A method for improving indoor positioning accuracy by fusing wi-fi, bluetooth and 4g signals

Abstract

This invention relates to the field of wireless positioning technology, specifically disclosing a method for improving indoor positioning accuracy by fusing Wi-Fi, Bluetooth, and 4G signals. The method first synchronously collects and timestamps-aligns Wi-Fi, Bluetooth, and 4G signals and data; then, it extracts distance or location observations and dynamically evaluates the uncertainty of each observation; next, it constructs a factor graph-based fusion model, applying the inertial factor generated by pre-integration and the observation factor generated by the wireless signal to the terminal state node to be determined; finally, through nonlinear optimization, it solves for the optimal motion trajectory in one step. This invention deeply fuses heterogeneous signals using a probabilistic graphical model, utilizes the wide-area coverage of 4G to compensate for the blind spots of Wi-Fi / Bluetooth, uses the continuity to smooth the fluctuations of wireless signals, and intelligently allocates weights through uncertainty modeling, ultimately achieving comprehensive coverage, high accuracy, and strong robustness in indoor positioning.

Description

Technical Field

[0001] This invention relates to the field of wireless positioning technology, specifically to a method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals. Background Technology

[0002] Indoor positioning is a key technology in fields such as the Internet of Things and smart cities, but single technologies have inherent drawbacks: Wi-Fi positioning typically uses fingerprinting or triangulation methods. Fingerprinting requires extensive preliminary data collection, and environmental changes can easily lead to decreased accuracy. Triangulation methods have significant errors due to the non-line-of-sight propagation and multipath effects of Wi-Fi signals; Bluetooth positioning, especially based on Bluetooth beacons (… The signal strength indication method of the receiver is simple to deploy, but the signal strength is easily affected by human body obstruction and environmental changes, resulting in poor stability and large fluctuations in accuracy. Outdoor positioning based on time difference of arrival and other methods has high accuracy, but indoor positioning accuracy is extremely low due to the long distance between base stations, large signal penetration loss, and the possibility of multiple cell handovers indoors. Inertial navigation has strong autonomy, but it has cumulative errors and will diverge in a short period of time.

[0003] Existing convergence solutions are mostly simple switching or weighted averaging, such as using Wi-Fi when the Wi-Fi signal is strong and Bluetooth when it is not. These solutions fail to fundamentally solve the inherent defects of various signals and cannot provide a reliable solution in environments with sparse or harsh signals.

[0004] Therefore, a collaborative localization method is needed that can deeply explore and integrate the advantages of multiple heterogeneous signals and intelligently compensate for their respective disadvantages. Summary of the Invention

[0005] To address the technical problems mentioned in the background section, the present invention aims to provide a method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals. This method constructs a unified, factor graph-optimized fusion framework that tightly couples the observations of different signals with inertial navigation data to achieve optimal estimation.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals includes:

[0008] S100: The terminal device synchronously receives and timestamps wireless signals from at least one Wi-Fi access point, at least one Bluetooth beacon, and at least one 4G base station, and synchronously collects built-in... Inertial data;

[0009] S200: Extracts the received Wi-Fi and Bluetooth signal strengths and converts them into distance observation values ​​respectively. At the same time, it quantifies the uncertainty parameters by using the fixed characteristics of Wi-Fi and the fluctuation characteristics of Bluetooth respectively. It analyzes the 4G cell identifier and signal parameters to determine the terminal location constraints and base station distance range, and assigns the corresponding uncertainty parameters for the distance range in combination with the low-precision characteristics of 4G.

[0010] S300: Construct a factor graph model based on Bayesian networks, while simultaneously... Inertial data is pre-integrated into relative constraints, which serve as inertial factors connecting state nodes at adjacent time points, providing continuous motion prediction.

[0011] Each of the various distance observations is modeled as a factor of the corresponding category, and the current state node is connected to the fixed node at the known position through the factors; at the same time, a noise model and information matrix are assigned to each factor according to the uncertainty parameters corresponding to each of the various distance observations; the constructed factor graph is solved by a nonlinear optimization method, and the optimal estimate of the state at all times is obtained in one optimization by maximizing the posterior probability, that is, the final motion trajectory.

[0012] S400: The optimized final motion trajectory is bound to the Wi-Fi fingerprint collected at that time, and the fingerprint database is dynamically updated or expanded for subsequent positioning and to improve the positioning accuracy of other terminals.

[0013] Furthermore, in step S300, the state variables of the factor graph model are the temporal position, velocity, and attitude of the terminal.

[0014] Furthermore, in step S100, the terminal device first initiates clock reference initialization to establish a unified time reference, assigning a unique timestamp to all data to be collected; for The acquired triaxial acceleration data underwent gravity separation processing, and the triaxial angular velocity data underwent zero-bias calibration processing, followed by... Using a unified sampling frequency, interpolation or point-sampling simplification processes were performed on Wi-Fi, Bluetooth, and 4G signal data respectively. Linear interpolation was used for interpolation, while point-sampling simplification employed a sliding window-based mean sampling algorithm to ensure all data timestamp intervals were uniformly matched. Collection cycle.

[0015] Furthermore, in step S200, the Wi-Fi distance observation is obtained by comparing the initial distance with the newly sampled distance. Value weighted average correction, corresponding uncertainty parameter Based on corrected distance observations Path loss index and shadow fading coefficient The quantification is performed using the following formula: .

[0016] Furthermore, in step S200, the Bluetooth distance observation value is obtained from the most recent three samplings. The mean is smoothed to obtain the smoothed Bluetooth distance observation. The corresponding uncertain parameters Combining signal intensity fluctuations from nearly 5 sampling points Historical stability coefficient Bluetooth path loss index The quantification is performed using the following formula: .

[0017] Furthermore, in step S200, through parsing The global identifier of the cell used to signal the coverage area of ​​the serving cell / neighboring cell is used as a location constraint. Based on the initial distance of the timing lead conversion, the corrected distance interval is determined. Then through the formula Assign corresponding uncertainty parameters ;in, This is the minimum value within the 4G corrected distance range. This is the maximum value.

[0018] Furthermore, in step S300, the 4G factor is modeled using distance range residuals, and non-zero residuals are generated only when the position estimate exceeds the distance range; the information matrix is ​​obtained by inverting the uncertainty parameters of each factor: the inertial factor information matrix is ​​a diagonal matrix, and the Wi-Fi, Bluetooth and 4G factor information matrices are the reciprocals of the corresponding distance uncertainty parameters; the global objective function is the weighted sum of squared residuals, which is solved iteratively using the Gauss-Newton method to obtain the optimal estimates of position, velocity and attitude at all times in one optimization.

[0019] Furthermore, in step S400, the Wi-Fi fingerprint construction employs anomaly removal, normalization, and confidence evaluation processes, specifically: through... Criteria for removing anomalies If an outlier is found, it will be replaced with the average of the last three values ​​for that access point, and then the valid values ​​will be updated. Value normalized to Finally, the confidence level of the fingerprint-location association pair is calculated within the interval.

[0020] Furthermore, in step S400, the fingerprint database uses a 0.5m grid index, and the maintenance strategy is as follows: when the confidence level is ≥0.7, a new fingerprint is added; when a record already exists, the similarity between the new fingerprint and the existing fingerprint is calculated. If the similarity is ≥0.8, the position and confidence level are updated with weights. If the similarity is <0.8 and the new confidence level is ≥0.7, a redundant fingerprint is added.

[0021] Furthermore, in step S400, the fingerprint database update verification mechanism is established as follows: for every 10 fingerprints updated, 20 records are randomly selected for simulated positioning and the average positioning error is calculated. If the average deviation between the test location and the actual location is <0.3m, the update is valid; otherwise, the confidence threshold for new fingerprints is lowered to 0.5 and the update is re-executed. If the average positioning error is ≥0.3m, the confidence threshold for new fingerprints is lowered to 0.5, and this threshold is maintained until the average error of the next simulated positioning verification is <0.3m. The threshold is then restored to 0.7, and the minimum confidence threshold is lowered to 0.5. If the error requirement cannot be met after the lowering, the fingerprint database recalibration process is triggered.

[0022] Compared with the prior art, the advantages of the present invention are as follows:

[0023] 1. This invention uses a factor graph optimization framework to tightly couple observation information with different accuracies and reliability to the motion model, achieving information complementarity, effectively suppressing fluctuations and errors of single signals, and stabilizing the positioning accuracy to the 1-3 meter level.

[0024] 2. This invention uses 4G signals as anchor points for wide-area coverage, providing the most basic location constraints that will not be completely lost. This effectively prevents the positioning from becoming completely scattered when Wi-Fi and Bluetooth signals are in dead zones or completely fail. At the same time, the system can intelligently and dynamically adjust the weights of each factor according to signal quality, ensuring that the positioning does not diverge in complex environments.

[0025] 3. This invention, through uncertainty modeling, enables the system to automatically reduce the weight of volatile Bluetooth signals or weak Wi-Fi signals and place more trust in stable observations, thereby making it more adaptable to environmental changes.

[0026] 4. This invention relies on the fingerprint database self-learning mechanism, which enables the system to use high-confidence positioning results to reinforce the fingerprint database, forming a positive feedback loop that becomes more accurate with use, thus reducing long-term maintenance costs. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the overall workflow of the method of the present invention;

[0029] Figure 2 This is a flowchart of the steps for calculating multi-source signal observations and modeling uncertainty in this invention;

[0030] Figure 3 This is a flowchart of the factor graph model fusion optimization steps of the present invention;

[0031] Figure 4 This is a schematic diagram of the factor graph model in step S300 of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] To achieve the above objectives, the present invention provides a method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals, such as... Figures 1-4 As shown, the system includes:

[0034] S100: The terminal device synchronously receives and timestamps wireless signals from at least one Wi-Fi access point, at least one Bluetooth beacon, and at least one 4G base station, and synchronously collects built-in... Inertial data.

[0035] S110: After the terminal device starts the positioning function, it performs a clock reference initialization operation, which is: starting the built-in clock module to establish a unified time reference, assigning a unique timestamp to all data to be collected, and completing the pre-processing for time alignment of multi-source data.

[0036] S1101: Synchronously receive Wi-Fi access point broadcast signals and extract received signal strength indication. (Unit: dBm) Record the known coordinates of the Wi-Fi access point corresponding to the currently received signal. The signal strength is converted into the initial distance between the terminal and the access point using a log-normal shadowing model. The calculation formula is ;in, The transmit power of the Wi-Fi signal at a reference distance of 1m (unit: dBm); Shadow fading factor (unit: dB); The Wi-Fi path loss index in an indoor environment; the initial distance between the terminal and the access point. As the basis for the distance observations in the S200 step described later;

[0037] In this embodiment, combining general technical knowledge in the field of wireless communication and industry-standard experience values ​​for indoor Wi-Fi signal propagation, the Wi-Fi path loss index is... The range of values ​​is ;

[0038] S1102: Synchronously receive Bluetooth beacon broadcast signals and extract received signal strength indication. (Unit: dBm) Record the known coordinates of the Bluetooth beacon corresponding to the currently received signal. Based on the free-space path loss model, the signal strength is converted into the initial distance between the terminal and the beacon. The calculation formula is: ;in, This represents the Bluetooth signal transmission power at a reference distance of 1m (unit: dBm). The Bluetooth path loss index is given in indoor obstructed scenarios; the initial distance between the terminal and the beacon. As the basis for the distance observations in the S200 step described later;

[0039] In this embodiment, industry-standard engineering parameters for Bluetooth indoor positioning are combined with... , The Bluetooth path loss index refers to the standard value range specified by Bluetooth chip manufacturers in their positioning solution design manuals. The range of values ​​is ;

[0040] S1103: Synchronously receive downlink signals from 4G base stations and analyze them to obtain timing advance. (Unit: chip) Record the known coordinates of the 4G base station corresponding to the currently received signal. The timing advance is converted into the initial distance between the terminal and the base station. The calculation formula is: ;in, The speed of electromagnetic wave propagation; The chip rate is 4G signal bit rate; the initial distance between the terminal and the base station. This serves as the basis for position constraints in step S200 described later;

[0041] In this embodiment, the speed of electromagnetic wave propagation in a vacuum, a universally accepted standard parameter in physics for calculating signal propagation distance in communication technology, is used. Values Combining the standard technical parameter requirements of 4G communication networks, and Formulated The standard for , The system downlink chip rate specification requirements, the 4G signal chip rate Values chip s;

[0042] S120: Synchronous Acquisition triaxial acceleration data (unit: ) and triaxial angular velocity data (Unit: rad / s);

[0043] Regarding the aforementioned triaxial acceleration data Gravity separation is performed, specifically through calculation. The effective acceleration after removing gravitational interference from the triaxial acceleration data was calculated. ;in, ; It is the acceleration due to gravity; for The unit gravity vector corresponding to the axis; finally, the effective triaxial acceleration data after removing gravity interference is obtained. ;

[0044] Regarding the aforementioned three-axis angular velocity data Perform angular velocity zero-bias calibration, specifically by calculating... Calculation Calibration angular velocity after eliminating zero drift in three-axis angular velocities ;in, The angular velocity zero bias value is obtained from the pre-static calibration; finally, the calibrated triaxial angular velocity data is obtained. ;

[0045] Finally, the preprocessed effective acceleration is... and the calibrated angular velocity As the basic material for motion constraint modeling in the S300 step described later;

[0046] S130: with sampling frequency To ensure a unified benchmark, the collected Wi-Fi, Bluetooth, and 4G signal data were processed accordingly. Specifically, if the target data collection frequency was lower than [a certain value], [further processing was performed]. If the target data collection frequency is higher than 100%, then interpolation is performed to supplement the data points; Then, it performs point sampling to simplify the data points, ultimately ensuring that the timestamp intervals of all the above data are uniformly matched. This is done to achieve time synchronization of multi-source data, ensuring that subsequent steps call upon multi-source information corresponding to the same moment.

[0047] S200: Extracts the received Wi-Fi and Bluetooth signal strengths and converts them into distance observation values ​​respectively. At the same time, it quantifies the uncertainty parameters based on the differences between the fixed characteristics of Wi-Fi and the fluctuation characteristics of Bluetooth. On this basis, it analyzes the 4G cell identifier and signal parameters, determines the terminal location constraints and base station distance range, and assigns the corresponding distance range uncertainty parameters in combination with the low precision characteristics of 4G.

[0048] S210: Based on the initial distance between the terminal and the access point in step S1101. Combined with the newly sampled received signal strength indication (Unit: dBm) Calculate Wi-Fi corrected distance observations The calculation formula is:

[0049]

[0050] in, The current sampled shadow fading coefficient; This refers to the Wi-Fi signal path loss index.

[0051] In this embodiment, since the sampled value of the shadow fading coefficient needs to fluctuate with the real-time environmental signal, and based on research on indoor Wi-Fi positioning, it is known that the single sampled value of the shadow fading coefficient in the 2.4GHz band under normal office conditions is typically distributed within ±3dB, with 1.8dB being one of the typical values ​​within this range. Therefore, the currently sampled shadow fading coefficient... The value is 1.8dB;

[0052] Based on the corrected distance observations and the shadow fading coefficient in step S1101 Through calculation formula The Wi-Fi corrected distance observation value was calculated. Corresponding uncertainty parameters ;

[0053] S220: Based on the initial distance between the terminal and the beacon in step S1102. Bluetooth path loss index Smoothing optimization and uncertainty assessment are performed, specifically as follows:

[0054] First, extract the average of the received signal strength indicators from the last three samples of the Bluetooth signal. (Unit: dBm), and calculated using the formula To calculate Bluetooth smoothed distance observations ;

[0055] Secondly, the signal strength fluctuations at nearly 5 sampling points were combined. With historical stability coefficient Through calculation formula The distance observation value was calculated. Corresponding uncertainty parameters ;

[0056] In this embodiment, since the historical stability coefficient is a commonly used parameter in wireless signal observation smoothing algorithms, research on indoor Bluetooth positioning indicates that the value range of this coefficient is typically set to [value range missing]. When the signal transmission environment is stable (such as an office area without frequent personnel movement), the coefficient value is mostly within... Between these values, 0.8 is a typical value for this scenario, hence the historical stability coefficient. The value is 0.8;

[0057] S230: Initial 4G distance obtained based on S1140 Combined with the analyzed service cell / neighbor cell Corresponding base station coordinates , Assign base station numbers, determine the regional constraint range (serving cell coverage area) of the terminal, and calculate the 4G corrected distance interval. The calculation formula is: ;in, This is the minimum value within the 4G corrected distance range. It is the maximum value;

[0058] Considering the low accuracy of 4G positioning, a calculation formula is used. The corrected distance interval was calculated. Corresponding uncertainty parameters ;

[0059] In this embodiment, the coefficients 0.9, 1.1 and 0.1 are obtained by fitting 100 sets of indoor 4G signal measured data. They are statistical correction results of the 4G initial distance observation error distribution and can cover the 4G distance observation fluctuation range in more than 95% of scenarios.

[0060] S240: Finally, the Wi-Fi corrected distance observation. and the corresponding uncertainty parameters The Bluetooth smoothed distance observation value and the corresponding uncertainty parameters The 4G corrected distance range and the corresponding uncertainty parameters ,right The observed data undergoes pre-integration processing to obtain displacement, velocity, and attitude residuals, and the corresponding standard deviations of uncertainty parameters are calculated, specifically: standard deviation of displacement residuals. Velocity residual standard deviation Posture residual standard deviation The above data will serve as the core input parameters for subsequent multi-source fusion positioning.

[0061] S300: Construct a factor graph model based on Bayesian networks, using the terminal's temporal position, velocity, and attitude as core state variables, and... Data pre-integration provides continuous motion prediction for connecting adjacent state nodes to inertial factors. Wi-Fi, Bluetooth distance observations, and 4G location distance observations are modeled as corresponding factors, connecting the current state node with known fixed nodes. Combining the uncertainty of dynamic evaluation of various observations, an appropriate noise model and information matrix are configured for each factor. The factor graph is solved by nonlinear optimization method to maximize the posterior probability to obtain the optimal state estimate for all time moments at once, and output the high-precision motion trajectory of the terminal.

[0062] S310: First, construct a Bayesian network factor graph model to clarify the state nodes and fixed nodes, specifically:

[0063] For the aforementioned state node, regarding the time sequence... , The total number of time steps is given, and the state node at each time step is defined as follows: ;in, Indicates that the terminal is in The three-dimensional position at time (unit: m), and ; Indicates that the terminal is in The three-dimensional velocity at time t (unit: m / s), and ; Indicates that the terminal is in The attitude quaternion at time, and Satisfying the normalization constraint ;

[0064] For the fixed node, the reference node at a known location is connected, including:

[0065] Wi-Fi access point coordinates , Total number of Wi-Fi access points; Bluetooth beacon coordinates , Total number of Bluetooth beacons; 4G base station coordinates , This represents the total number of 4G base stations and provides a reference benchmark for subsequent observation constraints.

[0066] S320: Based on the distance observations and corresponding uncertainty parameters in step S200, and the values ​​in step S100... Preprocess the data, converting different data sources into constraint factors of the factor graph to connect the state nodes and the fixed nodes;

[0067] S3201: Regarding step S100... Effective acceleration and calibrating angular velocity Perform pre-integration, where correspond At time t, the relative motion relationship between adjacent time points is obtained, and it is modeled as an inertia factor to connect adjacent state nodes. The calculation formula is:

[0068]

[0069] in, for The pre-integrated residual vector contains the displacement residuals. Velocity residual and attitude residual It is used to quantify the deviation between motion prediction and actual state; for and Sampling interval between adjacent time points (unit: s); This refers to quaternion multiplication operations; for Time-of-pose quaternions The conjugate quaternion is used for inverse attitude transformation; This is a quaternion exponential mapping used to convert the integral result of angular rate into attitude change quaternions;

[0070] S3202: Combine the Wi-Fi corrected distance observations from step S210. ,correspond Time terminal to the The distance observation value of each Wi-Fi node is The Wi-Fi factor is modeled to connect the current state node with the fixed Wi-Fi node, and the calculation formula is: ;in, for Time terminal to the The distance observation residuals of each Wi-Fi node reflect their three-dimensional location. and distance observations Deviation; This is the Euclidean distance operator used to calculate the 2-norm of a vector;

[0071] S3203: Based on the Bluetooth smoothed distance observation value in step S220 ,correspond Time terminal to the The distance observation value of each Bluetooth beacon is The Bluetooth factor is modeled to connect the current state node with the fixed Bluetooth node, and the calculation formula is:

[0072]

[0073] in, for Time terminal to the The distance observation residual of a Bluetooth beacon reflects the three-dimensional position. and distance observations Deviation;

[0074] S3204: Regarding the 4G-corrected distance range in step S230. ,correspond Time terminal to the The distance range of the 4G base stations is The 4G factor is modeled to connect the current state node with the 4G fixed node, and the calculation formula is as follows:

[0075]

[0076] in, for Time terminal to the The distance range residuals of each 4G base station are such that the residuals are 0 when the location estimate is within the distance range, and non-zero when it is outside the range, thereby achieving location constraints. for Time terminal to the The minimum distance between 4G base stations; for Time terminal to the The maximum distance between 4G base stations;

[0077] S330: Based on the uncertainty parameters and standard deviation data corresponding to the various observations output in step S240, configure an information matrix for each factor. The information matrix is ​​obtained by inverting the noise covariance matrix and is used to measure the confidence of the observation. The larger the value of the information matrix, the higher the confidence of the corresponding observation.

[0078] S3301: The formula for calculating the inertia factor information matrix is:

[0079]

[0080] in, Construct operators for diagonal matrices by using input elements as diagonal elements to generate the matrix;

[0081] S3302: The formula for calculating the Wi-Fi factor information matrix is ​​as follows: ;in, For the first One Wi-Fi access point in At any given time, correct the uncertainty parameter corresponding to the distance observation;

[0082] S3303: The formula for calculating the Bluetooth factor information matrix is ​​as follows: ;in, For the first A Bluetooth beacon in At time t, the uncertainty parameter corresponding to the smoothed distance observation;

[0083] S3304: The formula for calculating the 4G factor information matrix is ​​as follows: ;in, For the first 4G base stations At any given moment, the uncertainty parameters corresponding to the distance interval after 4G correction;

[0084] S340: Integrate the residuals of all factors with the information matrix to construct a global objective function, which is in the form of a weighted sum of squared residuals. Maximizing the posterior probability estimate in multi-source fusion localization is equivalent to minimizing this weighted sum of squared residual function, calculated as follows:

[0085] in, The optimal state estimate for all time points; For any time State Nodes (corresponding to the timeline in the previous text) (State sequence) Find the variable values ​​that minimize the objective function; the summation terms are the weighted sum of squared residuals of the inertia factor, Wi-Fi factor, Bluetooth factor, and 4G factor, respectively, reflecting the contribution of each constraint to the objective function;

[0086] To solve this nonlinear optimization problem composed of multiple factors, the Gauss-Newton method is used for iterative updating, and the corresponding iterative updating formula is:

[0087]

[0088] in, The incremental update vector of the state variables is expanded to represent all time steps. , , The increment; The global information matrix is ​​formed by piecing together the information matrices of each factor in a diagonal block format. The global residual vector is given by , , , It is assembled in sequence; This is a matrix transpose operation; Invert a matrix;

[0089] After the iterations converge, the optimal state estimate is extracted. Position sequence in This sequence represents the high-precision motion trajectory of the terminal; the convergence condition is the state increment. The length of the European modulus is less than the preset threshold. ;in, The optimal position at any given time satisfies ;

[0090] In this embodiment, considering both the numerical stability requirements of the Gauss-Newton method and the industry-standard multi-source fusion positioning, the selection of the convergence threshold for the Gauss-Newton method needs to balance computational efficiency and accuracy requirements. The magnitude is a typical value that balances both aspects, avoiding excessive iterations that lead to computational delays while ensuring that the accuracy of state estimation meets the error requirements of the positioning scenario; therefore, the threshold value is taken as... At this point, the magnitude of the decrease in the objective function has met the accuracy requirements.

[0091] S400: The optimized high-precision trajectory location is bound to the Wi-Fi fingerprint collected at that time, and the fingerprint database is dynamically updated or expanded for subsequent positioning and to improve the positioning accuracy of other terminals.

[0092] S410: The raw Wi-Fi signal synchronously acquired in step S100 is subjected to feature extraction and preprocessing to form a structured fingerprint, specifically:

[0093] S4101: Each fingerprint sample is a feature vector The expression is:

[0094]

[0095] in, For the first The physical address of each Wi-Fi access point; for Time of the first Signal strength of each access point (unit: dBm); For collection timestamps;

[0096] S4102: Use Criteria for removing anomalies Value, if Then replace it with the average value of the access point over the last 3 times. , The average of the existing data is taken; among them, , These are the historical records of this access point. The mean and standard deviation;

[0097] The mean The calculation formula is The standard deviation The calculation formula is ;in, This refers to the historical time sequence number of the access point. For the collected history The number of time points corresponding to the data;

[0098] S4103: Will Mapped to The interval, the formula is:

[0099]

[0100] in, for Time of the first Normalized results of signal strength of individual Wi-Fi access points; for Time of the first Raw signal strength observations for each Wi-Fi access point (unit: dBm); Minimum effective strength; Maximum effective strength;

[0101] In this embodiment, based on the general definition of the effective range of Wi-Fi signal strength and the selection of conventional parameters in the field of multi-source positioning, the minimum effective strength is... Values Maximum effective strength Values ;

[0102] Normalized fingerprints are ;in, For the first The physical address of each Wi-Fi access point; for The timestamp is used to mark the time when the fingerprint was collected;

[0103] S420: The steps in S340 are... optimal position at any time (unit: m) and Binding, forming an association pair ( ), and calculate the confidence score. The calculation formula is:

[0104]

[0105] in, Let be the uncertainty parameter for location estimation, and , Here is the location covariance matrix; for Time normalization Standard deviation;

[0106] The formula for calculating the standard deviation is: ;in, This corresponds to the mean;

[0107] In this embodiment, the confidence assessment of Wi-Fi fingerprint positioning in existing technologies is often achieved by fusing location estimation uncertainty and signal stability. For example, the location error is quantized using the covariance matrix trace based on Kalman filtering, and stability is measured using statistical characteristics of signal strength. Therefore, the confidence level... The closer the correlation is to 1, the more reliable the association pair is;

[0108] S430: Fingerprint Database Storage format is ;in, It serves as the grid index key for the database, used to quickly locate fingerprint records in the corresponding area; This is the standardized Wi-Fi fingerprint feature vector; For high-precision optimal position (unit: m); This represents the confidence value of the fingerprint-location association pair; This is the timestamp of the most recent update to the fingerprint record;

[0109] The following maintenance strategy is adopted:

[0110] S4301: Grid index by The grid is used to divide the positioning area, and the index key is:

[0111]

[0112] in, To locate the area The number of grid cells divided along the axis; To locate the area The number of grid cells divided along the axis; , and for The corresponding three-dimensional spatial coordinates;

[0113] S4302: The new logic is: if No record and Insert the associated pair directly;

[0114] S4303: Update logic is: If Existing records Calculate the similarity between the new fingerprint and existing fingerprints. ;in, A standardized fingerprint of an existing record in the database; The high-precision location corresponding to an existing record in the database; The confidence score is the score of an existing record in the database; The most recent update timestamp of an existing record in the database;

[0115] like Then the weighted calculation of the updated position The calculation formula is:

[0116] ;

[0117] like and Add a redundant fingerprint;

[0118] S440: For every 10 fingerprints updated, 20 records are randomly selected to simulate location verification, specifically:

[0119] The three most similar fingerprints are selected, and the localization result is calculated using a weighted average.

[0120]

[0121] in, The test location results obtained during the simulated positioning verification; For the first Similarity between candidate fingerprints and target fingerprints; For the first High-precision location corresponding to each candidate fingerprint;

[0122] The mean deviation between the test location and the actual location is calculated using the following formula:

[0123]

[0124] in, The spatial distance modulus; For the first The test location results obtained during the simulated positioning verification were recorded; For the first The actual location corresponding to each record;

[0125] If the average positioning error If the update is successful, then the confidence threshold for newly added fingerprints is lowered to 0.5; otherwise, the confidence threshold is lowered to 0.5. The confidence threshold for newly added fingerprints has been lowered to 0.5, and this threshold will be maintained until the next simulated location verification. The threshold is restored to 0.7, and the confidence threshold is lowered to a minimum of 0.5, and must not be lowered further. If the error requirement cannot be met after the reduction, the fingerprint database recalibration process is triggered.

[0126] This invention combines the wide reach and broad coverage of 4G, the solid foundation of Wi-Fi, and the precise and close-range capabilities of Bluetooth (especially at close range). The continuity is unified under a mathematical framework through the probabilistic graphical model. It is not a simple voting or switching, but a process of negotiation and compromise: the optimization algorithm will find an overall optimal solution. This solution may not completely match a certain observation (such as a obviously abnormal Bluetooth signal), but it is the overall best-fit solution for all observation information and motion constraints, thus achieving the effect of 1+1+1+1>4.

[0127] In this embodiment, existing fingerprint positioning systems often verify the effectiveness of database updates by combining simulated positioning with error threshold judgment. For example, setting centimeter-level / decimeter-level error thresholds to screen reliable fingerprints. In this embodiment, the design of using 0.3m as the average positioning error threshold and adjusting the newly added threshold to optimize the update logic is in line with the existing technical framework of error feedback iterative optimization of fingerprint database, which can effectively ensure the positioning accuracy of the database.

[0128] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0129] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals, characterized in that, include: S100: The terminal device synchronously receives and timestamps wireless signals from at least one Wi-Fi access point, at least one Bluetooth beacon, and at least one 4G base station, and synchronously collects built-in... Inertial data; S200: Extracts the received Wi-Fi and Bluetooth signal strengths and converts them into distance observation values ​​respectively. At the same time, it quantifies the uncertainty parameters by using the fixed characteristics of Wi-Fi and the fluctuation characteristics of Bluetooth respectively. It analyzes the 4G cell identifier and signal parameters to determine the terminal location constraints and base station distance range, and assigns the corresponding uncertainty parameters for the distance range in combination with the low-precision characteristics of 4G. S300: Construct a factor graph model based on Bayesian networks, while simultaneously... Inertial data is pre-integrated into relative constraints, which serve as inertial factors connecting state nodes at adjacent time points, providing continuous motion prediction. Each of the various distance observations is modeled as a factor of the corresponding category, and the current state node is connected to the fixed node at the known position through the factors; at the same time, a noise model and information matrix are assigned to each factor according to the uncertainty parameters corresponding to each of the various distance observations; the constructed factor graph is solved by a nonlinear optimization method, and the optimal estimate of the state at all times is obtained in one optimization by maximizing the posterior probability, that is, the final motion trajectory. S400: The optimized final motion trajectory is bound to the Wi-Fi fingerprint collected at that time, and the fingerprint database is dynamically updated or expanded for subsequent positioning and to improve the positioning accuracy of other terminals.

2. The method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals according to claim 1, characterized in that, In step S300, the state variables of the factor graph model are the terminal's temporal position, velocity, and attitude.

3. The method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals according to claim 1, characterized in that, In step S100, the terminal device first initiates clock reference initialization to establish a unified time reference, assigning a unique timestamp to all data to be collected; for The acquired triaxial acceleration data underwent gravity separation processing, and the triaxial angular velocity data underwent zero-bias calibration processing, followed by... Using a unified sampling frequency, interpolation or point-sampling simplification processes were performed on Wi-Fi, Bluetooth, and 4G signal data respectively. Linear interpolation was used for interpolation, while point-sampling simplification employed a sliding window-based mean sampling algorithm to ensure all data timestamp intervals were uniformly matched. Collection cycle.

4. The method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals according to claim 1, characterized in that, In step S200, the Wi-Fi distance observation is obtained by comparing the initial distance with the newly sampled distance. Value weighted average correction, corresponding uncertainty parameter Based on corrected distance observations Path loss index and shadow fading coefficient The quantification is performed using the following formula: .

5. The method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals according to claim 1, characterized in that, In step S200, the Bluetooth distance observation value is obtained from the last three samples. The mean is smoothed to obtain the smoothed Bluetooth distance observation. The corresponding uncertain parameters Combining signal intensity fluctuations from nearly 5 sampling points Historical stability coefficient Bluetooth path loss index The quantification is performed using the following formula: .

6. The method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals according to claim 1, characterized in that, In step S200, through parsing The global identifier of the cell used to signal the coverage area of ​​the serving cell / neighboring cell is used as a location constraint. Based on the initial distance of the timing lead conversion, the corrected distance interval is determined. Then through the formula Assign corresponding uncertainty parameters ;in, This is the minimum value within the 4G corrected distance range. This is the maximum value.

7. The method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals according to claim 1, characterized in that, In step S300, the 4G factor is modeled using distance range residuals, and non-zero residuals are generated only when the position estimate exceeds the distance range; the information matrix is ​​obtained by inverting the uncertainty parameters of each factor: the inertial factor information matrix is ​​a diagonal matrix, and the Wi-Fi, Bluetooth and 4G factor information matrices are the reciprocals of the corresponding distance uncertainty parameters; the global objective function is the weighted sum of squared residuals, which is solved iteratively using the Gauss-Newton method to obtain the optimal estimates of position, velocity and attitude at all times in one optimization.

8. The method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals according to claim 1, characterized in that, In step S400, the Wi-Fi fingerprint construction employs anomaly removal, normalization, and confidence evaluation processes, specifically: through... Criteria for removing anomalies If an outlier is found, it will be replaced with the average of the last three values ​​for that access point, and then the valid values ​​will be updated. Value normalized to Finally, the confidence level of the fingerprint-location association pair is calculated within the interval.

9. The method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals according to claim 1, characterized in that, In step S400, the fingerprint database uses a 0.5m grid index, and the maintenance strategy is as follows: when the confidence level is ≥0.7, a new fingerprint is added; when a record already exists, the similarity between the new fingerprint and the existing fingerprint is calculated. If the similarity is ≥0.8, the position and confidence level are updated with weight. If the similarity is <0.8 and the new confidence level is ≥0.7, a redundant fingerprint is added.

10. The method for improving indoor positioning accuracy by integrating Wi-Fi, Bluetooth, and 4G signals according to claim 1, characterized in that, In step S400, the fingerprint database update verification mechanism is established as follows: for every 10 fingerprints updated, 20 records are randomly selected for simulated positioning and the average positioning error is calculated. If the average deviation between the test location and the actual location is <0.3m, the update is valid; otherwise, the confidence threshold for new fingerprints is lowered to 0.5 and the update is re-executed. If the average positioning error is ≥0.3m, the confidence threshold for new fingerprints is lowered to 0.5 and maintained at this threshold until the average error of the next simulated positioning verification is <0.3m. The threshold is then restored to 0.7, and the minimum confidence threshold is lowered to 0.

5. If the error requirement cannot be met after the lowering, the fingerprint database recalibration process is triggered.

Images