A multipath-assisted Bluetooth AOA indoor positioning method based on crowdsourcing mechanism

Abstract

The present invention relates to a crowdsourcing-based multipath-assisted Bluetooth AOA indoor positioning method. Bluetooth base stations are deployed in an indoor environment, and the position of a user terminal changes as the user's movement path changes. The user terminal sends a signal to the Bluetooth base station. After receiving the signal transmitted by the user terminal, the Bluetooth base station uploads data information to a cloud server. The cloud server processes the message uploaded by the Bluetooth base station, performs Bluetooth AOA estimation for different user terminals and their corresponding Bluetooth base stations, extracts measurement values, and uses multipath information and geometric structures to resolve virtual anchor point status information generated by multipath components. The anchor point status is used as a feature in a wireless map, and an indoor wireless map with the anchor point status as a feature is constructed. Finally, the cloud server directly returns the user position calculated in real time to the user terminal, assisting the user terminal in performing indoor positioning in complex indoor environments.

Description

Technical Field

[0001] The present invention relates to the field of indoor positioning technology, and specifically to a multipath-assisted Bluetooth AOA indoor positioning method based on a crowdsourcing mechanism. The method can be applied to indoor positioning application scenarios with significant cost constraints and high positioning accuracy requirements, such as indoor location services, public safety, big data analysis, etc. Background Art

[0002] With the rapid development of mobile smart devices and wireless networks, all aspects of human production and life have been affected. In this era, location-based services (LBS) have become an indispensable part of people's lives.

[0003] Currently, thanks to the maturity of outdoor positioning technologies such as GPS and Beidou navigation, people's demand for outdoor positioning services has largely been met. However, for indoor positioning, a comprehensive solution still lacks. According to a report by the US Environmental Protection Agency, people spend approximately 70% to 90% of their time indoors. Combining indoor people or objects with spatial data can digitize their location information, making them searchable and locating. This has many application scenarios, such as indoor location services, public safety, and big data analysis.

[0004] However, due to the presence of complex conditions such as non-line-of-sight (NLOS) areas and multipath effects in indoor environments, signal propagation conditions become harsh. Therefore, positioning methods that successfully achieve high positioning accuracy in outdoor scenarios cannot provide sufficient positioning accuracy in indoor environments. Current high-precision indoor positioning technologies require expensive auxiliary equipment or a large amount of manual processing in the early stages. How to improve the accuracy of low-cost indoor positioning technology and how to reduce the cost of high-precision indoor positioning technology are prerequisites for the popularization of indoor positioning technology. The huge research significance and practical value of indoor positioning have attracted a large number of researchers, and a large number of methods capable of indoor positioning have emerged, becoming a hot topic of discussion and research.

[0005] Indoor positioning solutions are diverse and can be categorized by the positioning technology used: WiFi, ZigBee, RFID, vision, infrared, ultrasonic, ultra-wideband, inertial navigation, geomagnetism, Bluetooth, cellular networks, pseudolites, and others. Common indoor positioning methods can be further categorized as proximity detection, centroid positioning, multilateral positioning, triangulation, polar positioning, scene analysis, fingerprint positioning, and dead reckoning. Combining different positioning technologies and methods can yield a variety of indoor positioning use cases. Current indoor positioning requires low power consumption, low cost, and high accuracy. Comparisons of positioning technologies reveal that Bluetooth AOA has a significant market share in indoor positioning, but its positioning accuracy is susceptible to multipath fading, making it incapable of achieving high-precision positioning. Therefore, selecting appropriate positioning methods to assist Bluetooth AOA in achieving high-precision positioning in the presence of multipath indoors is of practical significance. Summary of the Invention

[0006] In view of the problems existing in the prior art, the object of the present invention is to provide a multipath-assisted Bluetooth AOA indoor positioning method based on a crowdsourcing mechanism, which has the advantages of low cost, low power consumption and high positioning accuracy.

[0007] To achieve the above object, the technical solution adopted by the present invention is:

[0008] A crowdsourcing-based multipath-assisted Bluetooth AOA indoor positioning method is implemented based on an indoor positioning system. The indoor positioning system includes a user terminal, a Bluetooth base station, and a cloud server. The user terminal sends a signal to the Bluetooth base station at regular intervals. The Bluetooth base station packages the received signals according to the sequence numbers of different user terminals and sends them to the cloud server.

[0009] The cloud server uses the Bluetooth base station as a physical anchor point and sets a virtual anchor point. The position of the virtual anchor point is a mirror image of the physical anchor point and is caused by reflection from a flat surface. For each reflection path, in the case of a two-dimensional plane, the distance from the user terminal to the physical anchor point is equal to the distance from the user terminal to the virtual anchor point.

[0010] The cloud server also stores a cloud wireless map, which is constructed by physical anchor points and virtual anchor points of user terminals that have completed positioning;

[0011] When the cloud server receives data sent by the Bluetooth base station, it estimates the specular multipath component parameters from the received data and processes the specular multipath component to obtain AOA information; then, the AOA information is used as the measurement value, and the physical anchor point and virtual anchor point position information are obtained from the cloud wireless map as the feature state, and there is a data association between the feature state and the measurement value; the cloud server calculates the user terminal position information corresponding to the measurement value and the corresponding physical anchor point and virtual anchor point position information based on the feature state and the measurement value using the BP algorithm; and returns the user terminal position information to the user terminal, and at the same time, updates the cloud wireless map using the calculated physical anchor point and virtual anchor point.

[0012] The process of calculating the user terminal location and the physical anchor point and virtual anchor point location information based on the BP algorithm is as follows:

[0013] (1) State transition:

[0014] The user terminal state x at time n-1 n-1 , measuring angle deviation α n-1 With the characteristic state y n-1 , the user terminal state x at time n is obtained by independent evolution according to Markov state dynamics n , measuring angle deviation α n and traditional characteristic states It can be expressed as a probability density function:

[0015]

[0016] Among them, f(x n |x n-1 ) is represented by the following motion model:

[0017]

[0018] In formula (15), x n-1 is the user terminal state at time n-1, α n-1 is the measured angle deviation at time n-1, y n-1 is the characteristic state at time n-1, x n is the user terminal state at time n, α n To measure the angle deviation at any moment, is the traditional characteristic state at time n, J is the number of PAs, is the number of traditional features, is the dth traditional feature vector generated by the jth PA at time n, that is, the traditional feature state generated by the PF with index (j, d), is the dth eigenvector generated by the jth PA at time n, that is, the characteristic state generated by the PF with index (j, d), represents the measurement error, where α n Represents the angular deviation of the user terminal device, is the path loss index, which changes according to the changes in the environment;

[0019] In formula (16), A represents the state transfer matrix, ΔT is the sampling period, and d n is a motion process with a mean of zero and a covariance of Gaussian distribution;

[0020] If a feature does not exist at time n-1, it cannot exist as a traditional feature at time n. You can define:

[0021]

[0022] Among them, f D (·) is the probability density function of any setting;

[0023] If a feature exists at time n-1, the probability that the feature exists at time n is determined by its production probability; when , you can define:

[0024]

[0025] Among them, P s (·) represents the survival probability of the feature;

[0026] (2) measurement and evaluation;

[0027] When the measured value is known, the user terminal state x n , eigenvector state Measurement angle error α n , data association vector Measurement quantity The probability measurement distribution of , its likelihood function is updated as:

[0028]

[0029] In formula (19), is the observation value generated by the jth PA at time n, x n is the user terminal state at time n, α n is the angle deviation measured at time n, is the characteristic state generated by the j-th PA at time n, is the feature-oriented data association vector of the j-th PA at time n, is the total number of measurements generated by the j-th PA at time n, is the data association vector represented by the dth feature vector generated by the jth PA at time n, f false (·) is the probability density distribution function of false alarms, is the traditional feature set that generates measurements at time n, Generate a new feature set of measurements at time n, is the total number of features, since is a constant, Under known conditions, that is, when the total number of measurements is known, measurement evaluation is defined as:

[0030]

[0031] (3) Data association; data association vector The joint prior probability density function, the number of measurements and the status of the new feature Depends on traditional feature status User terminal status x n and the unknown measured angle deviation α n , expressed as:

[0032]

[0033] in, is the data vector constraint, and Represent the average number of new features and false alarms, is the set of traditional features that have no measurement value generated at time n, is the set of new features that have no measurement values ​​generated at time n, P d (·)∈(0,1] represents the probability of the feature being detected, that is, whether a measurement value is generated; f new (·) represents the probability density function of the new detection feature;

[0034] (4) Data fusion: The joint posterior distribution of the user terminal state, measurement deviation, feature state, and data association vector depends on the measurement values ​​in all discrete time periods: f(x 1:N ,α 1:N ,y 1:N ,a 1:N ,b 1:N ∣z 1:N ), where z 1:N Represents the measured value from time 1 to N, x 1:N represents the user terminal status from time 1 to N, α 1:N Indicates the measurement angle error from time 1 to time N, y1:N Represents the characteristic state from time 1 to N, a 1:N 、b 1:N Represents the data association vector from time 1 to N;

[0035] set up is a set of traditional features that generate measurements at time n, so for have and is a set of traditional features that do not generate measurement values ​​at time n; therefore, for have and And when Sometimes, there are After finishing, we can get:

[0036]

[0037] In addition, It is defined as a set of new features that generate measurements at time n. have and is a set of traditional features that do not generate measurement values ​​at time n; therefore, for have and And when Sometimes, there are After finishing, we can get:

[0038]

[0039] Finally, the joint posterior probability density function is sorted out to obtain:

[0040]

[0041] According to formula (24), the location information of the user terminal at time n and the corresponding physical anchor point and virtual anchor point location information can be obtained.

[0042] The cloud server updates the cloud wireless map as follows:

[0043] Define a collection It contains the estimated feature state of user terminal k as the wireless feature map of user terminal k, as shown below:

[0044]

[0045] and

[0046]

[0047] Where n' represents the convergence time of the SLAM algorithm at user terminal k, Representation characteristics The existence probability of , the subscript k is used to distinguish different user terminals. In the cloud, additional definition of the set To share wireless maps, collect is a collection A weighted combination of, where K represents the total number of user terminals, the set It is a dynamic collection that needs to continuously receive data information from the Bluetooth base station for update;

[0048] Assuming K1 is the total number of user terminals at present, we can define:

[0049]

[0050]

[0051] The cloud server analyzes the features estimated by the user terminal k and compares them with the set Combine and generate a complete shared wireless map in the following ways

[0052]

[0053] Introducing the weight coefficient w k,n' , which indicates the reliability of the feature in the prior information obtained by the k-th user terminal at time n'. Therefore, the estimated feature state generated by the k-th user terminal is expressed as:

[0054]

[0055] in Representation characteristics reliability,

[0056] will be collected The weight in is set to be proportional to the upload time, that is, when n1<n2,

[0057] Let P sn,l For the collection The reliability of the lth feature in threshold is the threshold, If P sn,l <P threshold , then the lth feature will be from the set Deleted.

[0058] This invention leverages multipath information and geometric structures to transform multipath disadvantages in complex environments into multipath advantages. By using the state information of virtual anchor points corresponding to physical anchor points as features to construct a wireless map, Bluetooth AOA achieves high positioning accuracy even in complex indoor environments. Furthermore, to address the slow construction of wireless maps in large indoor spaces, a crowdsourcing approach is employed to enable decentralized collaboration among multiple users to jointly collect and update information for wireless map construction. This approach also provides a solution to the repeated calculations and data reliability analysis caused by differences in user terminals, leveraging data convergence and pruning algorithms to address this issue. This results in higher accuracy and efficiency for the entire indoor positioning system, enhancing the commercial value of the entire indoor positioning solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 Schematic diagram of the positioning system of the present invention;

[0060] Figure 2 This is a schematic diagram of an exemplary environment for a Bluetooth AOA positioning system;

[0061] Figure 3 This is a flowchart of the Bluetooth positioning process;

[0062] Figure 4 This is the flow chart of the BP algorithm state information stage;

[0063] Figure 5 Factorize factor graphs for data fusion;

[0064] Figure 6 Build and update cloud-based wireless maps for crowdsourcing mechanisms. DETAILED DESCRIPTION

[0065] The present invention discloses a multipath assisted Bluetooth AOA indoor positioning method, which is implemented based on the Bluetooth AOA positioning system. Figure 1 As shown in FIG, the Bluetooth AOA positioning system includes a user terminal, a Bluetooth base station and a cloud server.

[0066] The positioning system of the present invention is configured in the following exemplary environment: a room with three Bluetooth base stations, which serve as physical anchors (PAs): PA1, PA2, and PA3. The virtual anchor (VA) is a mirror image of the physical anchor PA, caused by reflection from a flat surface. Therefore, the position of the virtual anchor VA depends on the surrounding environment. For each reflection path, in a two-dimensional plane, the distance from the user terminal to the physical anchor PA is equal to the distance from the user terminal to the virtual anchor VA. As the user terminal moves, the position of the virtual anchor VA also remains stationary, provided the physical anchor PAs and the surrounding plane remain stationary. For example, a user terminal at an unknown location moves randomly in a room to form a user movement trajectory. For the sake of simplicity, we only take the physical anchor point PA3 at time n2 as an example. At time n2, the physical anchor point PA3 receives one line-of-sight (LOS) propagation path and two first-order specular non-line-of-sight (NLOS) propagation paths. Each first-order specular NLOS propagation path corresponds to a virtual anchor point. For the physical anchor point PA, the number of propagation paths received may change during the user's movement, and the existence of the virtual anchor point VA may also change, but the position of the original VA does not change, such as Figure 2 shown.

[0067] A signal is sent from the user terminal to the Bluetooth base station. After receiving the signal from the user terminal, the Bluetooth base station uploads the data information to the cloud server. The cloud server estimates the specular multipath component (MPC) parameters from the received data information. In this indoor positioning system, MPC parameters, namely AOA information, are used as input measurements. These measurements are data-associated (DA) with the physical anchor points PAs and virtual anchor points VAs. Therefore, the location information of the physical anchor points PAs, virtual anchor points VAs, and the user terminal can be calculated using the backpropagation (BP) algorithm. In a wireless map, when the locations of the physical anchor points PAs and virtual anchor points VAs are known, their location information can be used to achieve multi-point assisted positioning. The location information of the physical anchor points PAs and virtual anchor points VAs marked in the wireless map can be provided within the detection range of interest (ROI), facilitating rapid and accurate positioning of the user terminal. These estimated physical anchor points PAs and virtual anchor points VAs will be marked as potential features (PF) in the wireless map, so the wireless map can be constructed by recording the locations of physical anchor points PAs and virtual anchor points VAs.

[0068] In large, complex indoor spaces with demanding positioning times, estimating the coordinates of physical anchor points (PAs) and virtual anchor points (VAs) using only measurements from a single user terminal over a short period of time is extremely difficult. However, since the positions of physical and virtual anchor points (VAs) remain static in indoor environments, crowdsourcing can be used to leverage distributed collaboration among multiple users to collect a large number of measurements and obtain estimated positions. These estimates become increasingly accurate over time as more users are added to the system.

[0069] like Figure 3 As shown in the figure, in the entire indoor environment, the user terminal, Bluetooth base station, and cloud server are configured as follows:

[0070] User terminals: User terminals can be added to the positioning system at any time. Their number is variable and increases with the number of users. User terminal devices move with the user's movements. During movement, they send signals to the Bluetooth base station in discrete, fixed-length segments, waiting for the cloud server to return the calculated location information to the user terminal.

[0071] Bluetooth base stations are deployed indoors, with a fixed number of base stations. These base stations are fixed in position and only receive signals from user terminals. A user terminal in an indoor environment transmits signals to multiple Bluetooth base stations located within the environment. Upon receiving a signal from a user terminal, the base station packages the data according to the terminal's serial number and uploads it to a cloud server.

[0072] Cloud server: Receives data information uploaded by Bluetooth base stations, extracts MPC parameters, namely AOA information, from the data based on different user terminals and corresponding Bluetooth base stations, uses the AOA information as input measurement values, and uses the BP algorithm to solve the virtual anchor point VAs position and user terminal position corresponding to the physical anchor point PAs on the cloud server. The status information of the physical anchor point PAs and virtual anchor point VAs is used as features to build a wireless map. The status information of the physical anchor point PAs and virtual anchor point VAs in the wireless map is updated in real time through the data information continuously uploaded by the Bluetooth base station to achieve the purpose of updating the cloud wireless map. Finally, the user terminal location information calculated in real time by the cloud server is returned to the user terminal.

[0073] The specific implementation steps of the indoor positioning method are as follows:

[0074] Step 1: The user terminal that joins the positioning system sends a signal to the Bluetooth base station at discrete time periods. The transmitted signal is:

[0075] STX,n (t),

[0076] Where n=1…N, N is the number of discrete times.

[0077] Step 2: The Bluetooth base station receives the signal sent by the user terminal and uploads the received signals from different terminals to the cloud server.

[0078] The signal received by the Bluetooth base station is:

[0079]

[0080] Wherein j=1…J, J is the number of PAs, which is a known condition. is the total number of LOS and first-order mirror NLOS propagation paths, is the complex path gain, is the steering vector, For AOA, S TX,n (t) is the sending signal, For delay, is diffuse multipath interference, is additive white Gaussian noise.

[0081] Step 3: The cloud server receives the data information uploaded by the Bluetooth base station and extracts the MPC parameters from the data. Here, the MPC parameters are AOA information. The AOA estimation algorithm is used to extract the data and the AOA information corresponding to the different user terminal data and the Bluetooth base station is used as the input measurement value.

[0082] Step 4: The cloud server needs to solve the anchor point position and the user terminal position. The user terminal position is recorded as x n .

[0083] In the wireless map, the locations where anchor points may exist are recorded as PFs. For PFs, its index is defined as (j, k), where For each PAj, there exists PFs, and the PFs set contains the PAs themselves, and the position of the PF is expressed as In the positioning system, the user terminal position x n and anchor position Is an unknown quantity. Whether the (j,k)th PF is an actual feature is determined by the variable Indicates that, when When , it means that PF does not exist at time n. On the contrary, when When , it means that PF exists at time n. Considering its existence, the state of the feature can be defined as Also take into account the measurement error.

[0084] Therefore, let the characteristic state represents the measurement error, where α n Represents the angular deviation of the user terminal device, is the path loss exponent (PLE), which changes according to the environment.

[0085] In the entire indoor positioning system, considering the uncertainty of data association, the estimated measurement value can come from traditional features, new features, or no features. The value that does not come from any feature is called a false alarm. The traditional feature refers to the feature that exists at time n-1, and the new feature refers to the feature that does not exist at time n-1 but exists at time n. The measurement value sets of traditional features, new features, and false alarms are respectively denoted as The set of measurements can therefore be divided into Three subsets, the measurement value set is recorded as have:

[0086]

[0087] in, For the data association between measurements and features, it is assumed that in any discrete time period, time n, each PF can generate at most one measurement value, and each measurement value can be generated by at most one PF.

[0088] For formula (2), So you can set dimensional feature-oriented data association vector and dimensional measurement-oriented data association vector The specific expressions are as follows:

[0089]

[0090]

[0091] where a n and b n are equivalent, and the two can be deduced from each other, so the BP algorithm can be constrained, which is defined as follows:

[0092]

[0093] in,

[0094]

[0095] The likelihood function based on the position state measurement distribution of user terminals and features and the data association vector can be defined as:

[0096]

[0097]

[0098] where f false (·) represents the probability density function of each false alarm measurement value, when in and Represent the status of traditional features and new features respectively. and The distribution is expressed as

[0099] The joint posterior probability distribution function of the state information of the user terminal and the feature and the data association depends on the measurement data collected at time N and can be defined as:

[0100]

[0101] According to the Bayesian principle, the above formula can be described as follows using the BP algorithm:

[0102]

[0103] The implementation significance of this formula represents the result of data fusion. Indicates the number of measurements. (a), (b), and (c) in formula (9) have the following meanings:

[0104] Represents the process of state transition. Represents the process of measurement and evaluation, Represents the process of data association.

[0105] Given the posterior probability, in the discrete time period N, the user terminal state x n The minimum mean square error (MMSE) estimator of can be defined as:

[0106]

[0107] where f(x n ∣z 1:N )=∫ w f(x 1:N ,y1:N ,a 1:N ,b 1:N ∣z 1:N )dw is a marginal posterior distribution in formula (9), w=[x 1:N-1 ,y 1:N ,a 1:N ,b 1:N ]. Its posterior existence probability is expressed as:

[0108]

[0109] in is a marginal posterior distribution in formula (9). According to Bayes’ theorem, we can define:

[0110]

[0111] Feature Location The MMSE estimator can be approximately defined as:

[0112]

[0113] For formula (9), it is not possible to directly calculate the marginal posterior probability density function, so it is necessary to approximate the marginal posterior probability density function through the BP algorithm.

[0114] In the positioning system, the relationship between the AOA measurement value, the feature position, and the position of the user terminal is expressed as:

[0115]

[0116] The specific steps for calculating the location information of physical anchor points, virtual anchor points, and user terminals based on the BP algorithm are as follows:

[0117] (1) State transition: For the user terminal state x n and traditional characteristic states It can be expressed in terms of Markov state dynamics as:

[0118]

[0119] where f(x n |x n-1 ) is represented by the following motion model:

[0120]

[0121] Among them, A is the state transfer matrix, ΔT is the sampling period, d n is a motion process with a mean of zero and a covariance of Gaussian distribution.

[0122] Because α n The value of has nothing to do with the state of the virtual anchor point, so So there is Therefore, the state transfer function of the feature can be expressed as And at time n-1, if a feature does not exist, then at time n, it cannot exist as a traditional feature, so when You can define:

[0123]

[0124] Among them, f D (·) is the probability density function of any setting;

[0125] If a feature exists at time n-1, the probability that the feature exists at time n is determined by its production probability. , you can define:

[0126]

[0127] Among them, P s (·) represents the survival probability of the feature.

[0128] (2) Measurement evaluation: Since the measurement values ​​come from traditional features, new features, and false alarms, the likelihood function can be updated as:

[0129]

[0130] because is a constant, Under known conditions, that is, when the total number of measurements is known, measurement evaluation can be defined as:

[0131]

[0132] (3) Data association: data association vector The joint prior probability density function, the number of measurements and the status of the new feature Depends on traditional feature status User terminal status x n and the unknown measurement deviation α n , expressed as:

[0133]

[0134] in, and Represents the average number of new features and false alarms, P d(·)∈(0,1] represents the probability that the feature is detected, that is, whether a measurement value is generated. new (·) represents the probability density function of the new detection feature.

[0135] (4) Data fusion: The joint posterior distribution of the user terminal state, measurement deviation, feature state, and data association vector depends on the measurement values ​​in all discrete time periods: f(x 1:N ,α 1:N ,y 1:N ,a 1:N ,b 1:N ∣z 1:N ). According to formula (9), the product of formulas (15), (20), and (21) is the joint posterior distribution. Formulas (20) and (21) are expressed in a more concise way.

[0136] set up is a set of traditional features that generate measurements at time n, so for have and is a set of traditional features that do not generate measurement values ​​at time n. have and And when Sometimes, there are After finishing, we can get:

[0137]

[0138] In addition, It is defined as a set of new features that generate measurements at time n. have and is a set of traditional features that do not generate measurement values ​​at time n. have and And when Sometimes, there are After finishing, we can get:

[0139]

[0140] Finally, the joint posterior probability density function is sorted out to obtain:

[0141]

[0142] The factor graph of the factorization in formula (24) is as follows Figure 5As shown, during the entire process, messages are forwarded in a timely manner, data information is calculated in a default order, and the calculated data information can be presented to the user in a timely manner, realizing real-time positioning of the user terminal.

[0143] The data information will first go through the state transition stage, with the data information from time n-1 passing through the factor node f x 、f α 、 To generate prior information, the measurement evaluation is then calculated by the factor node of the traditional feature and parallel factor nodes for new features To complete. Factor node and The output data are passed to the data association variable node In the Iterative calculation is performed between nodes, which is called the cyclic data association phase. After the last iterative calculation process, the data information is sent from the node Transmit to node and slave nodes Transmit to node Data information is provided by factor nodes Update. Finally, the fusion node x n , α n 、 When the message is reliable, the confidence level will be close to the expected marginal posterior probability density function.

[0144] Step 5: Use the feature status information to construct and update the cloud wireless map, and return the calculated user terminal location to the user terminal.

[0145] In order to further improve the performance of indoor positioning systems, it is necessary to quickly build cloud-based wireless maps. Since the feature states observed by each user terminal in the indoor environment are incomplete and biased, a crowdsourcing mechanism is used to build and update wireless maps, such as Figure 6 As shown in the figure, a cloud-based wireless map is constructed and updated through multi-user decentralized collaboration.

[0146] Define a collection It contains the estimated feature state of user terminal k as the wireless feature map of user terminal k, as shown below:

[0147]

[0148] and

[0149]

[0150] Where n' represents the convergence time of the SLAM algorithm at user terminal k, Representation characteristics The existence probability of , the subscript k is used to distinguish different user terminals. In the cloud, additional definition of the set To share wireless maps, collect is a collection A weighted combination of, where K represents the total number of user terminals, the set It is a dynamic collection that needs to continuously receive data information from the Bluetooth base station for update.

[0151] Assuming K1 is the total number of user terminals at present, we can define:

[0152]

[0153]

[0154] Since different user terminals have different hardware conditions or join the positioning system at different time points, the prior information obtained from the cloud is different, resulting in different reliability of the feature states obtained. Therefore, in the crowdsourcing mechanism, it is necessary to introduce a weight coefficient w k,n' , which indicates the reliability of the feature in the prior information obtained by the k-th user terminal at time n'. Therefore, the estimated feature state generated by the k-th user terminal can be expressed as:

[0155]

[0156] in Representation characteristics reliability, as the estimation errors of different user terminals may cause the Different features in the

[15] point to the same feature location in reality. Through the data association proposed in step 4, the obtained measurement value can be associated with the more reliable traditional features, thereby filtering out unreliable traditional features and finally achieving the approximate total number of measurement values ​​of the feature number estimated by each user terminal.

[0157] The cloud server analyzes the features estimated by the user terminal k and compares them with the set Combine and generate a complete shared wireless map in the following ways

[0158]

[0159] As time goes by, the feature states estimated by different user terminals will converge to the same position, and the accuracy of the shared wireless map stored in the cloud will become higher and higher, thus providing more accurate prior information for the next user terminal target positioning. For the setting of weights, based on the observation that the feature value estimated by the latter user terminal is more accurate than the feature value estimated by the previous user terminal, the set The weight in is set to be proportional to the upload time, that is, when n1<n2,

[0160] But as time goes by, the set The amount of data in the collection is also increasing, so in order to prevent As the data in increases, pruning operations are performed based on the reliability of each feature.

[0161] Let P s n,l For the collection The reliability of the lth feature in threshold is the threshold, If P s n,l <P threshold , then the lth feature will be from the set By pruning, the storage space of storage resources can be greatly saved and unreliable features can be eliminated to improve the accuracy of positioning.

[0162] The above description is merely an embodiment of the present invention and does not limit the technical scope of the present invention. Therefore, any minor modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A crowdsourcing-based multipath-assisted Bluetooth AOA indoor positioning method, implemented based on an indoor positioning system. The indoor positioning system includes user terminals, Bluetooth base stations, and a cloud server. The user terminals periodically send signals to the Bluetooth access point. The Bluetooth base stations package the received signals according to the sequence numbers of different user terminals and send them to the cloud server. The method is characterized by: The cloud server uses the Bluetooth base station as a physical anchor point and sets a virtual anchor point. The position of the virtual anchor point is a mirror image of the physical anchor point and is caused by reflection from a flat surface. For each reflection path, in the case of a two-dimensional plane, the distance from the user terminal to the physical anchor point is equal to the distance from the user terminal to the virtual anchor point. The cloud server also stores a cloud wireless map, which is constructed by physical anchor points and virtual anchor points of user terminals that have completed positioning; When the cloud server receives data sent by the Bluetooth base station, it estimates the specular multipath component parameters from the received data and processes the specular multipath component to obtain AOA information; then, the AOA information is used as the measurement value, and the physical anchor point and virtual anchor point position information are obtained from the cloud wireless map as the feature state, and there is a data association between the feature state and the measurement value; the cloud server calculates the user terminal position information corresponding to the measurement value and the corresponding physical anchor point and virtual anchor point position information based on the feature state and the measurement value using the BP algorithm; and returns the user terminal position information to the user terminal, and uses the calculated physical anchor point and virtual anchor point to update the cloud wireless map. Specifically, the user terminal position information at time n and the corresponding physical anchor point and virtual anchor point position information are obtained through state conversion, measurement evaluation, data association and data fusion. (1) The state is converted to: The user terminal status at time n-1 , measure angle deviation With feature status , the user terminal state at time n is obtained by independent evolution according to Markov state dynamics , measure angle deviation and traditional characteristic states , expressed as a probability density function: in, The state transfer function of is represented by the following motion model: In formula (15), is the user terminal status at time n-1, is the measured angle deviation at time n-1, is the characteristic state at time n-1, is the user terminal status at time n, To measure the angle deviation at any moment, is the traditional characteristic state at time n, J is the number of PAs, is the number of traditional features, is the dth traditional feature vector generated by the jth PA at time n, that is, the traditional feature state generated by the PF with index (j, d), is the dth eigenvector generated by the jth PA at time n, that is, the characteristic state generated by the PF with index (j, d), , represents the measurement error, [ ],in Represents the angular deviation of the user terminal device, is the path loss index, which changes according to the changes in the environment; In formula (16), A represents the state transfer matrix, is the sampling period, is a motion process with a mean of zero and a covariance of Gaussian distribution.

2. The multipath-assisted Bluetooth AOA indoor positioning method based on crowdsourcing mechanism according to claim 1 is characterized in that: If a feature does not exist at time n-1, it cannot exist as a traditional feature at time n. ,definition: in, is the probability density function of any setting; If a feature exists at time n-1, the probability that the feature exists at time n is determined by its production probability; when When, define: in, represents the survival probability of the feature; (2) Measurement and evaluation; When the measured value is known, the user terminal status , eigenvector state , measurement angle error , data association vector , measurement quantity The probability measurement distribution of , its likelihood function is updated as: In formula (19), is the observation value generated by the jth PA at time n, is the user terminal state at time n, is the angle deviation measured at time n, The characteristic state of A generated by the j-th PA at time n, is the feature-oriented data association vector of the j-th PA at time n, is the total number of measurements generated by the j-th PA at time n, is the data association vector represented by the dth eigenvector generated by the jth PA at time n, is the probability density distribution function of false alarms, is the traditional feature set that generates measurements at time n, Generate a new feature set of measurements at time n, is the total number of features, since is a constant, Under known conditions, that is, when the total number of measurements is known, measurement evaluation is defined as: (3) Data association; data association vector The joint prior probability density function, the number of measurements and the status of the new feature Depends on traditional feature status , User terminal status and unknown measured angle deviation , expressed as: in, is the data vector constraint, and Represent the average number of new features and false alarms, is the set of traditional features that have no measurement value generated at time n, is the set of new features that have no measurement values ​​generated at time n, Represents the probability of a feature being detected, that is, whether a measurement value is generated; The probability density function representing the new detected feature; (4) Data fusion: The joint posterior distribution of the user terminal state, measurement deviation, feature state, and data association vector depends on the measurement values ​​in all discrete time periods: ,in Represents the measured value from time 1 to N, Indicates the user terminal status from time 1 to N, Indicates the measurement angle error from time 1 to time N, Represents the characteristic state from time 1 to N, Represents the data association vector from time 1 to N; set up is a set of traditional features that generate measurements at time n, so for have ; is a set of traditional features that do not generate measurement values ​​at time n; therefore, for have ; and when Sometimes, there are , we can get: In addition, It is defined as a set of new features that generate measurements at time n. have ; is a set of traditional features that do not generate measurement values ​​at time n; therefore, for have ; and when Sometimes, there are , we can get: Finally, the joint posterior probability density function is sorted out to obtain: According to formula (24), the location information of the user terminal at time n and the corresponding physical anchor point and virtual anchor point location information can be obtained.

3. The method for indoor positioning using multipath-assisted Bluetooth AOA based on crowdsourcing mechanism according to claim 2, characterized in that: The cloud server updates the cloud wireless map as follows: Define a collection , which contains the estimated feature state of user terminal k as the wireless feature map of user terminal k, as shown below: and in, represents the convergence time of the SLAM algorithm at user terminal k, Representation characteristics The existence probability of , the subscript k is used to distinguish different user terminals; in the cloud, an additional set is defined To share wireless maps, collect is a collection A weighted combination of, where K represents the total number of user terminals, the set It is a dynamic collection that needs to continuously receive data information from the Bluetooth base station for update; Assumptions is the total number of current user terminals, then define: The cloud server analyzes the features estimated by the user terminal k and compares them with the set Combine and generate a complete shared wireless map in the following ways (30)。 4. The multipath-assisted Bluetooth AOA indoor positioning method based on crowdsourcing mechanism according to claim 2 is characterized in that: Introducing weight coefficient , used to indicate the kth user terminal at time Therefore, the estimated feature state generated by the k-th user terminal is expressed as: in Representation characteristics Reliability will be collected The weight in is set to be proportional to the upload time, i.e. when hour, ; set up For the collection The reliability of the lth feature in , is the threshold, ,like , then the lth feature will be from the set Deleted.

Images