An indoor positioning method based on multi-chain interpolation to construct fingerprint map

Abstract

The invention discloses an indoor positioning method for constructing a fingerprint map based on multi-chain interpolation, and belongs to the technical field of wireless device positioning method research. The method comprises the following steps: sampling a sample target according to a sampling rule, and sampling to obtain a sampling point data set, wherein the sampling point data set comprises position coordinate information of the sampling point and an RSS value of the sampling point; inputting the sampling point data set into the system for constructing a fingerprint map based on multi-chain interpolation; the system for constructing a fingerprint map based on multi-chain interpolation performs internal processing to obtain a complete fingerprint map, wherein the complete fingerprint map comprises a sampling point data set and a data set of points to be inserted, wherein the data set of points to be inserted comprises position coordinate information of the points to be inserted and an RSS value of the points to be inserted; and obtaining the position coordinate information of the points to be detected through the complete fingerprint map and the RSS value of the points to be detected. Compared with a traditional indoor positioning method, the method of the invention improves positioning accuracy and stability.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless device positioning method research, and specifically relates to an indoor positioning method based on multi-chain interpolation to construct a fingerprint map. Background Art

[0002] Location-based services have gradually become an indispensable service in daily life. According to the implementation scenario, they are generally divided into outdoor positioning and indoor positioning. Traditional positioning methods include positioning methods using satellite positioning systems, and positioning methods using satellite positioning systems include the deployment of satellite positioning systems (such as Beidou, GPS, etc.). In outdoor positioning scenarios, the deployment of satellite positioning systems (such as Beidou, GPS, etc.) has enabled outdoor positioning and navigation to be well applied, and the positioning accuracy is very high, reaching the sub-meter level. In indoor positioning scenarios, due to the interference of various indoor obstacles on satellite signals, satellites cannot achieve accurate indoor positioning.

[0003] The widespread deployment of wireless devices in indoor environments has made indoor positioning services a research hotspot. For the study of indoor positioning methods, researchers usually use wireless signal fingerprints for matching positioning. The basic idea is to use the spatial differences of access points (APs) at different locations, and use the wireless signal characteristics of reference points as the characteristics of physical locations to determine the location. The wireless signal characteristics are called fingerprint characteristics, among which the received signal strength (RSS) is a typical fingerprint feature. Indoor positioning based on received signal strength is the key to providing accurate location services. This method first needs to build a fingerprint map (Site Survey) for the perception area, and needs to collect signal data for a large number of reference points in the perception area, which is extremely time-consuming and labor-intensive. To solve this problem, researchers collect the received signal strength (RSS) of some reference points and use them to interpolate the surrounding unsampled points to build a map of the entire area. This method can effectively reduce the time of map creation and update, but at the same time, according to the fingerprint positioning method, when there are fewer reference points, the positioning error will generally increase, which also reduces the positioning accuracy. In order to meet the positioning needs of large-scale scenarios, such as large commercial centers, parking lots, conference and administrative centers, etc., huge measurement costs are required. At the same time, in order to improve positioning accuracy, it is required to sample signal data from as many reference points as possible. At the same time, due to the inherent characteristics of wireless signals, the signals of reference points will change over time, resulting in inconsistencies in fingerprints at the same location during the sampling phase and the testing phase, making them impossible to match. For this reason, the fingerprint database needs to be maintained and updated at regular intervals, which further increases the positioning cost. Therefore, while ensuring positioning accuracy, it is crucial to reduce the workload of the fingerprint database establishment and update process.

[0004] For the fingerprint positioning method, creating or updating maps requires a lot of time and manpower. In actual sampling, when the positioning scene is large, it is obviously not practical to intensively sample the perception environment. To solve this problem, researchers use fingerprint map virtualization to build maps. Specifically, when creating a map, the perception environment is not fully sampled, but only some point signals are collected, and then these values ​​are used to simulate the signals of unsampled locations to expand the complete fingerprint map. This can greatly reduce the sampling volume and save sampling costs. The larger the area of ​​the positioning scene, the more resources the fingerprint map virtualization saves, and its importance is more reflected. This method reduces the workload of the map construction and update process, and guarantees the positioning accuracy to a certain extent. In the fingerprint virtualization construction, the fingerprint map construction for WiFi and radio frequency is a more representative method.

[0005] In the construction of WiFi fingerprint maps, the map construction method based on crowd intelligence perception is still the most common method to alleviate this problem. For example, more users are encouraged to participate in the crowdsourcing process in an incentive way, so that users can actively collect information; users are allowed to sample data unconsciously in an implicit or transparent way, and there is no need to incentivize users to achieve the purpose of explicitly collecting data; by sampling a small amount of identified fingerprint data (i.e., a data set containing RSS and location information) and a large amount of unidentified fingerprint data (i.e., a data set containing only RSS information but no location information), the workload can be reduced and the positioning accuracy can be compensated. On the other hand, many scholars build maps based on models, which mainly use a small amount of sampled data to fit the signals at other unsampled reference points according to the model, thereby filling the entire sampling area. For example, through the polynomial model, an algorithm for region segmentation and curve fitting is proposed, and the fitting function is obtained based on a small number of reference points in each sub-region to obtain the signal of the unsampled point; the weighted centroid of the sampled point fingerprint is used to estimate the location of the access point (AP), and the parameters of the signal transmission power and path loss model are evaluated using the Bayesian method. Gaussian process regression is used to fit a small amount of existing sampling data to generate the mean surface and standard deviation surface of the local area, and the particle filter is used to generate particles for weight assignment. In order to better update the fingerprint map, Gaussian process regression is combined with K-means.

[0006] In the construction of RF fingerprint maps, the LANDMARC algorithm, based on RFID positioning technology, adds landmarks to assist in updating the online update of the fingerprint database, and can cope with the impact of dynamic environments without obtaining maps in advance. Based on this, many scholars have proposed many improvement directions. Among them, the VIRE algorithm is a virtual tag positioning algorithm based on the LANDMARC algorithm, and the positioning accuracy is improved by 13% to 73% in different indoor environments. It introduces virtual reference tags. By adding virtual tags between fixed reference tags, virtual tags obtain the same effect as fixed reference tags, and excessive numbers will not cause signal interaction. Therefore, the use of spatial interpolation algorithms in the training stage to reduce the workload during sampling is gradually accepted by more and more researchers. In order to increase the density of fingerprints, a linear interpolation method is used to generate virtual sampling fingerprints and the accurate position is estimated through the Bayesian algorithm. Compared with the traditional Bayesian algorithm, the positioning accuracy is improved by 6% while reducing the time complexity of O(4n / 5). In addition, the multipath error is eliminated by introducing the variable coefficient into the traditional Leyte criterion, and then the natural neighbor interpolation is used for virtual reference marking to improve the stability and accuracy of the virtual label similarity measurement, ensuring that 80% of the test label errors are accumulated within 4cm and 90% within 10cm. In addition, the kriging interpolation algorithm and its several deformation algorithms are proposed and compared to affect the accuracy of map estimation. In addition, when constructing the fingerprint database, it is estimated Wall penetration loss and distance power loss, then insert RSS values ​​based on these loss models, apply the interpolation method of spatial interpolation to improve the fingerprint map, and improve the average positioning accuracy by about 25%; it is also proposed to replace the traditional static reference tags with dynamic particles in the traditional RFID virtual tag positioning, use the particle swarm optimization algorithm to update the Monte Carlo sample particle swarm, and assign different weights based on the signal strength difference between the sampled particles and the undetermined tags, so as to achieve the interpolation effect of unknown tags; and for the construction of virtual maps, it proposes a region-based repair method, mainly by limiting the path loss index used to construct virtual reference points and limiting the area to minimize the estimation error. These methods all involve the ideas of interpolation and virtual points to a certain extent, and optimize the workload of building fingerprint maps to a certain extent. However, the above methods have certain limitations and are difficult to be directly applied to WiFi fingerprint positioning.

[0007] Therefore, it is necessary to propose a new fingerprint map construction method to achieve indoor positioning. Summary of the invention

[0008] Purpose of the invention: In order to overcome the shortcomings of the prior art that cannot simultaneously improve accuracy and efficiency, the present invention provides an indoor positioning method based on multi-chain interpolation to construct a fingerprint map.

[0009] Technical solution: To achieve the above-mentioned purpose, the present invention provides an indoor positioning method for constructing a fingerprint map based on multi-chain interpolation, the method comprising the following steps: The method provides a system for constructing a fingerprint map based on multi-chain interpolation, the method comprising the following steps:

[0010] S1 samples the sample target according to the sampling rule to obtain a sampling point data set, which includes the location coordinate information of the sampling point and the RSS value of the sampling point;

[0011] S2 inputs the above-mentioned sampling point data set into the system for constructing fingerprint map based on multi-chain interpolation;

[0012] S3 is processed internally by a system that constructs a fingerprint map based on multi-chain interpolation to obtain a complete fingerprint map. The complete fingerprint map includes a sampling point data set and a data set of points to be inserted. The data set of points to be inserted includes the location coordinate information of the points to be inserted and the RSS value of the points to be inserted;

[0013] S4 obtains the location coordinate information of the point to be detected through the complete fingerprint map and the RSS value of the point to be detected

[0014] Furthermore, the system for constructing a fingerprint map based on multi-chain interpolation includes:

[0015] Input module: used to input the sampling point data set into the system for constructing fingerprint map based on multi-chain interpolation;

[0016] The RSS value estimation module of the point to be inserted is used to calculate the RSS value of the point to be inserted;

[0017] The position coordinate calculation module of the point to be inserted is used to calculate the position coordinates of the point to be inserted;

[0018] Data set merging module: taking the RSS value of the to-be-inserted point and the position coordinate information of the to-be-inserted point as the to-be-inserted point data set, and merging the to-be-inserted point data set and the sampling point data set as a complete fingerprint map;

[0019] Output module: used to output the complete fingerprint map.

[0020] Furthermore, the RSS value estimation module for the point to be inserted is used to calculate the RSS value of the point to be inserted, and the RSS value estimation module for the point to be inserted specifically includes:

[0021] Direction estimation module: determines the AP signal propagation path corresponding to the point to be inserted, performs interpolation calculation on the point to be inserted in the direction of each AP signal propagation path, and obtains the RSS value of the point to be inserted in each direction chain;

[0022] Strength estimation module: The weights on each direction chain are obtained by using the RSS values ​​of the above-mentioned points to be inserted on each direction chain and the inverse distance weighted method, and the estimated RSS values ​​of the points to be inserted are calculated by using the weights on each direction chain;

[0023] Weight optimization module: Use the data set of the neighboring sampling points of the point to be inserted on each directional chain to fit the signal attenuation model on each directional chain, calculate the fitted RSS value of the neighboring sampling points of the point to be inserted on each directional chain through the signal attenuation model, calculate the error of the signal attenuation model on each directional chain through the fitted RSS value of the neighboring sampling points, and recalculate the weight on each directional chain through the error of the signal attenuation model on each directional chain;

[0024] Adjustment valuation module: recalculate the RSS value of the point to be inserted through the RSS value of the point to be inserted on each direction chain and the weight of each direction chain;

[0025] Iterative expansion module: Repeat the above module steps to obtain the RSS values ​​of all points to be inserted.

[0026] Furthermore, the module for calculating the position coordinates of the point to be inserted calculates the position coordinates of the point to be inserted by pre-defining the perceived environment.

[0027] Furthermore, the sampling rule adopts a vertex sampling rule or a vertex center full sampling rule.

[0028] Furthermore, there are 4 AP signal propagation paths corresponding to the point to be inserted.

[0029] Furthermore, the point to be inserted has at least two adjacent sampling points on each direction chain.

[0030] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0031] 1. According to the propagation characteristics of wireless signals, the signal of the insertion point is formed by the superposition of signals from multiple directions. It is not possible to only consider the correlation of surrounding neighboring points, but should be carried out in different directions. The method of the present invention simultaneously considers the influence of signals from multiple directions, improves the positioning accuracy, evaluates the signal superposition of different links, constructs an accurate fingerprint map, and improves the positioning accuracy and stability.

[0032] 2. The method of the present invention designs a fingerprint map construction method of multi-chain interpolation (Multi-ChainInterpolation, MCI). Compared with the traditional sampling method, it uses multiple direction chains to estimate each point to be inserted according to the signal propagation characteristics in the actual environment. Specifically, interpolation calculations are performed on different directions under a given sampling rule, and the estimated value of the point to be inserted is obtained by inverse distance weighting. Using this value, the corresponding signal attenuation model is fitted in each direction and the error is calculated. The directional weight is re-obtained, and the estimated value of the point to be inserted is obtained. Finally, a fingerprint database consisting of real points and virtual points is formed, which effectively reduces the maintenance and update costs of the fingerprint database. According to experimental verification, the MCI method improves positioning accuracy while reducing the measurement workload.

[0033] 3. The method of the present invention uses two sampling modes ("Corner Sampling" and "Corner & Centre Sampling"), and only a small amount of fingerprint point signals need to be collected to build a complete fingerprint map of the perception area. The sampling amounts of the two modes are 25% and 50% of the full sampling, respectively, that is, the workload of the fingerprint map construction process is reduced by 75% and 50%, respectively, which greatly reduces the workload of the map construction and update process.

[0034] 4. The method of the present invention has been verified by large-scale experiments. The multi-chain interpolation map construction method MCI has improved positioning accuracy and good stability. The two sampling methods of the above MCI improve the positioning accuracy by 13.58% and 4.74% respectively compared with the full sampling method. Compared with the classical interpolation method, the MCI method has better stability, especially when the sampling amount is low. When the sampling is only 25% of the full sampling amount, the average accuracy is improved by 18.50% compared with the classical interpolation method. The method of the present invention has good scalability.

[0035] 5. The method of the present invention is aimed at constructing a map for WiFi fingerprint positioning, which reduces cost consumption, fully considers the signal propagation characteristics, evaluates the signal superposition of different links, constructs an accurate fingerprint map, and improves positioning accuracy and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of the steps of the method of the present invention.

[0037] Figure 2 It is a system structure diagram of the present invention for constructing a fingerprint map based on multi-chain interpolation.

[0038] Figure 3 It is a structural diagram of the RSS value estimation module of the point to be inserted in the present invention.

[0039] Figure 4It is a schematic diagram of the present invention when there are 4 AP signal propagation paths corresponding to the point to be inserted.

[0040] Figure 5 It is a schematic diagram of the vertex angle sampling rule adopted in the present invention.

[0041] Figure 6 It is a schematic diagram of the vertex center full sampling rule adopted by the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the embodiments and drawings.

[0043] Embodiment 1:

[0044] The indoor positioning method of this embodiment based on multi-chain interpolation to construct a fingerprint map is different from the traditional interpolation method. The traditional interpolation method is based on the signal propagation theory, that is, the closer the physical positions of two sampling points are, the more similar their intensity values ​​are. Therefore, the traditional interpolation method uses this principle to estimate the intensity of the interpolated points. Accordingly, the designed interpolation method uses the nearest neighbors as a reference. Obviously, this interpolation method only performs interpolation calculations from the perspective of numerical similarity, and does not consider the signal propagation situation in the actual environment.

[0045] According to actual sampling, the distance from the propagation source to the point to be inserted on the propagation path is close to the distance from the propagation source to the neighbor point. Therefore, the correlation between the point to be inserted and the neighbor point is based on the propagation direction of the signal, rather than the neighbors surrounding the point to be inserted. This embodiment uses this signal propagation characteristic to construct a fingerprint map based on multi-chain interpolation (MCI) when performing signal interpolation, based on multiple propagation links, rather than only considering the surrounding neighbors of the point to be inserted, thereby achieving indoor positioning.

[0046] The method steps of this embodiment refer to Figure 1 , specifically including the following steps:

[0047] Firstly, a system for constructing a fingerprint map based on multi-chain interpolation is provided. The system is used to construct a complete fingerprint map. Multi-chain means that when performing signal interpolation, it is based on multi-propagation links.

[0048] Then, the sample target is sampled according to the sampling rules to obtain a sampling point data set, which includes the location coordinate information of the sampling point and the RSS value of the sampling point. The sampling points are removed from the sample target to leave the points to be inserted. The obtained sampling point data set not only serves as a reference data set for solving the data set of the points to be inserted, but also constitutes a complete fingerprint database, namely a fingerprint map, together with the data set of the points to be inserted.

[0049] Then, the sampling point data set is input into the system for constructing fingerprint map based on multi-chain interpolation;

[0050] Then, the system that constructs the fingerprint map based on multi-chain interpolation performs internal processing to obtain the data set of points to be inserted, and then obtains the complete fingerprint map. The complete fingerprint map includes a sampling point data set and a data set of points to be inserted. The data set of points to be inserted includes the position coordinate information of the points to be inserted and the RSS values ​​of the points to be inserted, and the sampling point data set includes the position coordinate information of the sampling points and the RSS values ​​of the sampling points.

[0051] Finally, the complete fingerprint map obtained through the above steps and the RSS values ​​of the points to be detected that need to be tested in the test phase are used to obtain the position coordinate information of all the points to be detected in the test phase. Specifically, the complete fingerprint map obtained through the above steps contains the position coordinate information of all points and the RSS value of the point. According to the RSS value of the point to be detected, the position coordinate information of the point to be detected can be obtained by comparison.

[0052] Embodiment 2:

[0053] The indoor positioning method of this embodiment based on multi-chain interpolation to construct a fingerprint map is based on Example 1, referring to Figure 2 ,The system for constructing fingerprint maps based on multi-chain interpolation includes:

[0054] Input module: used to input the sampling point data set into the system for constructing fingerprint map based on multi-chain interpolation;

[0055] The RSS value estimation module of the point to be inserted is used to calculate the RSS value of the point to be inserted;

[0056] The position coordinate calculation module of the point to be inserted is used to calculate the position coordinates of the point to be inserted;

[0057] Data set merging module: taking the RSS value of the to-be-inserted point and the position coordinate information of the to-be-inserted point as the to-be-inserted point data set, and merging the to-be-inserted point data set and the sampling point data set as a complete fingerprint map;

[0058] Output module: used to output the complete fingerprint map;

[0059] The above-mentioned RSS value estimation module for the point to be inserted is used to calculate the RSS value of the point to be inserted, referring to Figure 3 The RSS value estimation module of the point to be inserted specifically includes:

[0060] Direction estimation module: determine the AP signal propagation path corresponding to the point to be inserted. Usually, there are at least two APs corresponding to the point to be inserted, that is, there are at least two AP signal propagation paths corresponding to the point to be inserted. Perform multi-chain interpolation calculation on the point to be inserted in the direction of each AP signal propagation path. In this embodiment, multi-chain interpolation calculation means performing interpolation calculation based on multiple AP signal propagation paths corresponding to the point to be inserted, and obtaining the RSS value of the point to be inserted in each direction chain. Here, one direction chain refers to one AP signal propagation path.

[0061] Strength estimation module: The weights on each direction chain are obtained by using the RSS values ​​of the above-mentioned points to be inserted on each direction chain and the inverse distance weighted method, and the estimated RSS values ​​of the points to be inserted are calculated by using the weights on each direction chain;

[0062] Weight optimization module: Use the data set of the sampling points adjacent to the point to be inserted on each directional chain to fit the signal attenuation model on each directional chain, calculate the fitted RSS value of the sampling points adjacent to the point to be inserted on each directional chain through the signal attenuation model, calculate the error of the signal attenuation model on each directional chain through the fitted RSS value of the adjacent sampling points, and recalculate the weight on each directional chain through the error of the signal attenuation model on each directional chain;

[0063] Adjust the valuation module: recalculate the RSS value of the point to be inserted through the RSS value of the point to be inserted on each direction chain and the weight of each direction chain; the method of this embodiment takes into account that the point to be inserted will receive wireless signals in different directions, that is, the AP signal propagation paths corresponding to the point to be inserted include multiple, and the signal strength of the point to be inserted is affected by the superposition of multiple signals. Since the signal propagation in each direction is not the same, in order to more reasonably estimate the signal strength of the point, that is, the RSS value, it is necessary to consider the signal attenuation in each direction.

[0064] Iterative expansion module: Repeat the above module steps to obtain the RSS values ​​of all points to be inserted.

[0065] The RSS value of the point to be inserted is calculated by the above-mentioned RSS value estimation module, and then the data set of the point to be inserted is obtained.

[0066] Embodiment 3:

[0067] In this embodiment, an indoor positioning method for constructing a fingerprint map based on multi-chain interpolation is used to calculate the RSS value of the point to be inserted through an RSS value estimation module for the point to be inserted, based on Embodiment 2. The RSS value estimation module for the point to be inserted includes a direction estimation module, an intensity estimation module, a weight optimization module, an adjustment estimation module and an iterative expansion module.

[0068] The RSS value of the point to be inserted in each direction chain is solved by the direction estimation module, which specifically includes the following steps:

[0069] Suppose the point to be inserted is P, then the position coordinates of the point to be inserted P are P(x,y), determine the direction link of the point to be inserted P(x,y), that is, the above direction link, suppose the point to be inserted P(x,y) has n direction links, and the n direction links are marked as (r1,r2,…,r n ), suppose the point to be inserted P(x,y) is at r k The number of neighbor links on the directional link is l(k)(1≤k≤n), and the point to be inserted P(x,y) is at r k The set of neighbor points on the directional link is marked as is the point to be inserted P(x,y) in r k The nearest neighbor point on the directional link, Is the insertion point In r k The nearest neighbor point on the directional link, yes In r k The nearest neighbor point on the directional link, and so on;

[0070] The point to be inserted P(x,y) is at r k RSS value on the direction link Calculated by formula (1):

[0071]

[0072] In the above formula (1), Yes The RSS value of Yes The RSS value of is the point to be inserted P(x,y) and its position in r k The nearest neighbor point on the directional link The distance between Yes and Point The distance between points RSS value point RSS value can be obtained through the sampling step. and All can be calculated through the position coordinates of the point;

[0073] The RSS value of the point to be inserted in each direction chain is calculated through the above steps.

[0074] The estimated RSS value of the point to be inserted by the strength estimation module is first calculated through the above steps to obtain the weights of the chains in each direction, specifically:

[0075] Assume that the point to be inserted P(x,y) is at r k The weight of the directional link is w(r k ), then w(r k ) is calculated according to formula (2):

[0076]

[0077] In the above formula (2), Yes The RSS value of Yes The RSS value of Yes and Point The distance between points RSS value point RSS value can be obtained through the sampling step. All can be calculated through position coordinates;

[0078] Through the above steps, we can get the point P(x,y) to be inserted in r k The weight of the directional link is w(r k ).

[0079] According to formula (3), the RSS estimated value RSS of the point to be inserted P(x, y) is calculated P ;

[0080]

[0081] In the above formula (3), Calculated by formula (1), w(r k ) is calculated by formula (2), where n is a natural number, 1≤k≤n;

[0082] Through the above steps, we can get the RSS estimate RSS of the point to be inserted P(x,y) P .

[0083] The weights on each direction chain are recalculated through the weight optimization module, as follows:

[0084] First, the signal attenuation model on each direction chain is fitted using the data set of the neighboring sampling points of the point to be inserted on each direction chain. Suppose the point to be inserted P(x, y) is at r k Its adjacent sampling points on the direction chain are marked as The RSS values ​​of neighboring sampling points are marked as m is the number of neighboring sampling points. The position coordinate information of the neighboring sampling points is obtained through the sampling step. The parameters of the signal attenuation model formula (4) are obtained by fitting the signal attenuation model formula (4) with the neighboring sampling point data set. And the value of parameter σ:

[0085]

[0086] Place the point to be inserted P(x,y) at r k The neighboring sampling point data set on the direction chain is introduced into the above formula (4). The neighboring sampling point data set includes the RSS value and position coordinate information of the neighboring sampling points, and the result is

[0087] In the above formula, Yes The distance to the point to be inserted P(x,y), Yes The distance to the point to be inserted P(x,y), Yes The distance to the point to be inserted P(x,y), and so on. The parameters are calculated by the above formula and the value of parameter σ, thus fitting to obtain r k Signal attenuation model formula on the directional chain

[0088] Then, the signal attenuation model formula obtained by fitting and the RSS value of the sampling point adjacent to the point to be inserted on each direction chain are used. To calculate the fitted RSS value of the sampling points adjacent to the point to be inserted on each direction chain, the fitted RSS value of the sampling points adjacent to the point to be inserted is marked as The fitted RSS values ​​of the sampling points adjacent to the point to be inserted on each direction chain are calculated by the following formula:

[0089] In the above formula, the RSS value of the adjacent sampling point is Through the sampling step, the parameters The value of and parameter σ is calculated by formula (4), Yes The distance to the point to be inserted P(x,y), Yes The distance to the point to be inserted P(x,y), Yes The distance to the point to be inserted P(x,y), and so on.

[0090] Then, according to formula (5), calculate kError in the signal attenuation model on the directional link:

[0091]

[0092] In the above formula (5), α is based on the point The distance to the point to be inserted is defined, and β is based on the point The distance to the point to be inserted is defined, and γ is based on the point The distance to the point to be inserted is defined as m, which is the distance between the point to be inserted P(x,y) and r. k The number of adjacent sampling points on the direction chain, m is not less than 2;

[0093] Finally, according to formula (6), the insertion point P(x, y) is recalculated in r k The optimized weight w′(r k ):

[0094]

[0095] In the above formula (6), D(r k ) is calculated by formula (5), where n is a natural number, 1≤k≤n;

[0096] Recalculate the optimized RSS value RSS′P of the point to be inserted P(x, y) according to formula (7):

[0097]

[0098] In the above formula (7), Calculated by formula (1), w′(r k ) is calculated by formula (6), where n is a natural number, 1≤k≤n;

[0099] The optimized RSS value RSS′ of the point to be inserted P(x, y) is obtained P As the final RSS value of the point to be inserted P(x, y), the final RSS values ​​of all the points to be inserted in the sample target are obtained through all the above steps.

[0100] Embodiment 4:

[0101] In this embodiment, an indoor positioning method based on multi-chain interpolation to construct a fingerprint map is based on Embodiment 3. In the direction estimation module, refer to Figure 4 , Figure 4It shows that there are 4 AP signal propagation paths corresponding to the point to be inserted, that is, the point to be inserted receives signals from 4 AP directions, and the signal of the point to be inserted is the superposition of the signals from the 4 AP directions. The 4 APs are AP1, AP2, AP3 and AP4, and the signal of the point to be inserted is the superposition of the signals from AP1, AP2, AP3 and AP4. Then the point to be inserted P(x, y) includes the RSS values ​​on the 4 directional links, and the point to be inserted P(x, y) includes the weights on the 4 directional links. The RSS estimated value RSS of the point to be inserted P(x, y) is calculated by the RSS values ​​on the 4 directional links and the weights on the 4 directional links. P , and further obtain the optimized RSS value RSS′ of the insertion point P(x,y) P .

[0102] Embodiment 5:

[0103] In this embodiment, an indoor positioning method for constructing a fingerprint map based on multi-chain interpolation is provided. Based on Embodiment 3, in a weight optimization module, a data set of adjacent sampling points of a point to be inserted in each direction chain is used to fit a signal attenuation model on each direction chain. The fitted RSS value of the adjacent sampling points of the point to be inserted in each direction chain is calculated by the signal attenuation model. The error of the signal attenuation model on each direction chain is calculated by the fitted RSS value of the adjacent sampling points. The weights on each direction chain are recalculated by the error of the signal attenuation model on each direction chain. In this embodiment, the signal attenuation model on each direction chain is firstly fitted by the data set of adjacent sampling points of the point to be inserted in each direction chain. Assume that the point to be inserted P(x, y) is in r k Its adjacent sampling points on the direction chain are marked as The fitting result is r k Signal attenuation model formula on the directional chain Then calculate the fitted RSS values ​​of the sampling points adjacent to the point to be inserted on each direction chain, and the fitted RSS values ​​of the sampling points adjacent to the point to be inserted are marked as Then, according to formula (5), calculate k Error in the signal attenuation model on the directional link:

[0104]

[0105] In the above formula (5), α is based on the point The distance to the point to be inserted is defined, and β is based on the point The distance to the point to be inserted is defined, and γ is based on the point The distance to the point to be inserted is defined as m, which is the distance between the point to be inserted P(x,y) and r. k The number of adjacent sampling points on the direction chain, m is equal to 2, then formula (5) is transformed into:

[0106]

[0107] In the above formula, α is based on the point The distance to the point to be inserted is defined, and β is based on the point Definition of the distance to the point to be inserted.

[0108] Through the above transformation formula, we can get D(r k ), and finally recalculate the insertion point P(x, y) in r according to formula (6) k The optimized weight w′(r k ):

[0109]

[0110] Finally, the optimized RSS value RSS′ of the point to be inserted P(x, y) is recalculated according to formula (7): P :

[0111]

[0112] The optimized RSS value RSS′ of the point to be inserted P(x, y) is obtained P As the final RSS value of the point to be inserted P(x, y), the final RSS values ​​of all the points to be inserted in the sample target are obtained through all the above steps.

[0113] Example 6

[0114] The present embodiment is an indoor positioning method for constructing a fingerprint map based on multi-chain interpolation. Based on Example 3, the position coordinate calculation module of the point to be inserted calculates the position coordinates of the point to be inserted by pre-defining the perceived environment, that is, when the sample target is sampled according to the sampling rule, the sampling obtains a sampling point data set, and the position coordinates of the point to be inserted are calculated by pre-defining the perceived environment, thereby obtaining a data set of the point to be inserted.

[0115] Example 7

[0116] In this embodiment, an indoor positioning method for constructing a fingerprint map based on multi-chain interpolation is used. Based on Embodiment 6, in sampling the sample target according to the sampling rule, the sampling rule adopts the vertex sampling rule. When the vertex sampling rule is adopted, the corresponding sampling map is divided into a number of grids, and fingerprints are collected evenly at intervals on the grids at a certain distance. Figure 5 The specific rule is to divide the map into multiple "3×3" squares, sample the four corners (four vertices) of each square, and finally fill the entire sampling space. This sampling mode is called "corner sampling mode (Corner Sampling)", as shown in the figure, where the four vertices in each "3×3" square are real sampling points, and the remaining 5 points in the "3×3" square are to be inserted. To ensure interval sampling, each "3×3" square slides 2 sampling points, Figure 5 The 10m×10m map is sampled using this mode, and the actual sampling points only account for 25 / 100=25% of all the points.

[0117] Example 8

[0118] In this embodiment, an indoor positioning method for constructing a fingerprint map based on multi-chain interpolation is based on Example 6. In sampling the sample target according to the sampling rule, the sampling rule adopts the vertex center full sampling rule. It is known that the more sampling points there are, the better the effect of the constructed fingerprint map is. At the same time, the fingerprints of neighboring points are more similar. Based on the above actual situation, on the basis of the "corner sampling mode", in order to further improve the positioning accuracy and stability, the sampling points are increased by the center points of the grid. Specifically, when the vertex center full sampling rule is adopted, in addition to collecting the four vertex corners of the "3×3" grid, its center point is also included. The sampling form is as follows: Figure 6 This sampling rule is called "Corner & Centre Sampling". As shown in the figure, the four vertices and one center point in each "3×3" square are real sampling points, and the remaining four points in the "3×3" square are points to be inserted. Similar to the "Corner Sampling" mode, the square slides 2 sampling points each time. In this sampling mode, Figure 6 In the 10m×10m map, the actual sampling points account for ((25+25)) / 100=50% of all the points.

[0119] Example 9

[0120] This embodiment is an indoor positioning method based on multi-chain interpolation to construct a fingerprint map. Based on Example 7, this embodiment experimentally verifies the performance of the MCI method using the "Corner Sampling mode". Experiments are conducted in large-scale scenarios to verify the performance of the MCI method. The experimental scenario is an indoor environment of 3,000 square meters. 50 iBeacons of equal height are placed at a height of 2 meters from the ground as APs. In order to verify the stability of MCI performance in different devices, this embodiment uses 6 smartphones for sampling, namely HUAWEI mate7, HUAWEI mate20, HUAWEI honor, VIVOx20, VIVO x23 and YAAO. The sampling point interval is 1 meter, and each sampling point collects 1 minute of signal. The entire sampling process takes about 250 hours. Table 1 is obtained through experimental comparison:

[0121] Table 1

[0122]

[0123] Table 1 shows the comparison results of the MCI method with other methods when the “Corner Sampling” mode is used for sampling. It can be seen from Table 1 that the average positioning accuracy of the MCI method is 4.74% higher than that of the full sampling method (i.e., sampling all points in the sample target).

[0124] Example 10

[0125] This embodiment is an indoor positioning method for constructing a fingerprint map based on multi-chain interpolation. Based on Example 8, this embodiment experimentally verifies the performance of the MCI method using the "Corner & Centre Sampling" mode, and conducts experiments in large-scale scenarios to verify the performance of the MCI method. The experimental scene is an indoor environment of 3,000 square meters. 50 iBeacons of equal height are placed at a height of 2 meters from the ground as APs. In order to verify the stability of MCI performance in different devices, this embodiment uses 6 smartphones for sampling, namely HUAWEI mate7, HUAWEI mate20, HUAWEIhonor, VIVO x20, VIVO x23 and YAAO. The sampling point interval is 1 meter, and each sampling point collects 1 minute of signal. The entire sampling process takes about 250 hours. Table 2 is obtained through experimental comparison:

[0126] Table 2

[0127]

[0128] Table 2 shows the comparison results of the MCI method with other methods when the “Corner & Centre Sampling” is used for sampling. The average positioning accuracy of the MCI method is 13.58% higher than that of the full sampling method (i.e., sampling all points in the sample target).

[0129] According to experimental verification, in two sampling modes, when the sampling points are reduced by 50% and 75% compared with the full sampling method, the positioning accuracy of MCI is improved by 13.58% and 4.74% respectively. Compared with the classical interpolation method, the MCI method has better stability, and the advantage of positioning accuracy is more obvious when the sampling amount is small. When the sampling is only 25% of the full sampling amount, the average accuracy is improved by 18.50% compared with the classical interpolation method. Therefore, the MCI method can better ensure the positioning accuracy while greatly reducing the workload of the map construction process.

[0130] The above is only a preferred embodiment of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. An indoor positioning method for constructing a fingerprint map based on multi-chain interpolation, characterized in that: The method provides a system for constructing a fingerprint map based on multi-chain interpolation, and the method comprises the following steps: S1 samples the sample target according to the sampling rule to obtain a sampling point data set, which includes the location coordinate information of the sampling point and the RSS value of the sampling point; S2 inputs the above-mentioned sampling point data set into the system for constructing fingerprint map based on multi-chain interpolation; S3 is processed internally by a system that constructs a fingerprint map based on multi-chain interpolation to obtain a complete fingerprint map. The complete fingerprint map includes a sampling point data set and a data set of points to be inserted. The data set of points to be inserted includes the location coordinate information of the points to be inserted and the RSS value of the points to be inserted; S4 obtains the position coordinate information of the point to be detected through the complete fingerprint map and the RSS value of the point to be detected; the system for constructing a fingerprint map based on multi-chain interpolation includes: Input module: used to input the sampling point data set into the system for constructing fingerprint map based on multi-chain interpolation; The RSS value estimation module of the point to be inserted is used to calculate the RSS value of the point to be inserted; The position coordinate calculation module of the point to be inserted is used to calculate the position coordinates of the point to be inserted; Data set merging module: taking the RSS value of the to-be-inserted point and the position coordinate information of the to-be-inserted point as the to-be-inserted point data set, and merging the to-be-inserted point data set and the sampling point data set as a complete fingerprint map; Output module: used to output the complete fingerprint map; The RSS value estimation module for the point to be inserted is used to calculate the RSS value of the point to be inserted, and the RSS value estimation module for the point to be inserted specifically includes: Direction estimation module: determines the AP signal propagation path corresponding to the point to be inserted, performs interpolation calculation on the point to be inserted in the direction of each AP signal propagation path, and obtains the RSS value of the point to be inserted in each direction chain; Strength estimation module: The weights on each direction chain are obtained by using the RSS values ​​of the above-mentioned points to be inserted on each direction chain and the inverse distance weighted method, and the estimated RSS values ​​of the points to be inserted are calculated by using the weights on each direction chain; Weight optimization module: Use the data set of the neighboring sampling points of the point to be inserted on each directional chain to fit the signal attenuation model on each directional chain, calculate the fitted RSS value of the neighboring sampling points of the point to be inserted on each directional chain through the signal attenuation model, calculate the error of the signal attenuation model on each directional chain through the fitted RSS value of the neighboring sampling points, and recalculate the weight on each directional chain through the error of the signal attenuation model on each directional chain; Adjustment valuation module: recalculate the RSS value of the point to be inserted through the RSS value of the point to be inserted on each direction chain and the weight of each direction chain; Iterative expansion module: Repeat the above module steps to obtain the RSS values ​​of all points to be inserted.

2. The indoor positioning method based on multi-chain interpolation to construct fingerprint map according to claim 1 is characterized in that: The position coordinate calculation module of the point to be inserted calculates the position coordinates of the point to be inserted by pre-defining the perceived environment.

3. The indoor positioning method based on multi-chain interpolation to construct fingerprint map according to claim 2 is characterized in that: The sampling rule adopts the vertex sampling rule or the vertex center full sampling rule.

4. The indoor positioning method based on multi-chain interpolation to construct fingerprint map according to claim 3 is characterized in that: There are 4 AP signal propagation paths corresponding to the point to be inserted.

5. The indoor positioning method based on multi-chain interpolation to construct fingerprint map according to claim 4 is characterized in that: The point to be inserted has at least two adjacent sampling points on each direction chain.

Images