A long-term WiFi fingerprint map construction and maintenance method and system
By automatically inferring the AP location and combining spatiotemporal Gaussian process regression and sparse variational Gaussian process for incremental updates, the problems of high labor intensity, poor environmental adaptability and high maintenance cost of existing WiFi fingerprint positioning technology are solved, and high-precision, low-cost long-term stable indoor positioning is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUNAN UNIV
- Filing Date
- 2026-03-11
- Publication Date
- 2026-07-10
AI Technical Summary
Existing WiFi fingerprint positioning technology suffers from problems such as high labor intensity, poor scalability, susceptibility to dynamic environmental changes, and high maintenance costs, making it difficult to achieve long-term stable indoor positioning.
By combining WiFi signal backpropagation mechanism with maximum a posteriori probability estimation and path loss model, the location of access points (APs) can be automatically inferred. Spatiotemporal Gaussian process regression and sparse variational Gaussian process are used for incremental updates to build a long-term WiFi fingerprint map, thereby reducing deployment and maintenance costs and improving anti-interference capability and stability.
It achieves the elimination of manual AP location surveying, reducing deployment costs, resisting dynamic environmental interference, possessing low-cost incremental maintenance capabilities, high and long-term stable positioning accuracy, suitable for various indoor scenarios, reducing positioning error by more than 22%, and its positioning accuracy is superior to existing technologies.
Smart Images
Figure CN122372928A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of indoor positioning technology, specifically to a method for constructing and dynamically maintaining a long-term WiFi fingerprint map, and a system for implementing the method. Background Technology
[0002] Indoor positioning is a core supporting technology for fields such as the Internet of Things (IoT) and smart environments. Among these, WiFi fingerprint positioning has become one of the most widely used indoor positioning solutions due to its combination of accuracy, cost-effectiveness, and reusability with existing infrastructure. However, existing WiFi fingerprint positioning technology has the following key drawbacks that restrict its long-term stable application: 1. Traditional WiFi fingerprint positioning relies on manual on-site surveys to construct received signal strength (RSS) fingerprint maps, which suffers from high labor intensity and poor scalability; 2. Purely data-driven automated map building methods lack physical constraints and are easily affected by dynamic environmental changes (such as object movement and structural modifications), leading to fingerprint map degradation over time and increased spatial ambiguity. 3. The map maintenance method based on the signal propagation model assumes that the location of the access point (AP) is known. However, in actual applications, the AP location metadata is often missing or inaccurate, which leads to ambiguity in the relationship between signal sources and significant map maintenance errors. 4. Existing methods cannot effectively distinguish between temporary signal noise and permanent AP displacement, making it difficult to achieve long-term incremental maintenance of fingerprint maps. Frequent model retraining is required, resulting in high maintenance costs.
[0003] Therefore, there is an urgent need for a long-term WiFi fingerprint map construction and maintenance technology that can automatically infer the physical location of APs, resist dynamic environmental interference, and support incremental updates, in order to solve the stability and maintainability problems of existing technologies. Summary of the Invention
[0004] The purpose of this invention is to provide a method and system for constructing and maintaining a long-term WiFi fingerprint map that can automatically infer the location of access points (APs), resist dynamic environmental changes, and support low-cost incremental updates.
[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A method for constructing and maintaining a long-term WiFi fingerprint map includes the following steps: Step S1: Deploy reference points at preset intervals in the target indoor area, collect WiFi received signal strength RSS data, global coordinates and timestamps of each reference point, and construct an initial fingerprint dataset.
[0006] Step S2: Based on the WiFi signal backpropagation mechanism and maximum a posteriori probability (MAP) estimation, combined with the path loss model, the spatial coordinates of each access point (AP) are automatically inferred. This step eliminates the need for manual AP location surveying, reducing deployment costs.
[0007] Step S3: Decompose the RSS observation of each AP into the mean path loss and residual term based on the AP coordinates obtained in Step S2; model the residual term using spatiotemporal Gaussian process regression (GPR). The spatiotemporal kernel function used in this modeling integrates the spatial radial basis function kernel and the temporal kernel, and optimizes the kernel hyperparameters. By providing a stable reference through the physical path loss model and combining it with spatiotemporal residual learning to capture environmental dynamics, the map's anti-interference ability and long-term stability are significantly improved.
[0008] Step S4: Introduce an inducement point set based on the Sparse Variational Gaussian Process (SVGP), and achieve incremental updates of the fingerprint map by maximizing the lower bound of evidence (ELBO). The number of inducement points in the set... satisfy , This represents the total number of training samples. This incremental update strategy only needs to update local model parameters, without requiring a full retraining, greatly reducing the computational and time costs of long-term maintenance.
[0009] Step S5: Based on the fingerprint map updated in step S4, the optimal location estimate of the terminal is output by minimizing the loss function between the terminal's observed RSS value and the map's predicted RSS value, thereby achieving high-precision and long-term stable indoor positioning.
[0010] Accordingly, the present invention also provides a long-term WiFi fingerprint positioning system, comprising: The data acquisition unit is used to collect WiFi RSS data, inertial measurement unit (IMU) positioning data, and timestamps; The AP coordinate estimation module is used to execute the AP coordinate estimation method. The spatiotemporal learning and maintenance module is used to execute the spatiotemporal residual modeling and incremental map maintenance methods. Fingerprint map storage unit, used to store map data and model parameters; A terminal positioning unit is used to execute the terminal positioning method.
[0011] Compared with existing technologies, the long-term WiFi fingerprint map construction and maintenance method and system provided by the present invention have the following significant advantages: (1) No manual survey of AP location is required: The AP spatial coordinates are automatically inferred through the APLoc module, which solves the problem of existing technologies relying on AP location metadata and reduces deployment costs; (2) Strong resistance to dynamic environmental interference: The physical path loss model provides a stable physical reference, and the spatiotemporal residual learning captures dynamic environmental changes, enabling the fingerprint map to resist interference such as object movement and AP displacement, and has excellent long-term stability. (3) Low incremental maintenance cost: Based on the incremental update strategy of SVGP, only local model parameters need to be updated when new data is introduced, without the need for full retraining, which significantly reduces the computational and time costs of long-term maintenance; (4) High positioning accuracy and long-term stability: It achieves a positioning RMSE of about 3.5m in dynamic environments, which is more than 22% lower than the existing advanced technology. It also maintains stable accuracy in cross-year scenarios. The average AP positioning error is only 2.57m, providing reliable support for fingerprint map calibration. (5) High versatility: It is suitable for various indoor scenarios, requires no special hardware, can reuse existing WiFi infrastructure, is compatible with conventional terminal equipment, and has a wide range of application prospects. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 This is a flowchart illustrating the long-term WiFi fingerprint map construction and maintenance method provided by the present invention.
[0013] Figure 2 The structural block diagram of the long-term WiFi fingerprint map construction and maintenance system provided by the present invention.
[0014] Figure 3 The flowchart is for the long-term WiFi fingerprint map construction and maintenance method provided in Embodiment 1 of the present invention. Detailed Implementation
[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0016] Please see Figure 1 As shown, this invention provides a method for constructing and maintaining a long-term WiFi fingerprint map, including the following steps: Step S1: Deploy reference points at preset intervals in the target indoor area, collect WiFi received signal strength RSS data, global coordinates and timestamps of each reference point, and construct an initial fingerprint dataset. Step S2: Based on the WiFi signal back propagation mechanism and maximum a posteriori probability estimation, combined with the path loss model, automatically infer the spatial coordinates of each access point (AP). Step S3: Decompose the RSS observation of each AP into the mean path loss and residual term based on the AP coordinates obtained in step S2; use spatiotemporal Gaussian process regression to model the residual term, wherein the spatiotemporal kernel function used for modeling integrates the spatial radial basis function kernel and the time kernel, and optimizes the kernel hyperparameter; Step S4: Introducing a set of inducement points based on a sparse variational Gaussian process, and incrementally updating the fingerprint map by maximizing the lower bound of evidence, wherein the number of inducement points... satisfy , This represents the total number of corresponding training samples; Step S5: Based on the fingerprint map updated in step S4, the optimal location estimate of the terminal is output by minimizing the loss function between the terminal's observed RSS value and the map's predicted RSS value.
[0017] In step S2, the preset interval is 50cm.
[0018] The path loss model is expressed as follows: ; in, To predict signal strength; The RSS value at a reference distance of 1 meter; The path loss index; For the first The coordinates of the reference points; For the first The coordinates of each AP; This represents the Euclidean distance between the reference point and AP.
[0019] The optimal estimated coordinates for each AP are estimated by minimizing the following objective function. : ; in, Indicates the total number of reference points; In the first The first reference point observed The RSS value of each AP; The predicted RSS value calculated based on the path loss model; For the first The approximate prior location of an AP; As a priori accuracy parameter, the Determined through k-fold cross-validation.
[0020] In step S3, the spatiotemporal kernel function is: ; in, and These are the coordinates and timestamps of two spatiotemporal points, respectively. It is a spatial length scale; It is a time length scale; Let V be the variance of the spatiotemporal residual signal.
[0021] In step S4, the maximized evidence lower bound ELBO is expressed as: ; in, For the first The streaming mini-batch sample index set corresponding to each AP; For the first The AP in the first The residual target for each sample; For latent functions; To observe the noise variance; Induced variable The variational posterior distribution; Induced variable The prior distribution; This represents the Kullback-Leibler divergence.
[0022] Combined Figure 2 As shown, the present invention also provides a long-term WiFi fingerprint map construction and maintenance system for executing the aforementioned long-term WiFi fingerprint map construction and maintenance method, comprising: Data acquisition unit 1 is configured to acquire WiFi received signal strength RSS data, inertial measurement unit (IMU) positioning data, and corresponding timestamps at the reference point and the terminal.
[0023] AP coordinate estimation module 2 is configured to process the data collected by the data acquisition unit based on the WiFi signal back propagation mechanism and maximum a posteriori probability (MAP) estimation, combined with the path loss model, so as to automatically infer the spatial coordinates of each access point (AP).
[0024] The spatiotemporal learning and maintenance module 3 is configured to decompose the RSS observations of each AP into the mean path loss and residual terms based on the AP coordinates output by the AP coordinate estimation module; model the residual terms using spatiotemporal Gaussian process regression (GPR); and implement incremental updates of the fingerprint map based on sparse variational Gaussian process (SVGP). The number of induced point sets used... satisfy , This represents the total number of corresponding training samples.
[0025] The fingerprint map storage unit 4 is configured to store and update the fingerprint map data and related model parameters constructed and maintained by the spatiotemporal learning and maintenance module.
[0026] Terminal positioning unit 5 is configured to calculate and output the terminal's location estimate based on the latest map in the fingerprint map storage unit by minimizing the loss function between the terminal's observed RSS value and the map's predicted RSS value.
[0027] The AP coordinate estimation module 2 is specifically configured to minimize the objective function. To estimate the optimal coordinates of AP ,in, Indicates the total number of reference points; In the first The first reference point observed The RSS value of each AP; The predicted RSS value calculated based on the path loss model; For the first The approximate prior location of an AP; For prior accuracy parameters, Determined through k-fold cross-validation.
[0028] The spatiotemporal learning and maintenance module 3 uses spatiotemporal Gaussian process regression (GPR) to model the signal residuals, and the spatiotemporal kernel function used is: ; in, and These are the coordinates and timestamps of two spatiotemporal points, respectively. It is a spatial length scale; It is a time length scale; Let V be the variance of the spatiotemporal residual signal.
[0029] The terminal positioning unit 5 is configured to achieve millisecond-level real-time positioning in dynamic environments.
[0030] The present invention will be further described in detail below with reference to Embodiment 1, but the implementation of the present invention is not limited thereto.
[0031] Example 1 Please combine Figure 3 As shown, Example 1 provides a method for constructing and maintaining a long-term WiFi fingerprint map, including the following steps: 1. Data acquisition configuration.
[0032] Deploy access points (APs) compliant with the IEEE 802.11 standard in the target area (e.g., an office area). Use mobile terminals with integrated WiFi and IMU to set up reference points (RPs) at preferred intervals of 50cm. At each RP, collect 10-20 sets of RSS data from all detectable APs, and simultaneously record the global coordinates of the RPs calculated by the IMU or preset. and the corresponding timestamp To form the initial dataset ,in Let J be the RSS vectors of the J APs. This dense sampling strategy provides a sufficient data foundation for subsequent model training.
[0033] 2. Implementation of AP coordinate estimation (APLoc module).
[0034] This step aims to automatically and accurately estimate AP locations, with the benefit of completely eliminating reliance on manual surveys or predictable AP location metadata.
[0035] First, an RSS prediction function is constructed based on the path loss model: ; in, For the first The actual coordinates of each AP; The RSS value at a reference distance of 1 meter; This is the path loss index.
[0036] The RSS observations are modeled as the sum of the predicted values and the shadow fading: ,in Zero-mean Gaussian noise; The standard deviation represents the fading of the shadow.
[0037] Introducing the prior distribution of AP location ,in This is a rough priori location; For prior accuracy; It is an identity matrix.
[0038] The optimal estimated coordinates of AP are obtained by maximum a posteriori probability (MAP) estimation. That is, to minimize the following objective function: ; Prior accuracy parameters From the grid using k-fold cross-validation (k=5) The selection criteria were based on minimizing the root mean square error (RMSE) of the residuals in the validation set. Experiments show that this method can achieve an average AP positioning accuracy of approximately 2.57m, providing reliable physical structure calibration for fingerprint maps.
[0039] 3. Implementation method of spatiotemporal residual modeling and incremental maintenance (PGR-STL module).
[0040] The core of this step is to combine physical laws with data-driven learning. Its beneficial effect is that it can ensure the physical interpretability of the model, flexibly adapt to environmental changes, and achieve long-term stable map maintenance.
[0041] 3.1 Decomposition of Measurement Values: Using the estimated coordinates p̂_j output by the APLoc module, the RSS observations are decomposed into: .in, The path loss prediction based on the estimated AP location is used as a physical guidance term; The residual objective includes shadow fading, multipath effects, and environmental dynamics that the model did not capture.
[0042] 3.2. Spacetime Gaussian Process (GPR) Modeling: Model the residual field yij for each AP.
[0043] The space kernel uses radial basis functions (RBF): ; in, The signal variance of the spatial residual process. For spatially relevant length scales; introduce a Gaussian prior for the AP location. The AP location and kernel hyperparameters are optimized by regularized logarithmic posterior. ,in To observe the noise variance, Determined through k-fold cross-validation; The spacetime kernel is obtained by multiplying the space kernel and the time kernel: ; in, and These are the coordinates and timestamps of two spatiotemporal points, respectively. It is a spatial length scale; It is a time length scale; Let V be the variance of the spatiotemporal residual signal.
[0044] 3.3 Incremental Update Based on SVGP: To handle continuous data flow and reduce computational costs, a Sparse Variational Gaussian Process (SVGP) is employed. A set of induced points is introduced for each AP. The number of induced points satisfy ( (For example, setting the number of historical samples). =50. Define the induced variable. and variational posterior Approximates the true posterior. Incremental learning is performed by maximizing the evidence lower bound (ELBO) with respect to the variational parameters: ; in, This is a mini-batch sample index set for the current streaming data. This strategy enables fast and efficient incremental updates of the fingerprint map by updating the variational parameters based on the current mini-batch data when new data arrives, avoiding global retraining.
[0045] 4. Terminal positioning implementation method.
[0046] During location, based on the maintained and updated fingerprint map: The terminal collects the RSS vector of the current location in real time. .
[0047] Call the pre-trained spatiotemporal GPR model in the PGR-STL module to predict at any candidate location At the current time RSS mean With variance .
[0048] Construct and minimize the following loss function To obtain a location estimate: ; This positioning method makes full use of the signal strength prediction and uncertainty information provided by the map, achieving a positioning accuracy of about 3.5m in dynamic environments, and the response time can reach the millisecond level, meeting the real-time requirements.
[0049] 5. System operation process.
[0050] Initial phase: Collect data and run the APLoc module to obtain the initial coordinates of the AP.
[0051] Training phase: Using the initial dataset, run the PGR-STL module to train the spatiotemporal GPR model and determine the optimal hyperparameters.
[0052] Incremental maintenance phase: New data is collected periodically (e.g., daily or weekly) to trigger the SVGP incremental update process of the PGR-STL module and update the map model.
[0053] Location service phase: The terminal positioning unit responds to the location request in real time and calls the latest map model for calculation.
[0054] The method of this invention is highly versatile, requires no special hardware, can reuse existing WiFi facilities, and has broad application prospects.
[0055] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the claims.
Claims
1. A method for constructing and maintaining a long-term WiFi fingerprint map, characterized in that, Includes the following steps: (1) Set up reference points in the target indoor area at preset intervals, collect WiFi received signal strength RSS data, global coordinates and timestamps of each reference point, and construct an initial fingerprint dataset; (2) Based on the WiFi signal back propagation mechanism and maximum a posteriori probability estimation, combined with the path loss model, the spatial coordinates of each access point (AP) are automatically inferred; (3) Decompose the RSS observation of each AP into the mean path loss and residual term based on the AP coordinates obtained in step (2); use spatiotemporal Gaussian process regression to model the residual term, wherein the spatiotemporal kernel function used for modeling integrates the spatial radial basis function kernel and the time kernel, and optimizes the kernel hyperparameter; (4) An inducement point set is introduced based on a sparse variational Gaussian process, and incremental updates of the fingerprint map are achieved by maximizing the lower bound of evidence. The number of inducement points in the inducement point set is... satisfy , This represents the total number of corresponding training samples; (5) Based on the fingerprint map updated in step (4), the optimal location estimate of the terminal is output by minimizing the loss function between the terminal's observed RSS value and the map's predicted RSS value.
2. The method for constructing and maintaining a long-term WiFi fingerprint map according to claim 1, characterized in that, In step (1), the preset interval is 50cm.
3. The method for constructing and maintaining a long-term WiFi fingerprint map according to claim 1, characterized in that, In step (2), the path loss model is expressed as: ； in, To predict signal strength; The RSS value at a reference distance of 1 meter; The path loss index; For the first The coordinates of the reference points; For the first The coordinates of each AP; This represents the Euclidean distance between the reference point and AP.
4. The method for constructing and maintaining a long-term WiFi fingerprint map according to claim 3, characterized in that, In step (2), the optimal estimated coordinates of each AP are estimated by minimizing the following objective function. : ； in, Indicates the total number of reference points; In the first The first reference point observed The RSS value of each AP; The predicted RSS value calculated based on the path loss model; For the first The approximate prior location of an AP; As a priori accuracy parameter, the Determined through k-fold cross-validation.
5. The method for constructing and maintaining a long-term WiFi fingerprint map according to claim 1, characterized in that, In step (3), the spatiotemporal kernel function is: ； in, and These are the coordinates and timestamps of two spatiotemporal points, respectively. It is a spatial length scale; It is a time length scale; Let V be the variance of the spatiotemporal residual signal.
6. The method for constructing and maintaining a long-term WiFi fingerprint map according to claim 1, characterized in that, In step (4), the maximized evidence lower bound ELBO is expressed as: ； in, For the first The streaming mini-batch sample index set corresponding to each AP; For the first The AP in the first The residual target for each sample; For latent functions; To observe the noise variance; Induced variable The variational posterior distribution; Induced variable The prior distribution; This represents the Kullback-Leibler divergence.
7. A long-term WiFi fingerprint map construction and maintenance system for executing the long-term WiFi fingerprint map construction and maintenance method according to any one of claims 1-6, characterized in that, include: The data acquisition unit is configured to acquire WiFi received signal strength RSS data, inertial measurement unit (IMU) positioning data, and corresponding timestamps at the reference point and the terminal. The AP coordinate estimation module is configured to process the data collected by the data acquisition unit based on the WiFi signal back propagation mechanism and maximum a posteriori probability (MAP) estimation, combined with the path loss model, so as to automatically infer the spatial coordinates of each access point (AP). The spatiotemporal learning and maintenance module is configured to decompose the RSS observation of each AP into the mean path loss and residual term based on the AP coordinates output by the AP coordinate estimation module; The residual terms are modeled using spatiotemporal Gaussian process regression (GPR), and incremental updates of the fingerprint map are achieved based on sparse variational Gaussian process (SVGP). The number of induced points used is... satisfy , This represents the total number of corresponding training samples; The fingerprint map storage unit is configured to store and update fingerprint map data and related model parameters constructed and maintained by the spatiotemporal learning and maintenance module; The terminal positioning unit is configured to calculate and output the terminal's location estimate based on the latest map in the fingerprint map storage unit by minimizing the loss function between the terminal's observed RSS value and the map's predicted RSS value.
8. The long-term WiFi fingerprint map construction and maintenance system according to claim 7, characterized in that, The AP coordinate estimation module is specifically configured to minimize the objective function. To estimate the optimal coordinates of AP ,in, Indicates the total number of reference points; In the first The first reference point observed The RSS value of each AP; The predicted RSS value calculated based on the path loss model; For the first The approximate prior location of an AP; As a priori accuracy parameter, the Determined through k-fold cross-validation.
9. The long-term WiFi fingerprint map construction and maintenance system according to claim 7, characterized in that, The spatiotemporal learning and maintenance module uses spatiotemporal Gaussian process regression (GPR) to model the signal residuals, and the spatiotemporal kernel function used is: ； in, and These are the coordinates and timestamps of two spatiotemporal points, respectively. It is a spatial length scale; It is a time length scale; Let V be the variance of the spatiotemporal residual signal.
10. The long-term WiFi fingerprint map construction and maintenance system according to claim 7, characterized in that, The terminal positioning unit is configured to achieve millisecond-level real-time positioning in dynamic environments.