A method for indoor positioning in multi-signal scenarios

Abstract

The present invention discloses an indoor positioning method in a multi-signal scenario, belonging to the field of indoor positioning. The method completes the construction and expansion of the fingerprint library in the offline stage, and uses the fingerprint clarity to characterize the degree of change of the path loss index, and combines the density peak clustering to realize the adaptive division of the target area; in the online stage, the clarity weight and the received signal strength RSS vector Euclidean distance weight are integrated to obtain the combined weight of each reference point, and the first E reference points with the highest combined weight are selected to participate in the target position calculation to obtain the estimated value of the target position. This application introduces the concept of fingerprint clarity, takes the regional fluctuation characteristics of the signal as the measurement standard, clusters the reference points and fits the sub-regional path loss index, effectively solving the problem of poor precision of single path loss simulation; further, this application uses fingerprint clarity as the second standard for selecting high-quality reference points, and integrates the RSS vector Euclidean distance to ensure the accuracy of position calculation.

Description

Technical Field

[0001] The invention relates to an indoor positioning method in a multi-signal scenario, and belongs to the field of indoor positioning. Background Art

[0002] Location Based Service (LBS) provides comprehensive and functional services guided by location or with added value based on the location information of mobile terminal users. In outdoor environments, the Global Position System (GPS) can provide users with accurate latitude and longitude information, but due to building obstructions and overlapping interference of channel multipath, GPS cannot accurately measure distance and position indoors. Therefore, the value of exploring and researching indoor positioning technology is becoming increasingly prominent.

[0003] Considering the ubiquitous Wi-Fi signals in indoor areas, which are easy to obtain, propagate far, and are widely distributed, fingerprint positioning technology based on Wi-Fi information has become a key research direction for indoor positioning. In the offline stage, this technology collects the Wi-Fi signals of the reference points (RP) in the positioning area as a fingerprint library. In the online stage, it calculates the estimated position of the target by matching the real-time positioning signals with the fingerprint library signals, thus achieving accurate positioning under indoor conditions.

[0004] At present, the mainstream method to reduce the cost of deploying fingerprint libraries is to simulate a large number of unsampled reference points in the target area by interpolating a small number of sampled reference points. Based on this idea, the geometric interpolation method is chosen to build a global fingerprint library. This method is simple and easy, but its nonlinear property of the received signal strength (RSS) represented by linear distance will inevitably reduce the interpolation accuracy. The statistical modeling method overcomes this shortcoming. The Kriging algorithm defines the sample weights based on the minimum variance and unbiased estimation, and reshapes the relationship between the physical space distance and the signal space distance of the sampling points by selecting a suitable variation function. However, for irregular spaces such as corridors and heterogeneous corners, the isotropy assumption of Kriging is not applicable, so its positioning accuracy will be affected.

[0005] In addition to the impact of the fingerprint construction method on positioning accuracy, whether the high-quality reference points in the fingerprint library can participate in the calculation is also an important factor affecting positioning accuracy. For different estimated positions, different reference points have different resolution capabilities. Involving all reference points in the expanded fingerprint library in positioning with the same candidate probability may have an adverse effect on positioning accuracy. In order to select high-quality reference points, the existing method uses the WKNN positioning algorithm to reversely weight the reference points according to the Euclidean distance between the reference point signal strength and the real-time RSS vector, and selects high-weight reference points to participate in the calculation. However, the particularity of the expanded fingerprint library makes it impossible for the reference point selection algorithm using Euclidean distance as the single measurement standard to bring about a significant improvement in accuracy. Summary of the invention

[0006] In order to further improve the indoor positioning accuracy in multi-signal scenarios, the present invention provides an indoor positioning method in multi-signal scenarios, takes the regional fluctuation characteristics of the signal as the measurement standard, introduces the concept of fingerprint clarity, clusters the sample reference points and fits the sub-regional path loss index, effectively solving the problem of poor accuracy of single path loss simulation; and when selecting high-quality reference points, fingerprint clarity is used as the second criterion to ensure the accuracy of position calculation.

[0007] An indoor positioning method in a multi-signal scenario, the method comprising an offline phase and an online phase;

[0008] The fingerprint library is constructed and expanded in the offline stage, and the change degree of the path loss index is characterized by the fingerprint clarity. The adaptive division of the target area is realized by combining the density peak clustering. The expanded fingerprint library is called the offline interpolation fingerprint library. The fingerprint clarity refers to the reference point RP. j The ratio of the sum of the signal differences to the adjacent reference points to the number of adjacent reference points; the comprehensive clarity weight wd in the online stage j and received signal strength RSS vector Euclidean distance weight wo j , get the reference point RP j The first E reference points RP with the highest combined weights are selected to participate in the target position calculation to obtain an estimated value of the target position, where the target is the point to be located; and the target area is the indoor area where the point to be located is located.

[0009] Optionally, the method comprises:

[0010] Step 1: Build a system model;

[0011] Arrange N wireless access points AP in the plane target space i , M reference points RP j ; Among them, the M reference points include R sampled reference points SRP r and Z reference points to be interpolated IRP z; Among them, 1≤i≤N, 1≤j≤M, 1≤r≤R, 1≤z≤Z;

[0012] Step 2: Fingerprint data preprocessing;

[0013] Describing SRP based on signal aggregation and frequency r From AP i signal;

[0014] Step 3: Expand the fingerprint database;

[0015] Step 3.1: Cluster and partition the sampled reference point SPR;

[0016] According to SRP r Its neighbor SRP r1 The signal difference and distance difference determine the SRP r From AP i Fingerprint clarity

[0017] According to each SRP r Fingerprint clarity And plane position Jointly construct the SRP mapping three-dimensional space and calculate any SRP r and SRP r1 The space between r,r1 ;

[0018] Calculate SRP according to formula (4) r The local density of the nearest neighbors at is:

[0019]

[0020] Among them, the cutoff distance f cut It is calculated by the ratio of the average expected number of neighboring SRPs of each SRP in the space to the total number of SRPs;

[0021] According to the local density ρ of each SRP neighbor r Sort each SRP and let {q1,q2,…,q c ,…q C} is the local density ρ of each SRP neighbor r The descending sequence number, where C = R, then SRP r The parent distance at is expressed as:

[0022]

[0023] Where ID(·) represents the variable number that satisfies the expression ·;

[0024] At this point, each SRP rThere is a pair of (ρ r ,θ r );

[0025] According to each SRP r The corresponding pair (ρ r ,θ r ) Clustering and partitioning all sampled reference point SPRs;

[0026] Step 3.2: SRP r The signal characteristics at the location are used as the initial signal strength, considering SRP r With IRP z Signal propagation loss between the two nodes, establish IRP z Relative to SRP r The difference neighbor signal model is used; and the interpolation reference point IRP is treated according to the unsupervised K nearest neighbor classification algorithm. z Partitioning; combining IRP z The SRP in the zone observed by the AP i The signal strength of the difference between the SRP and the neighbor signal model, for IRP z Signal strength Make predictions;

[0027] Step 4: Select high-quality reference points based on fingerprint clarity to calculate the target position.

[0028] Optionally, step 1 further includes:

[0029] definition Indicates the reference point RP j The received signal strength RSS value from N APs received at the location; Indicates the sampled reference point SRP r The received signal strength RSS value received at the location; Indicates the reference point to be interpolated IRP z The received signal strength RSS value received at the location;

[0030] definition represents the spatial position set of reference points RP, where Represents RP j The two-dimensional plane coordinates of Indicates SRP r The two-dimensional plane coordinates of represents the set of spatial locations of APs, where Indicates AP i The two-dimensional plane coordinates of

[0031] definition Represents the RSS vector of the point to be located measured in the online stage, where: It indicates the received signal strength RSS value of the i-th wireless access point AP received by the point to be located measured in the online stage.

[0032] Optionally, step 2 includes:

[0033] Step 2.1: Within a given unit time T, the sampled reference point SRP r Collected from AP i The H signal in Indicates SRP r Collected from AP i The hth signal of

[0034] Step 2.2: For the possible Q different RSS value samples, there is an SRP r About the Probability Distribution Set of RSS Its mean and median are μ r and δ r ;

[0035] Step 2.3: Describe the SRP based on the signal concentration and frequency r From AP i The signal

[0036]

[0037] in, is the original measurement set The average value μ is selected from r H is the threshold t A high frequency signal that satisfies H t ≤H.

[0038] Optionally, the SRP in step 3.1 r From AP i Fingerprint clarity The calculation formula is:

[0039]

[0040] in, Indicates SRP r The signal strength value of the reference point corresponding to the r1th number in the set A composed of the neighboring reference point numbers, Num(·) represents the number of elements contained in the set; Indicates SRP r The two-dimensional plane coordinates of Indicates SRP r1 The two-dimensional plane coordinates of .

[0041] Optionally, any SRP in step 3.1 r and SRP r1 The space between r,r1 Expressed in Euclidean distance, the calculation formula is as follows:

[0042]

[0043] in, Indicates SRP r1 From AP i Fingerprint clarity.

[0044] Optionally, the step 3.2 includes:

[0045] Calculate the sampled reference point SRP r From AP i Signal prediction value:

[0046]

[0047] in, Indicates the distance to the AP i The signal strength loss value at the default distance dis0, dis r Indicates SRP r With AP i spatial distance;

[0048] Calculate the reference point IRP to be interpolated z Relative to the sampled reference point SRP r The difference neighbor signal of:

[0049]

[0050] Investigation distance IRP z The classification of the most recent K SRPs, IRP z Assigned to the partition with the highest SRP share Inside;

[0051] make For partition Included R k The number sequence of SRPs, combined with the data from AP observed by SRP in the partition i The signal strength of the difference between the SRP and the neighbor signal model, for IRP z Signal strength Make a prediction:

[0052]

[0053] in, is the sampled reference point The corresponding path loss index is based on the same The signal strength of the known SRP in the partition is fitted with the distance between APs.

[0054] Optionally, step 4 includes:

[0055] Step 4.1: Obtain the RSS value of the received signal strength of all wireless access points AP received by the point to be located, form the RSS vector L of the point to be located, substitute the vector L of the point to be located into the position of each RP in the offline interpolation fingerprint library, and calculate the value of L at the RP. j From AP i Fingerprint clarity And with RP j Original fingerprint clarity Compare, set γ as the clarity difference threshold, if it satisfies Then RP j Join L's About AP i RP matching set

[0056] Step 4.2: Follow RP j The average clarity of all SRPs in the partition The pre-selected reference points in the set are sorted in descending order; the pre-selected reference points are given corresponding weights according to the average clarity and clarity difference.

[0057]

[0058] in, RP j The average clarity of the partition, σ is the clarity difference magnification factor;

[0059] Step 4.3: Let R s = {r s |1≤s≤S} represents the RP in step 4.1 j Available The corresponding AP number set; then each reference point RP j The total clarity weight wd j for:

[0060]

[0061] According to the obtained clarity weight wd j , take the first B RPs with the highest clarity weight as the support vector to be located L and the final RP set S L ;

[0062] Step 4.4: Calculate S LThe Euclidean distance weight of the inner RSS vector wo j , obtained by taking the inverse weighted Euclidean distance from L to the RSS vector of each reference point;

[0063] Step 4.5: Synthesis S L The clarity weight of the internal RP wd j and the RSS vector Euclidean distance weight wo j , get RP j The combination weights of the first E RPs with the highest combination weights are selected to participate in the target position calculation:

[0064]

[0065] Among them, X j Represents the spatial position of the jth reference point RP.

[0066] Optionally, the RP j The combined weight wc j The calculation formula is:

[0067]

[0068] Among them, 1≤b≤B.

[0069] The beneficial effects of the present invention are:

[0070] By introducing the concept of fingerprint clarity, using the regional fluctuation characteristics of the signal as a measurement standard, clustering the sample reference points and fitting the sub-regional path loss index, the problem of poor precision in the simulation of single path loss in the prior art is effectively solved; further, the present application uses fingerprint clarity as the second standard for selecting high-quality reference points, and integrates the Euclidean distance of the RSS vector to ensure the accuracy of position calculation. Furthermore, the present application generates data of other points by collecting a small number of sample points using interpolation, without the need to collect data of global reference points, thereby reducing the labor and time costs of collecting data and improving positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0072] Figure 1A This is a simulation diagram of the prediction results of the signal at the interpolation reference point IRP using a single model in the existing technology;

[0073] Figure 1BThis is a simulation diagram of the prediction results of the signal at the interpolation reference point IRP using the nearest neighbor model proposed in this application.

[0074] Figure 2 This is a flow chart of the fingerprint library expansion indoor positioning method DIFS in a multi-signal scenario proposed by the present invention.

[0075] Figure 3 The corresponding diagram of signal strength fluctuation and fingerprint clarity. Among them, (a) is the signal strength fluctuation diagram, and (b) is the fingerprint clarity diagram.

[0076] Figure 4A This is a schematic diagram of AP2 clustering effect;

[0077] Figure 4B Schematic diagram of AP3 clustering effect.

[0078] Figure 5A-5F Figure 2 is a schematic diagram of the residual box of the six APs under four interpolation algorithms. Figure 5A is a schematic diagram of AP1 residual. Figure 5B is a schematic diagram of AP2 residual. Figure 5C is a schematic diagram of AP3 residuals, Figure 5D This is a schematic diagram of AP4 residuals. Figure 5E This is a schematic diagram of AP5 residuals. Fig. 5F Schematic diagram of AP6 residual.

[0079] Figure 6 The following is a simulation diagram comparing the error accumulation probabilities of target positioning using the method of the present application and three existing methods. DETAILED DESCRIPTION

[0080] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0081] Embodiment 1:

[0082] This embodiment provides an indoor positioning method in a multi-signal scenario. Figure 2 , the method comprising:

[0083] Step 1: Construction of system model;

[0084] In the plane target space, N wireless access points (AP) are arranged, represented by AP = {AP i |1≤i≤N}, and R sampled reference points (Sample Reference Point, SRP), represented by SRP = {SRP r|1≤r≤R}, Z interpolated reference points (IRP), represented by IRP = {IRP z |1≤z≤Z}. SRP and IRP together form M reference points (Reference Point, RP), expressed as RP = {RP j |1≤j≤M}.

[0085] definition Represents RP j The received signal strength RSS value from N APs is defined as Indicates SRP r The RSS value received by the location, Represents IRP z The RSS value received at the location.

[0086] definition represents the spatial location set of RP, where Represents RP j The two-dimensional plane coordinates of Indicates SRP r The two-dimensional plane coordinates of represents the set of spatial locations of APs, where Indicates AP i The two-dimensional plane coordinates of . Indicates the RSS vector of the point to be located measured during the online phase.

[0087] Step 2: Fingerprint data preprocessing;

[0088] In multi-structure indoor partition environments such as circular corridors, multipath and shadow interference of wireless signals will show various changes such as superposition and repetition. Therefore, using traditional Gaussian distribution to fit RSS will simplify the research while bringing about large data deviation.

[0089] In a given unit time T, SRP r Collected from AP i The H signal For the possible Q different RSS value samples, there is an SRP r About the Probability Distribution Set of RSS Its mean and median are μ r and δ r .

[0090] Given that signal distribution characteristics can better reflect the universal characteristics of random signals, the probability distribution of RSS sets can be used as a scale to differentiate and describe the true characteristics of signals in a diverse environment, reducing the interference of individual outlier signals on the overall match. The outlier signal refers to a fingerprint signal whose signal characteristics do not match the true position. For high-frequency and stable signals, replacing the discrete probability description of the original individual with its characteristic distribution can filter out low-frequency and unstable signals to a certain extent.

[0091] For random signals, if the SRP r If the signal is stable, it comes from AP i The RSS value of fluctuates slightly and the values ​​are clustered, which will be reflected in probability as follows Some signal values ​​appear stably with high frequency, raising the mean of probability distribution, which is reflected in the data as μ r ≥δ r Therefore, without loss of generality, it is believed that the current high-frequency signal can represent SRP r On the contrary, if SRP r The signal has a high randomness, and the RSS value is naturally divergent. The signal values ​​in the SRP are decentralized. r The original signal has a similar predictive value, and it is more suitable to use the overall signal value mean to represent SRP. r Therefore, based on the aggregation and frequency of signals, SRP is re-described r From AP i Signal:

[0092]

[0093] in, is the original measurement set The average value μ is selected from r H is the threshold t A high frequency signal that satisfies H t ≤H.

[0094] Step 3: Fingerprint database expansion;

[0095] (1) Fingerprint clarity matching clustering

[0096] In positioning and ranging, a consistent path loss model is often used to associate the received signal power with the distance, so as to estimate the target distance by signal strength. However, the mutual coupling interference between multiple signal sources makes the path loss of the same AP in different areas or different APs in the same area inconsistent, and the global path loss index is also affected by the height of the AP. Due to this volatility, the Wi-Fi signals from multiple APs in a complex indoor environment cannot be fitted with a fixed or single path loss model; on the other hand, simply dividing the target area and selecting the optimal path loss index based on the signal change characteristics is not theoretically convincing.

[0097] Therefore, in order to accurately and objectively fit the distribution of Wi-Fi signals in complex indoor environments, this application introduces the concept of fingerprint clarity, which indirectly reflects the degree of change of the path loss index, and combines density peek classification (DPC) to achieve adaptive division of the target area.

[0098] 1) Fingerprint clarity

[0099] Fingerprint clarity refers to the reference point RP j The ratio of the sum of the signal differences to the adjacent reference points to the number of adjacent reference points; SRP r Can be based on adjacent SRP r1 The signal difference and distance difference determine the SRP r From AP i Fingerprint clarity As shown in formula (2), it indirectly reflects that different SRPs in the sample area r The signal fluctuation degree of the location to reflect the regional changes in the path loss exponent.

[0100]

[0101] In the above formula, Indicates SRP r The signal strength value of the reference point corresponding to the r1th number in the set A composed of the neighboring reference point numbers, Num(·) represents the number of elements contained in the set. Adaptive adjustment of the capacity of the set A can change the SRP r The range of clarity of perception of surrounding space.

[0102] Generally speaking, the signal difference and distance difference between SRPs in areas with similar signal changes should be close to each other, so that the fingerprint clarity between reference points also tends to be consistent. Therefore, it is generally believed that the path loss index at reference points with similar fingerprint clarity also has a high similarity. Therefore, the clustering based on signal fluctuation discrimination and path loss index characteristics can be transformed into a clustering process based on fingerprint clarity.

[0103] 2) Density peak clustering

[0104] The traditional fingerprint clustering method is changed from partitioning based on all received signal features of SRP to clustering and partitioning based on the similarity of fingerprint clarity of signals sent by a certain AP at different SRPs. Due to the complexity of indoor building structures and materials and the influence of occasional measurement errors, there will inevitably be extreme cases where the signal strength of a certain SRP has no obvious correlation with the surrounding signal strength values. For such abnormal clarity fingerprint points, in order to prevent partitioning errors, the spatial correlation between reference points should also be used as one of the clustering indicators. Therefore, the algorithm generalizes the reference points in three dimensions according to the fingerprint clarity and plane coordinates of the SRP to achieve clustering and partitioning.

[0105] For the sampled reference point set SRP to be clustered, each SRP r Fingerprint clarity And plane position The three-dimensional mapping space of SRP is constructed jointly. Then any SRP r and SRP r1 The space between r,r1 It can be expressed as Euclidean distance, as shown in formula (3).

[0106]

[0107] In the above formula, in order to ensure and The consistency of the impact on clustering results will be Convert it into a unit value constrained by the extreme difference of fingerprint clarity, It is converted into a unit value constrained by the extreme difference of the plane position, so as to realize the unified processing of clustering parameters at different scales.

[0108] SRP r The local density of the nearest neighbors at is:

[0109]

[0110] Among them, the cutoff distance f cut It is calculated by the ratio of the average expected number of neighboring SRPs of each SRP in the space to the total number of SRPs.

[0111] Let {q1,q2,…,q c ,…q C} is the local density ρ of each SRP neighbor r The descending sequence number, where C = R, then SRP r The parent distance at can be expressed as:

[0112]

[0113] Where ID(·) represents the variable number that satisfies the expression ·. So far, each SRP r There is a pair of (ρ r ,θ r ).

[0114] 3) SRP Clustering

[0115] In three-dimensional space, SRP r The local density ρ r It is positively correlated with the number of SRPs with similar signal fluctuations around it, so a high ρ with more similar SRPs r SRP value r , its fingerprint clarity can better reflect the signal change characteristics of most SRPs in the area. At the same time, θ r Describes SRP r The spatial interval of the high local density SRP with the closest distance. Therefore, in order to meet the characteristics of SRP aggregation at the cluster center and dispersed distribution between cluster centers, the local density ρ should be selected r Greater than the upper quarter digit, parent distance θ r The SRPs that are greater than the upper quartile are taken as cluster centers.

[0116] Due to the heterogeneity of AP devices in most indoor scenarios, the signal fluctuations from different APs vary. In order to make the path loss index of different partitions reflect the signal change characteristics within the partition as much as possible, the number of partitions should be able to adaptively follow the changes in environmental signals. This method ensures that the coarse-grained SRP clustering center has the characteristics of high local density and long parent distance, while the inter-class distance Γ between each partition r,r1 It should also not be lower than the high discrimination threshold ν, which is 0.05 times the maximum distance between the partition classes to filter out the cluster centers of the high coupling classes.

[0117] (2) Signal prediction

[0118] The traditional path loss model directly predicts the signal strength through the signal characteristics of the reference points around the AP and the distance between the AP and the interpolation point. However, when there are obstacles such as corners and walls in the signal propagation path, the signal fluctuation is aggravated and the fitting error jumps significantly. In order to avoid the interference of unknown factors between the AP and the IRP on the prediction results, the algorithm uses the RSS characteristics at the SRP as the initial signal value of the path loss model to strengthen the influence of the SRP signal strength in the model prediction. At the same time, it constructs the difference expression of the relationship between the IRP and SRP signal strength and distance, replacing the traditional path loss model and indirectly eliminating the influence of obstacles.

[0119] Figure 1A and Figure 1B A comparison between the traditional path loss model prediction and the nearest neighbor interpolation prediction is given. Figure 1A Select AP i A single neighboring SRP interpolates IRPs at different locations. Figure 1B The signal at the IRP is predicted and calculated using the SRP in the same partition as the IRP, ensuring that the IRP and SRP have similar signal strength fluctuation characteristics while effectively avoiding interference from factors such as walls during signal propagation.

[0120] 1) Difference neighbor signal model

[0121] If the reference point has been sampled, SRP r From AP i The signal prediction uses the traditional path loss model:

[0122]

[0123] in, Indicates the distance to the AP i The signal strength loss value at the default distance dis0, dis r Indicates SRP r With AP i spatial distance.

[0124] In indoor environments, there is often a physical barrier between the signal propagation of the AP and the point to be interpolated, which results in a large deviation in the interpolation prediction based on the RSS signal strength at the AP on a logarithmic distance scale. Since indoor positioning focuses on the relative amount of signal strength, the traditional path loss model can be constructed as the IRP of the point to be interpolated. z With the sampled reference point SRP r The difference in signal strength is expressed in terms of SRP r The signal characteristics at the location are used as the initial signal strength, and only SRP needs to be considered r With IRP z The signal propagation loss between the two. At this time, IRPz The predicted signal strength at the SRP depends more on the signal characteristics at the neighboring SRP. z Relative to SRP r The difference neighbor signal model expression is as follows:

[0125]

[0126] 2) Prediction interpolation

[0127] To ensure IRP z The signal fluctuation characteristics are similar to those of the neighboring SRPs involved in the interpolation calculation. z It also performs partition classification. z The signal prediction of IRP depends on its neighboring SRP. z The partitioning follows the "herd" principle similar to the unsupervised K nearest neighbor classification algorithm. That is, the distance IRP z The classification of the most recent K SRPs, IRP z Assigned to the partition with the highest SRP share Inside.

[0128] Prediction with a single SRP may result in interpolation errors caused by signal drift. All SRPs in the IRP z The above drift is smoothed by taking the mean of the interpolation results. For partition Included R k The number sequence of SRPs, combined with the data from AP observed by SRP in the partition i The signal strength of the difference between the SRP and the neighbor signal model, for IRP z Signal strength Make a prediction:

[0129]

[0130] in, is the sampled reference point The corresponding path loss index can be obtained from the same The signal strength of the known SRP in the partition is fitted with the distance between APs.

[0131] Step 4: Reference point optimization and position calculation.

[0132] Traditional positioning uses Euclidean distance matching between offline fingerprint database and online RSS. However, due to uncertain factors such as signal fluctuation and measurement error, there will be inverse RSS vector pairs with long spatial distance but short Euclidean distance, resulting in reduced positioning accuracy. At the same time, the signal strength difference between the interpolated fingerprint database and the real fingerprint database will increase the probability of the occurrence of inverse vector pairs.

[0133] Considering that fingerprint clarity can not only describe the ability of RSS signal at RP to distinguish the position, but also can be used as one of the indicators to distinguish RP in different partitions, a reference point screening and positioning algorithm based on fingerprint clarity matching is proposed. By matching fingerprint clarity, abnormal reference points are eliminated, and high-resolution RP subsets are retained, the resolution of the interpolated fingerprint library is improved.

[0134] (1) RP screening

[0135] Since the calculation of fingerprint clarity needs to comprehensively consider the information of the RP itself and its neighboring RPs, that is, it has regional correlation. When the RSS vector L to be located and the RSS vector stored at a certain RP in the fingerprint library are an anisotropic vector pair, the fingerprint clarity calculated using L as the RSS at the RP point will be significantly different from the fingerprint clarity calculated using the original RSS, that is, vector L is close to the RP in RSS, but far away from the RP in fingerprint clarity. Based on this regional correlation, the anisotropic RP of vector L is eliminated, and the reference point set is screened and corrected. The specific steps are as follows:

[0136] Step 1: Substitute the vector L of the point to be located into the position of each RP in the offline interpolation fingerprint library and calculate the position of L at the RP. j From AP i Fingerprint clarity And with RP j Original fingerprint clarity Compare, let γ be the clarity difference threshold, if it satisfies Then RP j Join L's About AP i RP matching set

[0137] Step 2 Follow RP j The average clarity of all SRPs in the partition The pre-selected reference points in the set are sorted in descending order. The pre-selected reference points are given corresponding weights based on the average clarity and clarity difference.

[0138]

[0139] in, RP jThe average clarity of the partition, σ is the clarity difference amplification factor. The selection of the amplification factor depends on the placement granularity of the RP and the degree of signal fluctuation, and is set according to the actual environment.

[0140] Step 3: Let R s = {r s |1≤s≤S} indicates that RP in step 1 j Available The corresponding AP number set. Then each reference point RP j The total clarity weight wd j It can be expressed as:

[0141]

[0142] According to the obtained clarity weight wd j , take the first B RPs with the highest weights as the support vector to be located L and the final RP set S L .

[0143] (2) Position calculation

[0144] The RP that finally participates in the target position calculation should satisfy the approximation of the vector to be located L on both the fingerprint clarity and the Euclidean distance scales. L The clarity weight of the internal RP wd j and the RSS vector Euclidean distance weight wo j ,wo j The inverse weighted calculation of the Euclidean distance from L to the RSS vector of each reference point is obtained; RP is further obtained j The combined weight wc j , the combined weight expression is as follows:

[0145]

[0146] Select the first E combined weights wc j The highest RP participates in the target position calculation.

[0147] The position X of the positioning vector L to be L The estimated expression is as follows:

[0148]

[0149] Embodiment 2

[0150] This embodiment provides an indoor positioning method in a multi-signal scenario for positioning in the corridor of a teaching building, and verifies the data RSS of a complete floor of the School of Internet of Things Engineering of Jiangx University. The indoor environment is a 60m×42m circular corridor. M=370 RPs are evenly set along the corridor, and 370×pt SRPs are extracted from the 370 RPs in proportion pt, and the remaining RPs are used as the verification set of IRPs.

[0151] In the offline phase, fingerprint data is collected 30 times at each RP, with a sampling interval of 2.3s. N = 6 APs of different brands and models are arranged at the four corners of the circular corridor and the middle of the north-south corridor to reflect the differences and fluctuations of AP devices. The undetected AP signal strength value at each RP is represented by -100dbm. In the online phase, the tester walks along the corridor with a handheld smart device to obtain a rectangular test trajectory, and records the positions of 187 test points on the trajectory and the corresponding RSS.

[0152] In the scenario, the principle of neighboring RPs with neighboring numbers is adopted to measure and count the changes in the signal strength and fingerprint clarity of all 370 RPs under a given AP1, such as Figure 3 shown. Figure 3 (a) The RSS of RP in the medium-smooth area fluctuates in the low-intensity range, with a small RSS span and a low path loss index, making it difficult to distinguish in the actual positioning process; the RSS of RP in the jump area fluctuates in the medium-high intensity range, with significant changes. It is easier to accurately locate the target by matching the RP in the jump area in the positioning area. Corresponding to RPs with high ambiguity and poor positioning accuracy, the fingerprint clarity of RP in the flat area in the fingerprint clarity map is also low, while the fingerprint clarity of RP in the jump area is generally high, reflecting that the RP in this interval has the characteristics of high RSS span and high positioning resolution. This proves the effectiveness of using fingerprint clarity to describe the regional signal fluctuation characteristics.

[0153] According to the fingerprint clarity clustering process, 370×0.167=62 RP data are uniformly extracted from the offline reference points at a sampling ratio of pt=0.167 to form the initial fingerprint library SRP. Based on 6 different APs, {Ω1,Ω2,…,Ω6} is generated for SRP. Due to the different performance of the AP itself and the different locations, the clustering partitioning of RP for each AP is quite different. The SRP partitioning of randomly selected AP2 and AP3 is shown in the figure below. Figure 4A and Figure 4B As shown in the figure. AP2 is an exposed Wi-Fi access point with an external mobile power bank, and AP3 is a daily home router. Objectively, the device stability of AP2 is lower than that of AP3. Figure 4AIn the figure, the SRPs receiving signals from AP2 are clustered into five regions, indicating that the signal fluctuations of AP2 are more diverse in this ring corridor environment. Figure 4B The number of partitions in AP3 is reduced by 2 compared with AP2. From the experimental environment itself, Figure 4A and Figure 4B The changes in the clustering area mostly occur at the corners of the corridor or where there are obvious obstructions. The existence of the fire door leads to Figure 4A and Figure 4B The sub-area boundaries are generated at the dotted boxes.

[0154] In order to verify the accuracy of the proposed fingerprint library expansion algorithm, the performance of the proposed interpolation algorithm in terms of interpolation accuracy was compared and tested between three interpolation algorithms: inverse distance weighted interpolation, Kriging interpolation, and traditional path loss model prediction. FIG. 5A to FIG. 5F The box plots of the fitting residuals of the four interpolation algorithms for the true value and the predicted value at each IRP at each AP are given. It can be seen that the inverse distance weighted algorithm has a large fitting error at each AP point, resulting in the largest error extreme value, because it only performs a simple linear processing on the distribution characteristics of the RSS signal. The traditional path loss model does not consider the change of the path loss index in the entire area, resulting in poor prediction effect in complex indoor environments. The Kriging interpolation algorithm cannot consider the different signal propagation characteristics of different APs in a specific indoor environment when fitting the variogram. The prediction accuracy at some APs is significantly lower than that of the proposed algorithm, and there are many outliers. The interpolation accuracy of the proposed interpolation algorithm is better than that of the other three interpolation algorithms in terms of error extreme value, median and number of error outliers. At the best AP2, the average fitting residual of the DIFS algorithm reached 2.79dbm, which was 18.3%, 33.2% and 64.2% higher than that of the Kriging, path loss model and IDW algorithms respectively. At other APs, the DIFS algorithm also has significant advantages.

[0155] In order to verify the overall performance of the method of this application, the performance of the method of this application is compared with three combination algorithms: Kriging interpolation algorithm and LWKNN positioning algorithm, path loss model interpolation algorithm and MCMC algorithm, and original fingerprint library and WKNN algorithm:

[0156] Kriging interpolation algorithm and LWKNN positioning algorithm can be found in Wang Xuan, Chen Guoliang, Cao Xiaoxiang, Yang Mengyi, and Wang Rui.

[0157] WLAN indoor positioning method based on adaptive K value and fingerprint library expansion[J]. Journal of Surveying and Mapping Science, 2020, 45(07): 26-32.

[0158] For the path loss model interpolation algorithm and MCMC algorithm, please refer to Wang Yue, Ba Bin, Cui Weijia, Lu Zhiyu. Markov Monte Carlo indoor positioning algorithm[J]. Journal of Xidian University, 2016, 43(02): 145-149.

[0159] To ensure the fairness of the experimental comparison, the other three combination algorithms all select the adaptive positioning algorithm of the current fingerprint library, and the four positioning algorithms all select four nearest neighbor reference points to participate in the position calculation.

[0160] Figure 6 The cumulative probability distribution of errors of four database expansion positioning algorithms is given. It can be seen that the number of test points with an estimated error within 3m for the method of this application reaches 62%, while the other algorithms are less than 60%. Table 1 gives the position estimation error values ​​of the four database expansion positioning algorithms. The average positioning error and maximum positioning error of the method of this application are smaller than those of the other three database expansion positioning algorithms. The positioning accuracy of the method of this application is improved by 16%, 14% and 19% respectively compared with the other three database expansion positioning algorithms. When performing position estimation, the method of this application has the highest positioning accuracy, which reflects that the DIFS algorithm can reduce labor costs in complex indoor environments while maintaining good positioning accuracy.

[0161] Table 1: Position estimation errors of four database expansion localization algorithms

[0162] m

[0163]

[0164] Some steps in the embodiments of the present invention may be implemented using software, and the corresponding software program may be stored in a readable storage medium, such as a CD or a hard disk.

[0165] 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 principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An indoor positioning method in a multi-signal scenario, characterized in that: The method comprises an offline phase and an online phase; In the offline stage, the fingerprint library is constructed and expanded, and the change degree of the path loss index is characterized by the fingerprint clarity. The adaptive division of the target area is realized by combining the density peak clustering. The expanded fingerprint library is called the offline interpolation fingerprint library. In the online stage, the comprehensive clarity weight wd j and received signal strength RSS vector Euclidean distance weight wo j , get the reference point RP j The first E reference points RP with the highest combined weights are selected to participate in the target position calculation to obtain an estimated value of the target position, where the target is the point to be located; The method comprises: Step 1: Build a system model; Arrange N wireless access points AP in the plane target space i , M reference points RP j ; Among them, the M reference points include R sampled reference points SRP r and Z reference points to be interpolated IRP z ; Among them, 1≤i≤N, 1≤j≤M, 1≤r≤R, 1≤z≤Z; Step 2: Fingerprint data preprocessing; Describing SRP based on signal aggregation and frequency r From AP i signal; Step 3: Expand the fingerprint database; Step 3.1: Cluster and partition the sampled reference point SPR; According to SRP r Its neighbor SRP r1 The signal difference and distance difference determine the SRP r From AP i Fingerprint clarity According to each SRP r Fingerprint clarity And plane position Jointly construct the SRP mapping three-dimensional space and calculate any SRP r and SRP r1 The space between r,r1 ; Calculate SRP according to formula (4) r The local density of the nearest neighbors at is: Among them, the cutoff distance f cut It is calculated by the ratio of the average expected number of neighboring SRPs of each SRP in the space to the total number of SRPs; According to the local density ρ of each SRP neighbor r Sort each SRP and let {q1,q2,…,q c ,q C } is the local density ρ of each SRP neighbor r The descending sequence number, where C = R, then SRP r The parent distance at is expressed as: Where ID(·) represents the variable number that satisfies the expression ·; At this point, each SRP r There is a pair of (ρ r ,θ r ); According to each SRP r The corresponding pair (ρ r ,θ r ) Clustering and partitioning all sampled reference point SPRs; Step 3.2: SRP r The signal characteristics at the location are used as the initial signal strength, considering SRP r With IRP z Signal propagation loss between the two nodes, establish IRP z Relative to SRP r The difference neighbor signal model is used; and the interpolation reference point IRP is treated according to the density peak clustering algorithm. z Partitioning; combining IRP z The SRP in the zone observed by the AP i The signal strength of the difference between the SRP and the neighbor signal model, for IRP z Signal strength Make predictions; Step 4: Select high-quality reference points based on fingerprint clarity to calculate the target position; SRP in step 3.1 r From AP i Fingerprint clarity The calculation formula is: in, Indicates SRP r The signal strength value of the reference point corresponding to the r1th number in the set A composed of the neighboring reference point numbers, Num(·) represents the number of elements contained in the set ·; Indicates SRP r The two-dimensional plane coordinates of Indicates SRP r1 The two-dimensional plane coordinates of .

2. The method according to claim 1, characterized in that The step 1 also includes: definition Indicates the reference point RP j The received signal strength RSS value received from N APs at the location; Indicates the sampled reference point SRP r The received signal strength RSS value received at the location; Indicates the reference point to be interpolated IRP z The received signal strength RSS value received at the location; definition represents the spatial position set of reference points RP, where Represents RP j The two-dimensional plane coordinates of Indicates SRP r The two-dimensional plane coordinates of represents the set of spatial locations of APs, where Indicates AP i The two-dimensional plane coordinates of definition Represents the RSS vector of the point to be located measured in the online stage, where: It indicates the received signal strength RSS value of the i-th wireless access point AP received by the point to be located measured in the online stage.

3. The method according to claim 2, characterized in that The step 2 comprises: Step 2.1: Within a given unit time T, the sampled reference point SRP r Collected from AP i The H signal in Indicates SRP r Collected from AP i The hth signal of Step 2.2: For the possible Q different RSS value samples, there is an SRP r About the Probability Distribution Set of RSS Its mean and median are μ r and δ r ; Step 2.3: Describe the SRP based on the signal concentration and frequency r From AP i The signal in, is the original measurement set The average value μ is selected from r H is the threshold t A high frequency signal that satisfies H t ≤H.

4. The method according to claim 3, characterized in that Any SRP in step 3.1 r and SRP r1 The space between r,r1 Expressed in Euclidean distance, the calculation formula is as follows: in, Indicates SRP r1 From AP i Fingerprint clarity.

5. The method according to claim 4, characterized in that The step 3.2 comprises: Calculate the sampled reference point SRP r From AP i Signal prediction value: in, Indicates the distance to the AP i The signal strength loss value at the default distance dis0, dis r Indicates SRP r With AP i spatial distance; Calculate the reference point IRP to be interpolated z Relative to the sampled reference point SRP r The difference neighbor signal of: Investigation distance IRP z The classification of the most recent K SRPs, IRP z Assigned to the partition with the highest SRP quantity ratio Inside; make For partition Included R k The number sequence of SRPs, combined with the data from AP observed by SRP in the partition i The signal strength of the difference between the SRP and the neighbor signal model, for IRP z Signal strength Make predictions: in, is the sampled reference point The corresponding path loss index is based on the same The signal strength of the known SRP in the partition is fitted with the distance between APs.

6. The method according to claim 5, characterized in that The step 4 comprises: Step 4.1: Obtain the RSS value of the received signal strength of all wireless access points AP received by the point to be located, form the RSS vector L of the point to be located, substitute the vector L of the point to be located into the position of each RP in the offline interpolation fingerprint library, and calculate the value of L at the RP. j From AP i Fingerprint clarity And with RP j Original fingerprint clarity Compare, set γ as the clarity difference threshold, if it satisfies Then RP j Join L's About AP i RP matching set Step 4.2: Follow RP j The average clarity of all SRPs in the partition The pre-selected reference points in the set are sorted in descending order; the pre-selected reference points are given corresponding weights according to the average clarity and clarity difference. in, RP j The average clarity of the partition, σ is the clarity difference magnification factor; Step 4.3: Let R s = {r s |1≤s≤S} represents the RP in step 4.1 j Available The corresponding AP number set; then each reference point RP j The total clarity weight wd j for: According to the obtained clarity weight wd j , take the first B RPs with the highest clarity weight as the support vector to be located L and the final RP set S L ; Step 4.4: Calculate S L The Euclidean distance weight of the inner RSS vector wo j ; Step 4.5: Synthesis S L The clarity weight of the internal RP wd j and the RSS vector Euclidean distance weight wo j , get RP j The combination weights of the first E RPs with the highest combination weights are selected to participate in the target position calculation: Among them, X j Represents the spatial position of the jth reference point RP.

7. The method according to claim 6, characterized in that The RP j The combined weight wc j The calculation formula is: Among them, 1≤b≤B.

Images