75% quantile, and 90% quantile of the RTT and RSS data sequences as statistical features; combine the normalized RTT / RSS data, statistical features, and scene classification labels to form a data set for extreme learning machine model learning.

4. The Wi-Fi RTT / RSS collaborative positioning implementation method taking into account scene recognition according to claim 1, characterized in that: In step B, the steps to obtain the optimal Wi-Fi AP layout are as follows: S21, based on the relationship between RSS and distance, the Wi-Fi signal path loss model is fitted as follows: R i =R0+10βlog 10 (D i )+e Among them, R i and R0 represent the distance D i and Wi-Fi RSS value at D0, β is the path loss coefficient, and ε represents the measurement random error; S22, gridding the target area at fixed intervals, obtaining the coordinates of the gridded points, calculating the theoretical real distance d between the grid points and the Wi-Fi AP, and simulating the real-time ranging data d according to the RTT ranging error model i :d i =d+δ i , where δ i Represents the ranging error given by the RTT ranging error model; based on the simulation results of each grid point, the positioning error is calculated using the least squares method; S23, calculate the average error value according to the positioning error of each grid point; and randomly generate N AP layout methods, initialize the genetic algorithm parameters, the population individuals in the genetic algorithm correspond to one AP layout method, select the calculated positioning error average function as the fitness function, and use the genetic algorithm to find the AP layout corresponding to the minimum positioning error as the optimal AP layout.

5. The Wi-Fi RTT / RSS collaborative positioning implementation method taking into account scene recognition according to claim 1, characterized in that: In step D, the artificial fingerprint database includes a Wi-Fi RTT ranging fingerprint library and a Wi-Fi RSS fingerprint library; Suppose there are k APs, then the RTT ranging fingerprint Θ i It is expressed as follows: Θ i ={d1,…,d k ,x i ,y i } Among them, (x i ,y i ) represents the coordinates of the i-th grid point, d k is the distance of the ith grid point relative to the kth AP, (X k ,Y k ) represents the coordinates of the kth AP; Then the RSS ranging fingerprint Ξ i It is expressed as follows: Ξ i ={r1,…,r k ,x i ,y i } Among them, (x i ,y i ) represents the coordinates of the i-th grid point, r k is the RSS value of the ith grid point relative to the kth AP.

Images