Indoor partition positioning method for reducing measurement information

Abstract

The present invention discloses an indoor zoning positioning method for streamlining measurement information, belonging to the field of indoor positioning. First, the present invention combines the traditional fingerprint positioning method and the compressed sensing technology to construct a positioning system model; then, it uses the spatial position characteristics to complete the fingerprint division of the fuzzy area and effectively screen APs according to multiple scales; finally, it calculates the target position and sparsely reconstructs the position information of the point to be located. The present invention performs compressed sampling on the measurement information, and then restores a nearly complete signal value through a limited number of measurement values, avoiding the acquisition and storage of samples of the complete signal. At the same time, the reduction of the sample size reduces the overhead during online calculation. Overall, APs outside the area with less influence are filtered out effectively, and individually, an AP subset with intensive information in the area is selected by comprehensively considering the discrimination and measurability. The SBL algorithm is selected to handle the influence of high correlation between fingerprint information, and the sorting of the AP information sequence is introduced to enhance the weight of adjacent RPs, effectively improving the positioning accuracy.

Description

Technical Field

[0001] This invention relates to an indoor zoning positioning method that simplifies measurement information, belonging to the field of indoor positioning technology. Background Technology

[0002] Given the dissimilarity of RSSI signal distributions across different spatial locations, this location-specific, unique RSSI characteristic information can serve as a fingerprint. Fingerprint localization technology determines the approximate location of the target by matching its fingerprint information with that of a known reference point. This data comparison method avoids multipath interference and fading errors during signal propagation. In the fingerprint localization process, an offline phase records the RSSI information received by the reference point (RP) from the access point (AP) to build a fingerprint database. An online phase matches the measurement information of the target point (TP) with the offline fingerprint database to determine the location of the TP.

[0003] In recent years, Compressive Sensing (CS) has been introduced into the field of indoor positioning. CS extracts sparse signals containing redundancy from the natural environment during sampling to obtain a sparse basis. During data dimensionality reduction and compressed sampling, CS maximizes information density in finite dimensions. When recovering the original signal, it accurately reconstructs the signal by selecting correlated information from the sparse basis. Linking indoor fingerprint positioning with the CS field transforms the dual challenges of offline fingerprint database construction and online target location restoration into a mathematical problem of compressed sampling and reconstruction of sparse signals.

[0004] Sparse signal compression sampling corresponds to the fingerprint database construction problem in the context of localization. By compressing and sampling fingerprint information simultaneously, sparse bases are established during the compression sampling process. Localization accuracy is closely related to the quality of the sparse bases and largely depends on the degree to which the fingerprint information replicates the real environment. To reduce the size of the fingerprint database and computational costs, clustering methods such as K-Means and Affine Propagation Clustering (APC) are applied to partition small-volume fingerprint databases in the offline stage. However, this single-data-perspective processing lacks a close fit to the actual environment, and the hard partitioning method of single RP assignment can lead to the loss of potential reference RP points on the other side of the cluster edge. To compress and extract high-localization-value information from fingerprints, a screening mechanism is introduced to remove unstable individuals in the AP, thereby increasing the information density of the fingerprint. Different methods have been used to measure redundant AP from various perspectives: sorting by the average intensity of offline data; selecting information entropy as the AP contribution scale; incorporating historical fingerprint database information into the offline stage for pre-measurement based on the Fisher criterion; and determining the AP value by the variance of AP change with respect to RP. These methods simplify fingerprint information from different angles, but they do not assess the regional impact of AP signal attenuation, nor do they take into account the RP's ability to identify near-range APs.

[0005] CS sparse reconstruction selects existing fingerprint information and reconstructs the original TP measurement signal to obtain the position vector, indirectly determining the nearest neighbor reference points. The recovery effect of the reconstruction algorithm plays a crucial role in localization and can be divided into greedy algorithms represented by Orthogonal Matching Pursuit (OMP) and convex optimization algorithms represented by Basis Pursuit (BP). In real-world environments, the signal variation of nearby RP points is small, and fingerprint information has strong correlation. The algorithms based on greedy iterative approximation and convex optimization linear programming are difficult to effectively distinguish and match highly correlated information, resulting in poor reconstruction accuracy. Summary of the Invention

[0006] To address the current problems of low fingerprint information density and insufficient accuracy in reconstructing target locations, this invention provides an indoor zone positioning method that simplifies measurement information. The specific technical solution is as follows:

[0007] The first objective of this invention is to provide an indoor positioning method, comprising:

[0008] First, a positioning system model is constructed based on fingerprint positioning methods and compressed sensing technology;

[0009] Then, spatial location features are used to complete the fingerprint segmentation of fuzzy regions and effective AP screening is performed based on multiple scales;

[0010] Finally, the target position is calculated, and the position information of the point to be located is sparsely reconstructed.

[0011] Optional, specifically including:

[0012] Step 1: Construct a CS-based positioning system model;

[0013] Step 1.1: Set up M AP signal sources and N RP reference points in the area to be located; in the area located at (x i ,y i A reference point RP i At this location, the RSSI matrix originating from the AP signal source was measured. Where i = 1, 2, ..., N; the initial fingerprint data is Ω = {Ω1, Ω2, ..., Ω}. N};

[0014] Record RP i The nth measurement vector for the j-th AP signal source is Where j = 1, 2, ..., M; Ω i The set of absolute median deviations of M APs, where yes The absolute median deviation; Let Ω be the fingerprint set of K regions after being divided, and N be the fingerprint set of these regions. k Let be the number of RPs in the k-th region; Let be... c k For the corresponding Cluster centers;

[0015] At the test point TP located at (x, y), the received measurement value vector from the AP signal source is:

[0016] Step 1.2: Using the following formula, obtain the position vector f of the test point TP through the reconstruction algorithm:

[0017]

[0018] Where P is the number of AP signal sources after filtering, ε is the noise figure, and Θ P×M For the measurement matrix, Φ M×N It is a sparse basis matrix, i.e., a set of regional fingerprints; Ψ P×N The sensing matrix in CS is the information-dense fingerprint set. To pass through the sensing matrix Ψ P×N And the low-dimensional observation vector measured and processed at point TP;

[0019] Step 2: Addressing the measurement uncertainties and spatial non-uniformity of the RSSI signal, the initial fingerprint data Ω = {Ω1, Ω2, ..., Ω...} is... NThe RSSI uncertainty filtering and region segmentation processes are performed sequentially to establish a stable and effective regional fingerprint set Φ. M×N ;

[0020] Step 3: Perform multi-scale screening of effective APs based on validity, discriminability, and measurability to obtain the information-dense fingerprint set.

[0021] Step 4: Calculate the target location and sparsely reconstruct the location information of the point to be located.

[0022] Optionally, step 2 includes:

[0023] Step 2.1: Filtering out signal uncertainties;

[0024] Given the reference point RP i nth measurement vector of the j-th AP signal source Obtain signal source AP j In RP i corresponding median and absolute median deviation To fit the RSSI distribution in a real-world scenario, it is determined that... Out-of-range RSSI values ​​are considered outliers and are filtered out; the processed values ​​are then... average As valid fingerprint data, RP is obtained. i fingerprint information All RPs constitute the fingerprint database.

[0025] Step 2.2: Spatial feature correction region division;

[0026] Step 2.2.1: Calculate the set of dissimilarity similarities between paired reference points RP, given reference points. fingerprint information and By introducing RP spatial location feature weights as scene factors to construct the difference similarity between RPs, the influence of data bias on the original similarity is corrected, and a reference point is defined. and The similarity is:

[0027]

[0028] In the formula, Let be the spatial location feature weights, and satisfy the following condition: If The spatial distance between Ω′ and the remaining RP is divided into U intervals. and spatial distance Located in the u-th interval, as shown in the following formula:

[0029]

[0030] In the formula, u=1,...,U,

[0031] Step 2.2.2: Merge RP spatial features to obtain cluster centers C = APC(S,Ω'), and perform initial region division according to criterion 1;

[0032] Step 2.2.3: Determine the cluster edge points based on criterion 2, and update the fuzzy region determination of the initial region fingerprint set;

[0033] The first criterion is:

[0034] The second criterion is: Guidelines

[0035] Where γ1 is the similarity threshold, They are respectively regions and The regional center;

[0036] The Ω region fingerprint set is obtained by dividing Ω using criteria 1 and criterion 2. Where N k The number of RPs contained in the k-th subregion.

[0037] Optionally, the validity calculation process in step 3 includes:

[0038] Given an AP signal strength threshold γ2, if the following conditions are met... Then AP j In RP i Valid mark T i j =1, otherwise T i j =0; Define AP j In the region The regional reliability is:

[0039]

[0040] Select AP j right Inside N k The set of absolute median deviations of each RP Reflecting AP j Calculate the average value of the signal fluctuation deviation at different RP points. As AP j The higher the reliability metric value, the more drastic the AP fluctuations and the lower the stability.

[0041] Normalization yields AP j exist The regional stability is:

[0042]

[0043] In the formula, σ>0 and σ→0;

[0044] Considering both regional reliability and regional stability, AP j For the current region The validity is:

[0045]

[0046] AP j of The higher the value, the better it is in the region. The more stable and reliable the fingerprint information, the better.

[0047] Optionally, the calculation process for the discrimination in step 3 includes:

[0048] Define AP j In RP i The fluctuation range of the nth signal measurement set at the location is

[0049]

[0050]

[0051] At the reference point and Between, AP j The overlap of the measurement set intervals is represented by the extended Jaccard coefficient:

[0052]

[0053] In the formula, |·| represents the length of the interval, AP j The final discrimination is taken from the region Mean overlap between all RP pairs

[0054] Optionally, the measurability calculation process in step 3 includes:

[0055] via AP to the area Average RSSI intensity distribution of internal RP Reflecting the measurability of AP within the region, among which AP j The stronger the measurability, the corresponding The larger the value.

[0056] Optionally, step 3 further includes:

[0057] For sub-regions Based on the overall stability differences of APs across different regions, V regions with stable and reliable performance were selected from the original AP set. k A number of valid APs were obtained. Then, based on the individual localization effect of AP, the measurability and discriminability were weighed through comprehensive performance factors, and the system was further simplified. The number of APs in China up to P k Seeking To achieve further optimization under multiple gain conditions;

[0058] Find an information-dense fingerprint set that simultaneously satisfies the sparse recovery condition and reflects the true RSSI distribution characteristics in the scene. In sparse matrix form, combined The fingerprint database is simplified in size and its information density is amplified through matrix multiplication, which is characterized as follows: in Each vector contains only one 1 element and the rest are 0. The index of this non-zero element corresponds to the position of the effective AP in the original AP. The elements are multiplied by matrix multiplication. Information filtering of redundant APs;

[0059] The same optimization strategy is applied to all sub-regions in Φ. The linkage between internal and external regions enables multi-scale fingerprint set simplification based on effectiveness, discriminability, and measurability under the information-intensive driving force, ultimately resulting in an information-intensive fingerprint set.

[0060] Optionally, step 4 includes:

[0061] Step 41: Reconstruction and recovery of real-time position vector;

[0062] During the online phase, a measurement vector y is received from the signal source AP at a given test point TP. tp Cluster centers C = {c1, c2, ..., c K},right Calculating Euclidean Distance Based on RSSI Signal Variance of AP The result is used as the similarity between TP and each cluster center for coarse localization. The region with the highest similarity is determined as the sub-region to which TP belongs. The TP location vector f is then reconstructed and recovered using the following formula combined with the sparse Bayesian algorithm:

[0063]

[0064] Step 42: Define the RSSI order of APs in the RP fingerprint information as the information sequence, quantify the similarity between RP points and TPs by AP intensity ordering, and then reconstruct the position coefficients by weighting.

[0065] right Obtain its corresponding mapping RP in Φ i fingerprint information Ω′ i If AP m , exist and observation vector y tp If the RSSI intensity rankings are the same, then μ is considered to be... m,n =1;

[0066] For all APs, pairwise comparisons yield D = P k ·(P k -1) / 2 possible outcomes, RP i The degree of similarity to the target location TP is determined by the similarity coefficient IS of the information sequence. tp,i express:

[0067]

[0068] RP i The closer IS is to TP in spatial location, the better. tp,i The higher the value;

[0069] θ is calculated according to the following formula. tp The set of information sequence similarity values ​​corresponding to the l mapped RPs is IS = {IS tp,i |1≤i≤l}, during online positioning, the position coefficients are reconstructed by combining IS comparison weighting to obtain the TP position:

[0070]

[0071] A second objective of this invention is to provide an indoor positioning system, comprising:

[0072] At least one storage medium storing at least one instruction set for locating a point to be located; and at least one processor communicatively connected to the at least one storage medium.

[0073] When the indoor positioning system is running, the at least one processor reads the at least one instruction set and implements the indoor positioning method described above.

[0074] The beneficial effects of this invention are:

[0075] 1. Fingerprint information that more closely approximates the actual signal distribution is obtained through the Absolute Median Deviation (MAD) method. Leveraging the asymmetry of the APC similarity matrix and the stability of consistent results from multiple clustering iterations, spatial features are introduced to redesign the similarity criteria and clustering mechanism. This allows for adaptive partitioning results that fit the scene characteristics during the clustering process, reducing the fingerprint database size and improving computational efficiency. Simultaneously, the introduced fuzzy re-partitioning mechanism avoids the mismatch problem of edge RP (Resolution Mapping).

[0076] 2. Fingerprint information is extracted through a multi-scale AP screening method. Overall, APs outside the region with less influence are filtered out based on effectiveness, while individual APs are selected based on a combination of distinguishability and measurability to identify a subset of APs with dense information within the region. This two-step approach, linking internal and external methods, solves the problems of strong signal fluctuations in long-distance APs and weak identification ability of short-distance APs, effectively consolidating the information density of fingerprints.

[0077] 3. The SBL algorithm is used to weaken the influence of high correlation between fingerprint information, resulting in an effective set of nearest neighbor RPs. Addressing the critical issue of non-nearest neighbor interference RPs introduced by non-grid design in indoor positioning scenarios, an AP information sequence ranking method is proposed to enhance the weight of neighbor RPs, effectively improving positioning accuracy. Attached Figure Description

[0078] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0079] Figure 1 This is a schematic diagram of the CS-based positioning system model of the present invention.

[0080] Figure 2 This is a diagram illustrating the edge RP adaptation effect of the present invention.

[0081] Figure 3 This is a flowchart of the method of the present invention (FingerprintInformationDensityAggregationPositioningAlgorithmBasedOnCompressedSensing, abbreviated as FIDA)).

[0082] Figure 4 This is a comparison chart of the clustering effects of the method of the present invention with those of APC and K-Means.

[0083] Figure 5 Box plot of localization error for clustering algorithm.

[0084] Figure 6 This is a comparison chart of the cumulative probability of errors in multi-scale AP screening and various AP selection strategies.

[0085] Figure 7 The images show a comparison of the position coefficients before and after reconstruction. (a) is located at TP7, (b) is located at TP46, and (c) is located at TP107.

[0086] Figure 8 This is a comparison chart of the cumulative error analysis of FIDA and various algorithms. Detailed Implementation

[0087] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0088] Example 1:

[0089] This embodiment provides an indoor positioning method, including:

[0090] First, a positioning system model is constructed based on fingerprint positioning methods and compressed sensing technology;

[0091] Then, spatial location features are used to complete the fingerprint segmentation of fuzzy regions and effective AP screening is performed based on multiple scales;

[0092] Finally, the target position is calculated, and the position information of the point to be located is sparsely reconstructed.

[0093] Example 2:

[0094] This embodiment provides an indoor zoning positioning method that simplifies measurement information. See [link to relevant documentation]. Figure 3 The method includes the following steps:

[0095] Step 1: Construct a positioning system model;

[0096] Step 2: Fingerprint segmentation in the fuzzy region;

[0097] Step 3: Multi-scale screening of effective APs;

[0098] Step 4: Calculate the target location and sparsely reconstruct the location information of the point to be located.

[0099] The specific process of step one is as follows:

[0100] (1) Spatial Model

[0101] There are M AP signal sources and N RP reference points within the area to be located. In the area located at (x... i ,y i A reference point RP i At this location, the RSSI matrix originating from the signal source AP was measured. Where i = 1, 2, ..., N. The initial fingerprint data is Ω = {Ω1, Ω2, ..., Ω}. N}. Record RP i The nth measurement vector for the j-th AP is Where j = 1, 2, ..., M; Ω i The set of absolute median deviations of M APs, where yes The absolute median deviation; Let N be the set of fingerprints for the K sub-regions after Ω is divided. k Let C = {c1, c2, ..., c} be the number of RPs in the k-th region. K c in} k For the corresponding The cluster center; located at the test point TP at (x,y), the received measurement vector from AP is

[0102] (2) Compressed Sensing Theory

[0103] According to the CS theoretical model, when When the signal has a sparsity of X, it can be obtained through the sensing matrix. and the low-dimensional observation vector measured and processed at point TP When Ψ satisfies the sparse constraint, based on the CS sparse reconstruction principle, the corresponding reconstruction algorithm is selected through equation (1) to obtain f.

[0104]

[0105] If we map equation (1) to fingerprint localization, let P represent the number of APs after screening, N be the number of RPs, and ε be the noise coefficient. Θ is the measurement matrix, and Φ is the sparse basis matrix (referred to as the regional fingerprint set in the text). Based on CS theory, utilizing the natural sparse characteristics of location fingerprints, the establishment and optimization of the fingerprint database can be transformed into constructing the sensing matrix Ψ in CS (referred to as the information-dense fingerprint set in the text). During target localization, related to the principle of fingerprint localization in selecting neighboring RPs with similar fingerprints to determine the TP location, CS establishes the location vector f and the observation vector y′ through Ψ. tp The mapping relationship is used to obtain the target location by using f, which can characterize the similarity of RP fingerprints in Ψ.

[0106] (3) CS-based positioning model

[0107] Positioning is divided into two stages, such as Figure 3As shown. Offline, the region fingerprint set Φ is obtained through uncertainty filtering and fuzzy region segmentation sequentially Ω. Then, Φ is transformed into an information-dense fingerprint set Ψ by refining the AP positioning value. Online, coarse localization first determines the sub-region where the target is located, and then fine-grained localization is performed using the CS algorithm to determine the target's position.

[0108] 1) Information-intensive fingerprint sets

[0109] Combining CS theory with actual positioning scenarios, P < M < < N is usually present, therefore, determining equation (1) is a problem of solving an underdetermined equation. To ensure the uniqueness of the solution, Ψ should satisfy the sparse recovery condition, i.e., the constrained isometric property (RIP). When P and N satisfy P = O(Xlog(N / X)), the equation can obtain a unique solution. Considering that AP screening can both ensure the sparse recovery condition's limitation on the number of APs and remove redundant information in the process to improve the positioning effect, a measurement matrix Θ can be constructed by screening redundant APs, and the high-value positioning information in Φ can be aggregated by equation Ψ = Θ * Φ to obtain the information-dense fingerprint set Ψ.

[0110] 2) Sparsity of location fingerprints

[0111] In indoor positioning scenarios, since a target can only be in a unique location at any given time, this characteristic naturally results in sparsity of target positions. Let f represent the potential position of TP among N RPs, and be a sparse vector of 1. N×1 =(θ1,θ2,...,θ N ) T θ i =1 or 0 is the position coefficient, indicating whether TP is located at RP. i Place.

[0112] 3) Target location positioning

[0113] In application scenarios, the randomness of TP's movement within the region can cause the positioning result to lie between several reference points RP, resulting in f containing multiple non-zero θ. i The set of l non-zero position coefficients is θ tp ={θ1,θ2,...,θ l The position coordinates (x, y) of TP are represented by a weighted average of f and the position coefficient of RP:

[0114]

[0115] The specific process of step two is as follows:

[0116] To address the measurement uncertainty and spatial non-uniformity of RSSI signals, the initial fingerprint set was sequentially subjected to RSSI uncertainty filtering and region division processing to overcome environmental disturbances to the measured values ​​and the fluctuations in actual location caused by RSSI distribution, thereby establishing a stable and effective regional fingerprint set.

[0117] (1) Filtering out signal uncertainty

[0118] RSSI measurement data are subject to uncertainty due to environmental influences, primarily manifested as a small number of drastically fluctuating impulse errors. By using the median as a reference and taking the median deviation of the data from the median as the absolute median deviation, a relative deviation method can be adopted to avoid impulse error disturbances introduced by the standard deviation.

[0119] Given AP can be obtained j In RP i corresponding median and absolute median deviation To fit the RSSI distribution in a real-world scenario, it is determined that... RSSI values ​​outside the range are considered outliers and are filtered out. The processed values ​​are then... average As valid fingerprint data, RP is obtained. i fingerprint information A fingerprint database composed of all RPs

[0120] (2) Spatial feature correction region division

[0121] Real-world location scenarios often feature large areas and complex spatial structures. Dividing the scene into several differentiated sub-regions can effectively reduce the dimensionality of the fingerprint database (RP) and improve online matching efficiency. However, using only Euclidean distance as the similarity criterion during clustering while ignoring scene structure can lead to RPs being assigned to incorrect sub-regions. To address these issues,

[0122] By leveraging the Affinity Propagation Cluster (APC) algorithm, which utilizes the asymmetric similarity matrix and the consistency of multiple clustering results, and integrating the spatial location features of the Propagation Principle (RP), a clustering algorithm with improved similarity criteria and fuzzy region partitioning mechanism is proposed.

[0123] 1) Spatial location-corrected RP similarity

[0124] Given fingerprint information and The proximity of two locations can be determined by the Euclidean distance of their RSSI signals. However, this mapping is susceptible to signal fluctuations. Therefore, spatial location weights of the reference points (RPs) can be introduced as scene factors to construct a difference-based similarity between RPs, correcting the impact of data bias on the original similarity. Definition and The similarity is:

[0125]

[0126] In the formula, Let be the spatial location feature weights, and satisfy the following condition: If The spatial distance between Ω′ and the remaining RP is divided into U intervals. and spatial distance Located in the u-th interval, as shown in equation (4):

[0127]

[0128] In the formula, u=1,...,U,

[0129] 2) Fuzzy region division

[0130] Typically, when a TP is located at the edge of a region, online matching can only obtain the neighboring RPs of its sub-region, while losing the positional reference information of the neighboring RPs in the region on the other side of the edge. Figure 2 As shown.

[0131] Depend on Figure 2 It can be seen that the introduction of mismatch RP after fuzzy region segmentation effectively constrains the deviation distance of the localization results. To accommodate the influence of RP on the target from both sides of the edge, after the algorithm finishes execution and obtains the cluster center set C, the RP of the cluster edge points is adjusted. i A second sub-region is allocated based on a secondary decision, establishing a fuzzy region. The decision criteria are as follows:

[0132] Guideline 1:

[0133] Guideline 2: Guidelines

[0134] In the above formula, γ1 is the similarity threshold. They are respectively regions and The region center. Using criteria 1 and 2, Ω can be divided into K region fingerprint sets. The number of RPs contained in the k-th subregion.

[0135] 3) Clustering process

[0136] The clustering process is based on APC, which treats each RP as a potential cluster center. There is no need to set the characteristics of the initial cluster center. The spatial location features of RP in the actual scene are introduced to help correct the similarity. During the iteration process, the cluster center is adaptively determined by the scene features. The region division is completed by fuzzy discrimination criteria. This can effectively constrain the positional changes caused by RSSI distribution and make up for the edge mismatch caused by the single region affiliation of RP points.

[0137] The first step calculates the spatial feature position weights between all pairwise relative points (RPs), and then calculates the set of dissimilarity similarities between the pairwise RPs. The second step fuses the spatial features of the RPs to obtain cluster centers C = APC(S,Ω'), and performs initial region partitioning according to criterion 1. The third step determines the cluster edge points according to criterion 2, updating the initial region fingerprint set with fuzzy region determination. The spatial location-corrected clustering algorithm reduces the RP dimension of the fingerprint set, effectively reducing offline storage costs and online computational overhead by maintaining a smaller number of RPs within each partition. However, due to the presence of a large number of low-location-value access points (APs), the fingerprint still contains a lot of redundant information, requiring further AP filtering to refine the information density.

[0138] The specific process of step three is as follows:

[0139] Due to the logarithmic decay characteristic of signals, distant access points (APs) exhibit insufficient strength and severe fluctuations, while nearby APs show strong correlation but weak discriminative power. To address these issues and consolidate valuable fingerprint localization information, a multi-scale AP optimization strategy is proposed to evaluate and select the best APs based on three scales: effectiveness, discriminative power, and measurability, thereby constructing a fingerprint set Ψ with high information density and converging features. Without loss of generality, the following section selects a sub-region... Evaluate the original AP set Q = {AP1, AP2, ..., AP...} M The positioning value of individual APs in}.

[0140] (1) Effectiveness

[0141] An effective target location access point (AP) should provide a reliable and stable signal. Reliability is demonstrated by the AP signal strength meeting the requirements for effective positioning and being captured by different target acquisition points (RPs) within the area; stability is measured by the degree of AP signal fluctuation.

[0142] Given an AP signal strength threshold γ2, if the following conditions are met... Then AP j In RP i Valid mark T i j =1, otherwise T i j =0. Define AP j In the region The regional reliability is:

[0143]

[0144] Select AP j right Inside N k The set of absolute median deviations of each RP It can reflect AP j Signal fluctuation deviation values ​​at different RP points. Calculate the average value. As AP j A reliability metric; the higher the value, the more volatile the AP (Average Per Second) and the lower the stability. Normalization yields the AP. j exist The regional stability is:

[0145]

[0146] To avoid the denominator being 0, the formula must have σ > 0 and σ → 0. Considering both regional reliability and regional stability, AP... j For the current region The validity is defined as:

[0147]

[0148] AP j of The higher the value, the better it is in the region. The more stable and reliable the fingerprint information, the better.

[0149] (2) Discrimination

[0150] AP j The fingerprint recognition capability of an app is determined by its ability to distinguish between different fingerprint receptors (RPs). j The lower the overlap of RSSI measurement fluctuation ranges among all RPs, the higher the accuracy of identifying neighboring RPs when participating in localization. This discriminative power can be measured based on the overlap of RSSI intervals of AP measurements. To account for the difference between the algorithm's value range and the actual fluctuation range, AP is defined as follows: j In RP i The fluctuation range of the nth signal measurement set at the location is

[0151]

[0152] At the reference point and Between, AP j The overlap of the measurement set intervals is expressed using the extended Jaccard coefficient.

[0153]

[0154] In the formula, |·| represents the length of the interval. AP j The final discrimination is taken from the region Mean overlap between all RP pairs It can be seen that if AP j In RP distribution The lower the value, the higher the AP's ability to distinguish RPs during online positioning.

[0155] (3) Measurability

[0156] Considering that APs with stronger signals are easier to detect, the area can be monitored via APs. Average RSSI intensity distribution of internal RP It can effectively reflect the measurability of AP within the region, among which AP j The stronger the measurability, the corresponding The larger the value, the greater the value.

[0157] Based on the discrimination index, it can be determined Internal AP j Comprehensive performance factor in The lower the AP j In the region Among all RPs, the higher the combined performance of its own measurability and its ability to distinguish the nearest neighbor RPs of the TP, the better.

[0158] (4) Information-intensive fingerprint sets

[0159] For sub-regions In the actual positioning process, the stability differences of APs in different regions are taken into consideration first. Therefore, regions with stable and reliable V are selected from the original AP set. k A number of valid APs were obtained. Then, based on the individual localization effect of AP, the measurability and discriminability were weighed through comprehensive performance factors, and the system was further simplified. The number of APs in China up to P k Seeking Further optimization under multiple gain conditions.

[0160] After completing the AP set optimization, the goal is to obtain an information-dense fingerprint set that simultaneously satisfies the sparse recovery condition and reflects the true RSSI distribution characteristics in the scene. In sparse matrix form, combined Fingerprint database simplification and information density agglomeration achieved through matrix multiplication can be characterized as follows: ,in Each vector contains only one 1 element and all others are 0. The index of this non-zero element corresponds to the position of the effective AP in the original AP. Elements are multiplied by matrix multiplication. Information filtering for redundant APs.

[0161] The same optimization strategy is applied to all sub-regions in Φ. The linkage between internal and external regions enables multi-scale fingerprint set simplification based on effectiveness, discriminability, and measurability under the information-intensive driving force, ultimately resulting in an information-intensive fingerprint set.

[0162] The specific process of step four is as follows:

[0163] (1) Real-time position vector recovery

[0164] During the online phase, given y tp And C, to Calculating Euclidean Distance Based on RSSI Signal Variance of AP The result is used as the similarity between TP and each cluster center. Coarse localization is performed to determine the region with the highest similarity as the subregion to which TP belongs. The position vector is obtained by combining equation (1) with the reconstruction algorithm.

[0165] Due to the inherent environmental context of small-scale indoor spaces, fingerprint information exhibits strong correlation. Therefore, algorithms based on greedy iterative approximation and convex optimization linear programming struggle to effectively distinguish and match highly correlated information. Sparse Bayesian Learning (SBL) employs a parameterized Gaussian distribution as the prior distribution of unknown vectors. Under the Gaussian distribution assumption, it estimates the posterior distribution of unknown location vectors through Bayesian inference, effectively constraining the influence of fingerprint correlation. This method leverages this characteristic of SBL to reconstruct and recover the location vectors.

[0166] (2) Reconstruction of position coefficients based on AP information sequence

[0167] Due to environmental obstacles and the similarity of RSSI distributions in spatial locations, the set of non-zero position coefficients θ in f tp This introduces many interfering RP points that are not near the TP. To reduce the influence of interfering RP points, the positional weights of neighboring RP points are increased. The RSSI ranking of APs in the RP fingerprint information is defined as the information sequence. The similarity between RP points and TPs is quantified by ranking the AP strength, and then the positional coefficients are reconstructed using weighted methods.

[0168] right Its corresponding mapping RP in Φ can be obtained. i fingerprint information like exist and observation vector y tp If the RSSI intensity rankings are the same, then μ is considered to be... m,n =1. For all APs, pairwise comparisons yield D = Pk ·(P k -1) / 2 possible outcomes. RP i The similarity to the target location TP can be determined by the similarity coefficient IS of the information sequence. tp,i express:

[0169]

[0170] RP i The closer IS is to TP in spatial location, the better. tp,i The higher the value, the better. θ is calculated according to equation (11). tp The set of information sequence similarity values ​​corresponding to the l mapped RPs is IS = {IS tp,i |1≤i≤l}. During online positioning, the position coefficients are reconstructed by combining IS comparison with weighted factors to obtain the TP position:

[0171]

[0172] Example 3:

[0173] This embodiment provides a novel indoor positioning method for compressing and refining location information, applied to positioning within a teaching building corridor.

[0174] The experimental test scenario was a continuous corridor on a high-rise building, measuring 72m × 62m. Reference points (RPs) were evenly placed every 1m along the corridor path. A total of N = 268 RPs were set up, with each RP sampled n = 30 times at a time interval of 2.3s.

[0175] During the data collection phase, personnel walked around the scene with handheld devices, collecting data at RP points. The devices collected data from a total of 118 different AP base stations (M=118), which were then numbered and stored according to their different MAC addresses. The test data collection process was the same as the training data collection process, collecting a total of 133 TP points for testing.

[0176] To evaluate the impact of adaptive region partitioning and fuzzy mechanisms in the proposed spatial feature-corrected clustering algorithm on indoor positioning performance, three clustering methods—APC, K-Means, and the proposed spatial location-corrected clustering—were tested under the same scene and RP scale. The effectiveness of spatial location correction similarity and region edge fuzziness was verified by comparing the clustering outliers and sub-region partitioning results. After multiple comparisons, the optimal parameters without loss of generality were determined to be U = 10 and γ1 = 0.01.

[0177] Figure 4The main body of the paper presents the spatial feature-corrected clustering results. The K-Means and APC clustering results only show some anomalies, marking the locations of anomalies within the scene for comparison with the spatial feature-corrected clustering. The spatial feature-corrected clustering algorithm divides the test scene into K=7 regions, with the k-th sub-region denoted as Φ. k k∈[1,K]. The region is connected and RP is uniformly distributed, and the corridor extends into the region Φ. 1 Φ 2 Φ 4 Φ 7 The distinction between them is obvious, region Φ 3 Φ 5 Φ 6 Because it is located at the boundary, the blurred regions at both ends intersect. In contrast, since the initial cluster centers of the K-Means algorithm are randomly selected, the clustering results are not sensitive to scene features and will appear in Φ. 2 Φ 3 The case of discrete Resource Links (RPs) interspersed within region clustering. APC does not consider the similarity of RSSI distributions for dissimilar RPs; some RPs may appear in the region Φ marked in the figure. 5 RP is assigned to region Φ 6 The phenomenon of displacement was addressed. By improving similarity to enhance spatial correlation and incorporating a secondary clustering fuzzy discrimination mechanism to improve the margin for selecting region edges (RP), spatial feature-corrected clustering exhibited a clustering effect that better fits the scene characteristics.

[0178] To evaluate the contribution of reducing the RP dimensionality in the clustering stage to subsequent localization performance, in a fair scenario where the AP selection and online localization matching processes are consistent, the localization errors of offline clustering using APC and K-Means were compared. To ensure a fair comparison of the algorithms based on environmental influences and random noise, statistical data from 30 experiments were used as the basis for performance analysis. Figure 5 The experimental results were compared to determine the number of RP location anomalies and the average positioning error of different algorithms.

[0179] Depend on Figure 5 Analysis shows that adding spatial location correction in the clustering process improves the margin of RP selection and enhances the probability of selecting nearest neighbor RP, effectively reducing the upper and lower bounds of the error and reducing the number of individuals with extreme location errors to 2. The spatial location correction clustering algorithm generally outperforms APC and K-Means in terms of positioning performance, with significant reductions in both maximum and minimum errors compared to the other two algorithms. The average positioning error is 0.9667m, which is 0.22m and 0.32m lower than APC and K-Means, respectively.

[0180] Considering the differences between regions and the varying locational contribution value of individual APs in different regions, offline data can be used to complete the analysis of the core parameter V in the fingerprint dataset. k and Pk Optimized configuration. For each sub-region in the scene, the optimal AP set is filtered in two steps, considering the overall effectiveness across regions, and V is selected. k ∈[10,60], within the region, considering both discriminability and measurability, we take P. k ∈[5,V k Finally, the differentiated AP set of the regional fingerprint set was determined. Compared with the original fingerprint database, the RP and AP dimensions of all sub-regions have been significantly reduced, with an average number of 28 and 43 respectively. This effectively reduces the database construction and storage costs of the fingerprint set, and the reduction in data volume can also reduce online computing overhead in the future.

[0181] To evaluate the ability of multi-scale AP selection strategies to aggregate fingerprint set location value information density and improve the location accuracy of CS-based algorithms, this test experiment selected three AP selection algorithms—Fisher, InfoGain, and MaxMean—for simulation comparison. To ensure algorithmic fairness, all four algorithms used the spatial features proposed in this method to correct the clustering results, independently generating their own optimal AP subsets for each region. The localization algorithm selected was SBL. The localization and deviation of all TP point sets were tested, and statistical analysis was performed in the form of cumulative error. Figure 6 As shown.

[0182] The multi-scale AP selection strategy outperforms the other three algorithms in terms of overall average error. When the error is within 1m, the cumulative error probability reaches 63.43%, and within 1.6m, it reaches 87.97%. MaxMean selects APs with larger offline mean values, making it susceptible to high-value anomaly environmental noise, resulting in an average positioning error of 1.191m. InfoGain evaluates APs from the perspective of maximizing information gain, failing to guarantee the effectiveness of APs in different regions, resulting in an average positioning error of 1.222m. Fisher combines historical information to ensure AP stability but ignores the AP's ability to distinguish RPs, resulting in an average positioning error of 1.301m. The proposed method considers the overall value of APs in the region and the value of matching and locating individual APs, effectively filtering redundant AP information from the fingerprint set during the compressed sampling stage. The average positioning error is 0.967m, and the overall positioning performance is improved by 23.2%, 26.1%, and 34.5% compared to MaxMean, InfoGain, and Fisher, respectively.

[0183] Location coefficient reconstruction can constrain the influence of the environment on the RSSI distribution and reduce the weight of non-nearest relative points (RPs). To verify the effectiveness of the IS coefficients, location coefficient reconstruction was performed on the recovered location vector f for each TP point. Figure 7 Selected from Φ 4 Φ 6 Φ 7 TP7, TP 46 and TP107 Compare the changes in RP weights before and after IS coefficient reconstruction.

[0184] Depend on Figure 7 Taking TP7, which has many surrounding interference RPs, as an example, the neighboring RPs in its area 11 weight θ 11 Increased by 0.277, interference point RP 20 RP 29 RP 31 weight θ 20 θ 29 θ 31 These were reduced by 0.123, 0.056, and 0.077 respectively. TP 46 TP 107 With fewer interfering RPs, the weight change ratio is relatively low. It is evident that within the position vector f reconstructed from the IS information sequence, the position coefficient θ corresponding to the nearest neighbor RP of TP is effectively improved, and the influence of interfering points is effectively suppressed, verifying the effectiveness of the algorithm.

[0185] To verify the localization performance of the FIDA algorithm, the method selected OMP and BP to compare and analyze the differences in the ability of different CS reconstruction algorithms to handle fingerprint correlation. At the same time, WKNN and its improved algorithm VWKNN were added as a control group for uncompressed sensing localization algorithms to measure the differences in localization performance of algorithms in different fields. (For the OMP method, please refer to Li Xiaoqiang, Chen Jianfeng, Zhang Rongrong, Wen Yang, Tan Weijie. Target localization algorithm in wireless sensor network based on fast orthogonal matching pursuit [J]. Journal of Northwestern Polytechnical University, 2020, 38(01).) :31-39. For the BP method, please refer to Zheng Qian, Hu Jiusong, Liu Hongli, et al. Analysis of positioning performance of indoor positioning system based on compressed sensing [J]. Journal of Chongqing University of Posts and Telecommunications (Natural Science Edition), 2018, 30(06):768-775. For the VWKNN method, please refer to Xu Tian, ​​He Jingsha, Zhu Nafei, Deng Wanhang, Wu Shuang, Ta Yongjun. VWKNN location fingerprint positioning algorithm based on discrete coefficient improvement [J]. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48(07):1242-1251.). FIDA, OMP and BP all adopt the clustering and AP screening process proposed by the method and adaptively select the optimal parameter to ensure the fairness of comparison. WKNN and VWKNN use the parameter K=4 when the average positioning error is the lowest. The cumulative error probability and position error of the four algorithms are shown in Figure 8 .

[0186] Depend on Figure 8It is evident that, due to the weak fingerprint correlation handling capability of traditional CS-type algorithms and the impact of poor signal propagation conditions in corridors, the positioning performance of OMP and BP is consistently weaker than the other three algorithms. FIDA, however, effectively processes redundant information through the fingerprint set establishment and simplification stages, and introduces SBL to weaken fingerprint correlation, surpassing CS algorithms in the same domain. Its cumulative error probability is consistently higher than OMP and BP, reaching 64.66% within 1m and 92.48% within 2m. Minor noise interference causes FIDA's cumulative error probability to be higher than VWKNN within 1m, but its overall positioning performance is better than WKNN and VWKNN. Table 1 shows the position estimation errors of the four algorithms. FIDA has significantly lower average positioning error, maximum error, and error variance than the other three algorithms, reaching 0.967m, 4.161m, and 0.686m, respectively. The comprehensive experimental results show that FIDA, through the close integration of bidirectional fingerprint set simplification and fingerprint correlation weakening, consolidates high-density positioning information and employs a more effective positioning method, significantly improving positioning performance.

[0187] Table 1. Position estimation errors of the four algorithms

[0188]

[0189] Some steps in the embodiments of the present invention can be implemented using software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk.

[0190] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An indoor positioning method, characterized in that, include: First, a positioning system model is constructed based on fingerprint positioning methods and compressed sensing technology; Then, spatial location features are used to complete the fingerprint segmentation of fuzzy regions and effective AP screening is performed based on multiple scales; Finally, the target position is calculated, and the position information of the point to be located is sparsely reconstructed. The indoor positioning method includes: Step 1: Construct a CS-based positioning system model; Step 1.1: Set the area to be located One AP signal source and RP reference points; located at a certain reference point At this location, the RSSI matrix originating from the AP signal source was measured. ,in The initial fingerprint dataset is as follows: ; remember For the first One AP signal source The measurement vector is ,in ; for middle The set of absolute median deviations of APs, where yes The absolute median deviation; for After being divided fingerprint sets of each region For the first The number of RPs in each region; [record] middle For the corresponding Cluster centers; lie in At test point TP, the received measurement vector from the AP signal source is: ; Step 1.2: Using the following formula, obtain the position vector of the test point TP through a reconstruction algorithm. : in, The number of AP signal sources after filtering. Noise figure For the measurement matrix, It is a sparse basis matrix, i.e., a set of regional fingerprints; The sensing matrix in CS is the information-dense fingerprint set. To pass through the sensing matrix And the low-dimensional observation vector measured and processed at point TP; Step 2: Addressing the measurement uncertainties and spatial non-uniformity of the RSSI signal, the initial fingerprint dataset... By sequentially performing RSSI uncertainty filtering and region segmentation, a stable and effective regional fingerprint set is established. ; Step 3: Perform multi-scale screening of effective APs based on validity, discriminability, and measurability to obtain the information-dense fingerprint set. ; Step 4: Calculate the target location and sparsely reconstruct the location information of the point to be located.

2. The indoor positioning method according to claim 1, characterized in that, Step 2 includes: Step 2.1: Filtering out signal uncertainties; Given the reference point For the first One AP signal source Secondary measurement vector , obtain signal source exist corresponding median and absolute median deviation To fit the RSSI distribution in a real-world scenario, it is determined that... Out-of-range RSSI values ​​are considered outliers and are filtered out; the processed values ​​are then... average As valid fingerprint data, it is obtained fingerprint information All RPs form a fingerprint database. ; Step 2.2: Spatial feature correction region division; Step 2.2.1: Calculate the set of dissimilarity similarities between paired reference points RP, given reference points. , fingerprint information and By introducing the spatial location feature weights of RPs as scene factors to construct the difference similarity between RPs, the influence of data bias on the original similarity is corrected, and a reference point is defined. and The similarity is: In the formula, Let be the spatial location feature weights, and satisfy the following condition: If and The span of the spatial distance of the remaining RPs is divided into Each interval and spatial distance Located in the The intervals are as follows: In the formula, , , ; Step 2.2.2: Merge RP spatial features to obtain cluster centers The initial regional division is carried out according to criterion 1; Step 2.2.3: Determine the cluster edge points according to criterion 2, and update the fuzzy region determination of the initial region fingerprint set; The first criterion is: ; The second criterion is: ; in, For similarity threshold, They are respectively regions and The regional center; By using the aforementioned criteria 1 and criterion 2 Divided into fingerprint collection in each region ,in For the first k The number of RPs contained in each subregion.

3. The indoor positioning method according to claim 2, characterized in that, The validity calculation process in step 3 includes: Given AP signal strength threshold If satisfied , but exist Valid mark ,otherwise ;definition In the region The regional reliability is: Select right Inside The set of absolute median deviations of each RP ,reflect Calculate the average value of the signal fluctuation deviation at different RP points. As The higher the reliability metric value, the more drastic the AP fluctuations and the lower the stability. Normalization yields exist The regional stability is: In the formula and ; In summary, considering both regional reliability and regional stability, For the current region The validity is: of The higher the value, the better it is in the region. The more stable and reliable the fingerprint information, the better.

4. The indoor positioning method according to claim 3, characterized in that, The process of calculating the discrimination in step 3 includes: definition exist place The fluctuation range of the secondary signal measurement set is : At the reference point and between, The overlap of the measurement set intervals is represented by the extended Jaccard coefficient: In the formula, This indicates taking the length of the interval. The final discrimination is taken from the region Mean overlap between all RP pairs .

5. The indoor positioning method according to claim 4, characterized in that, The measurability calculation process in step 3 includes: via AP to the area Average RSSI intensity distribution of internal RP Reflecting the measurability of AP within the region, among which , The stronger the measurability, the corresponding The larger the value.

6. The indoor positioning method according to claim 5, characterized in that, Step 3 also includes: For sub-regions Based on the overall stability differences of APs across different regions, regions with stable and reliable APs are selected from the original AP set. A number of valid APs were obtained. Then, based on the individual positioning effect of AP, the measurability and discriminativeness are weighed through comprehensive performance factors, and the system is further simplified. Number of APs in China Seeking To achieve further optimization under multiple gain conditions; Find an information-dense fingerprint set that simultaneously satisfies the sparse recovery condition and reflects the true RSSI distribution characteristics in the scene. : In sparse matrix form, combined The fingerprint database is simplified in size and its information density is amplified through matrix multiplication, which is characterized as follows: ,in Each vector contains only one 1 element and all others are 0. The index of this 1 element corresponds to the position of the effective AP in the original AP. The elements are multiplied by matrix multiplication. Information filtering of redundant APs; right The same optimization strategy is applied to all sub-regions, and the linkage between internal and external regions enables multi-scale fingerprint set simplification based on effectiveness, discriminability, and measurability under the information-intensive driving force, ultimately resulting in an information-intensive fingerprint set. .

7. The indoor positioning method according to claim 6, characterized in that, Step 4 includes: Step 41: Reconstruction and recovery of real-time position vector; During the online phase, a measurement vector is received from the signal source AP at a given test point TP. and cluster center ,right Calculation of Euclidean Distance Based on RSSI Signal Variance of AP The results are used as the similarity between TP and each cluster center for coarse localization. The region with the highest similarity is determined as the sub-region to which TP belongs. The TP location vector is then calculated using the following formula combined with the sparse Bayesian algorithm. Reconstruction and recovery: Step 42: Define the RSSI order of APs in the RP fingerprint information as the information sequence, quantify the similarity between RP points and TPs by AP intensity ordering, and then reconstruct the position coefficients by weighting. right to obtain its Corresponding mapping fingerprint information ,like exist and observation vector If the RSSI intensity rankings are the same, then it is considered that... ; All APs can be compared pairwise. This result, The degree of similarity to the target location TP is determined by the similarity coefficient of the information sequence. express: The closer it is to TP in spatial location, the better. The higher the value; Calculate according to the following formula Mapped The set of information sequence similarities corresponding to each RP When using online positioning, combine By comparing and weighting the position coefficients, we can reconstruct the position coefficients to obtain the TP position: 。 8. An indoor positioning system, characterized in that, The indoor positioning system includes: At least one storage medium storing at least one instruction set for locating a point to be located; and at least one processor communicatively connected to the at least one storage medium. When the indoor positioning system is running, the at least one processor reads the at least one instruction set and implements the indoor positioning method as described in any one of claims 1-7.

Images