WiFi positioning precision optimization method suitable for fixed scene

Abstract

The invention discloses a WiFi positioning precision optimization method suitable for a fixed scene. According to the method, firstly, RSSI data are preprocessed through Kalman filtering, and the multipath effect and noise interference are effectively suppressed; and a target coarse position is calculated by combining a reference path loss coefficient and an ITU model, and the initial positioning precision is improved by using a distance reciprocal weighting mechanism. Meanwhile, a dynamic fingerprint database containing an obstacle material and a thickness attenuation model is constructed, and the fingerprint database is ensured to be matched with an actual environment through regular patrol and precision triggering dual-mechanism updating. During positioning, a WKNN algorithm is combined with scene passable region constraints, unreasonable positioning results are eliminated, and coarse positions are fused through adaptive unscented particle filtering. The method solves the problems of fingerprint database outdated, PDR error accumulation and the like caused by multipath effect and environment change of WiFi positioning in a fixed scene, reduces the positioning root mean square error to 0.99 m, greatly improves the precision compared with a traditional method, and is suitable for high-precision positioning requirements of indoor scenes such as shopping malls, office buildings and the like.

Description

Technical Field

[0001] The present invention relates to the fields of signal processing and information fusion, and in particular to a WiFi positioning accuracy optimization method applicable to fixed scenarios. Background Art

[0002] In fixed scenarios, WiFi positioning technology faces multiple technical bottlenecks. On the one hand, indoor building structures such as concrete walls and metal partitions can cause multipath effects and non-line-of-sight propagation of WiFi signals. Measured data show that in complex office building environments, the RSSI signal fluctuation amplitude can reach 10-15dBm, resulting in traditional fingerprint matching errors exceeding 3 meters. On the other hand, dynamic changes in the environment of fixed scenarios (such as store renovations in shopping malls and rearrangement of furniture in office buildings) can quickly cause pre-built WiFi fingerprint libraries to become invalid. Studies have shown that when the positions of 20% of the obstacles in the scene change, the positioning accuracy of the static fingerprint library drops by more than 50%.

[0003] PDR positioning technology is limited by the hardware accuracy of consumer-grade sensors. The zero bias error of the built-in accelerometer in smartphones is usually 0.1-0.5m / s. 2 The gyroscope drift rate is approximately 0.1-1° / s, resulting in step count errors exceeding 5 after walking 100 meters, step length estimation errors of 10-15%, and cumulative positioning errors exceeding 5 meters. Existing WiFi-PDR fusion solutions often use static weighting or traditional Kalman filtering, which cannot adapt to environmental changes. When the WiFi signal is suddenly blocked (such as by a crowd) or the PDR sensor is affected by changes in motion patterns (such as walking up and down stairs), the fusion algorithm has difficulty dynamically adjusting weights, resulting in jumps in positioning results. Literature shows that traditional fusion methods have a root mean square error (RMSE) of over 2.5 meters in complex scenarios, which cannot meet the requirements of high-precision positioning scenarios such as warehousing and logistics, and medical care. Summary of the Invention

[0004] The purpose of the present invention is to provide a WiFi positioning accuracy optimization method applicable to fixed scenarios to solve the existing problems.

[0005] The present invention solves the above technical problems with the following technical solutions: A method for optimizing WiFi positioning accuracy applicable to fixed scenarios, comprising the following steps:

[0006] Step 1: Perform Kalman filter preprocessing on the collected WiFi signal strength (RSSI) data to remove noise and multipath interference;

[0007] Step 2: Match the pre-established location-RSSI fingerprint database to obtain the target's preliminary location. Combine the reference path loss coefficient and the ITU model to calculate the weight coefficient of each AP, and then weight them to obtain the target's rough location.

[0008] Step 3: Collect the heading angle and acceleration values ​​of the target, smooth the original acceleration values, dynamically set the number of steps for calculating the state transition parameters, and calculate the single step length through a nonlinear step length estimation method;

[0009] Step 4: Combine the fixed scene map information and obstacle distribution to build a priori knowledge fingerprint library containing the obstacle attenuation model and set up a dynamic update mechanism;

[0010] Step 5: Using the weighted K-nearest neighbor (WKNN) algorithm, combined with the scene's traversable area constraints, the adaptive unscented particle filter algorithm fuses the coarse position, number of steps, and step length to obtain the precise target location. By integrating multiple technologies and implementing a dynamic optimization mechanism, the system comprehensively addresses issues such as multipath effects and environmental changes. This improves positioning accuracy by 30%-50% compared to traditional methods, and the root mean square error can be controlled within 1.5 meters. This also improves the environmental adaptability and robustness of the positioning system.

[0011] Further solution: The calculation formula of the reference path loss coefficient PL0 is: PL0 = 20log(f) + c(k,f) - 28, where f is the transmission signal frequency, c(k,f) is the ground loss permeability coefficient, and k is the number of floors between the transmitting end and the receiving end.

[0012] Further solution: Calculate the distance between the target and the AP based on the ITU model The formula is: in is the RSSI value of the i-th AP at time t, and α is the path loss index, which ranges from 2.5 to 4.0.

[0013] Further solution: AP weight coefficient The calculation formula is: The formula for calculating the target rough position (x, y) is: Where N is the number of APs, (x i ,y i ) is the initial position.

[0014] Further solution: The dynamic update mechanism includes: Regular patrol update: Detect obstacles or AP position changes every 7 days, and when the area of ​​new or removed obstacles exceeds 10m 2 The update is triggered when there is an obstacle or when the position adjustment of at least three APs exceeds 0.5m. Accuracy trigger update: When the root mean square error (RMSE) of 10 consecutive positioning times exceeds 1.5m, the global fingerprint library is updated. The formula for calculating the root mean square error is: is the real coordinate of the i-th positioning, is the estimated coordinate of the i-th positioning.

[0015] The method according to claim 5 is characterized in that, in the accuracy trigger update, when the RMSE exceeds the threshold, the full-scene RSSI data is re-collected according to a 1m×1m grid, and 20 groups of signal strength values ​​at different time periods are collected at each grid point.

[0016] Further solution: The method for constructing the prior knowledge fingerprint library includes: dividing the scene into 1m×1m grids, measuring the coordinates of each grid point, collecting RSSI data for multiple periods, and performing attenuation correction on the signal strength according to the material and thickness of the obstacle. The correction formula is: where β j is the attenuation coefficient of the jth type of obstacle, l j is the length of the signal passing through the jth type of obstacle.

[0017] Further solution: The implementation method of the passable area constraint of the scene is: based on the electronic map, the obstacle position is marked, the spatial intersection judgment of the positioning result and the obstacle polygon is performed, the coordinates of the non-passable area are excluded, and the height dimension constraint is performed.

[0018] Further solution: The calculation formula of the nonlinear step length estimation method is: Where k is the step size estimation parameter, a max and a min The maximum and minimum acceleration values ​​for the target to take one step.

[0019] Further solution: The fusion calculation steps of the adaptive unscented particle filter algorithm include: establishing a target motion model and state equation, observation equation, iteratively calculating the prior probability density function, likelihood function and posterior probability density function, until the preset number of iterations is reached to extract accurate motion state information.

[0020] The present invention has the following beneficial effects:

[0021] This technical solution uses Kalman filter preprocessing, weighted coarse positioning, and dynamic fingerprint library construction, combined with obstacle attenuation models and scenario constraints, to reduce the root mean square error of positioning from 2.95m in traditional methods to 0.99m, improving accuracy by 66.4%, effectively solving the problems of multipath effect and signal fluctuation.

[0022] The fingerprint library is dynamically updated using a dual mechanism of regular inspections and precision triggering. When the scene changes, the fluctuation of positioning accuracy is reduced by 80%. Combined with adaptive unscented particle filter fusion, PDR error accumulation is eliminated, and long-term continuous positioning error is controlled within 1.5m.

[0023] Double noise reduction processing and traversable area constraints increase positioning efficiency to over 99%. The nonlinear step length estimation reduces the error by 40% compared with traditional models. The overall solution has strong environmental adaptability and outstanding anti-error capabilities, meeting the needs of high-precision indoor positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Flowchart of the WiFi positioning accuracy optimization method.

[0025] Figure 2 Schematic diagram of the dynamic fingerprint library update mechanism. DETAILED DESCRIPTION

[0026] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0027] (1) Signal preprocessing and coarse positioning optimization

[0028] 1. Kalman filter preprocessing

[0029] Implementation process: Set the state equation: x k =Ax k-1 +Bu k +w k , where x k is the signal state vector at time k (including RSSI true value and change rate), A is the state transfer matrix, B is the control matrix, u k is the control vector, w k is the process noise (obeying N(0,Q)).

[0030] Set the observation equation: z k =Hx k +v k , where z k is the RSSI observation value, H is the observation matrix, v k is the observation noise (obeying N(0,R)).

[0031] Recursive computation: by predicting the step and update step K k =P k|k-1 H T (HP k|k-1 H T +R) -1 、 P k|k =(IK k H)P k|k-1 , perform real-time filtering on the RSSI data collected at 10Hz.

[0032] Example: In an office building scenario, the original RSSI data fluctuates from -75dBm to -55dBm. After Kalman filtering, the fluctuation range is reduced to -68dBm to -62dBm, and the noise suppression rate reaches 70%.

[0033] 2. Weighted coarse positioning calculation

[0034] Environmental beacon acquisition: The transmission signal frequency f = 2.4 GHz, and the ground loss permeability coefficient c(k,f) is determined to be 40 based on field measurements (k = 1 for a single-layer scenario). The path loss exponent α is set to 3.5 (medium obstacle density) based on the obstacle density of the scenario.

[0035] Calculation process: Reference path loss coefficient: PL0 = 20log(2400) + 40-28 ≈ 80.8dB. If the real-time RSSI value of AP1 is The distance If there are 5 APs in the scene, the distances between them are The weight coefficient is:

[0036]

[0037] The same logic applies If the preliminary positions are (10,15), (12,14), (9,16), (11,17), and (13,15), then the rough positions are:

[0038] x=0.245×10+0.183×12+0.309×9+0.134×11+0.129×13≈10.78y=0.245×15+0.183×14+0.309×16+0.134×17+0.129×15≈15.52

[0039] That is, the rough position is (10.78, 15.52), with an error of about 1.2m.

[0040] (2) Dynamic fingerprint library construction and update

[0041] 1. Construction of prior knowledge fingerprint library

[0042] Grid division and coordinate measurement: The 200m×100m shopping mall scene was divided into 20,000 grids of 1m×1m. The coordinates of each grid point were measured using a Leica TS60 total station, with a planar accuracy of ±0.05m and an elevation accuracy of ±0.1m.

[0043] Multi-period signal collection: On weekdays from 8:00 to 22:00, RSSI data is collected for 5 minutes at each grid point (sampling frequency 10 Hz) every 2 hours. A total of 7 periods (8:00, 10:00, 12:00, 14:00, 16:00, 18:00, and 20:00) are collected. 300 sets of data are collected in each period, and the average value is taken as the fingerprint value for that period.

[0044] Obstacle attenuation correction: Establish an obstacle attenuation model: concrete wall (thickness 20cm) attenuation coefficient β1 = 8dB / m, glass wall (thickness 10cm) attenuation coefficient β2 = 3dB / m, metal shelf (thickness 50cm) attenuation coefficient β3 = 15dB / m. If a grid point and the AP pass through a 2m concrete wall and a 1m glass wall, the signal correction value is: S ′ =S-(8×2+3×1)=S-19dBm.

[0045] 2. Dynamic update mechanism

[0046] Regular inspection and update: Use FARO Focus S70 LiDAR to scan the scene every week to detect obstacle changes: When a new group of 30m obstacles is found in the clothing area, 2 Metal shelves (more than 10m 2 Threshold), re-collect signals from 500 grid points within 20m around the shelf, and collect 20 sets of data from each point (morning, noon and evening time periods).

[0047] Accuracy trigger update: When the RMSE exceeds 1.5m for 10 consecutive positioning attempts, a global update is triggered: the scene is re-divided into a 1m×1m grid, and coordinates are remeasured using a total station. Twenty sets of data are collected at each grid point (2 sets per hour, 8:00 AM–6:00 PM on weekdays), for a total of 10 periods, to establish a new fingerprint database.

[0048] (3) Correction algorithm and fusion calculation

[0049] 1. Scene constraints and WKNN matching;

[0050] Electronic map modeling: Use AutoCAD to build an electronic map of the shopping mall, mark obstacles such as walls, elevators, and stairs, and represent the passable area as a polygon set with an accuracy of ±0.5m.

[0051] WKNN matching process: The RSSI data collected and filtered in real time is matched with the fingerprint database, the Euclidean distance similarity with each grid point is calculated, and the K=5 nearest neighbor points with the highest similarity are selected.

[0052] Dynamically adjust weights: Consider the impact of obstacles. If there is a concrete wall between the neighboring point and the target, the weight is multiplied by a decay factor of 0.5; if there is no obstacle, the weight is multiplied by 1.2.

[0053] Spatial constraint judgment: If the weighted calculated position is within the wall polygon, then exclude the position and select the suboptimal neighbor point to recalculate until the position within the passable area is obtained.

[0054] 2. Adaptive Unscented Particle Filter Fusion

[0055] Motion model establishment: State equation: x k =[x k-1 +v k-1 cos(θ k-1 )Δt,y k-1 +v k-1 sin(θ k-1 )Δt,v k-1 ,θ k-1 ] T +w k , where (x, y) is the position, v is the velocity, θ is the heading angle, and Δt is the time interval (0.1s).

[0056] Observation equation: y k =[d1,d2,…,d N ,step size ,θ meas ] T +v k , where d i is the AP distance, step size is the step size, θ meas is the measured heading angle.

[0057] Iterative calculation process: Initialization: Set 1000 particles, evenly distributed around the rough position, with weights of 0.001. Prediction: Update particle positions and velocities according to the state equation. Calculate the likelihood function: Where S is the covariance matrix. Update weights: And normalize. Resampling: discard low-weight particles and replicate high-weight particles, keeping the total number of particles unchanged. After 50 iterations, calculate the particle mean as the precise position, and the error can be reduced to within 0.8m.

[0058] Comparison and advantages of existing technologies

[0059] 1) Comparison of technical solutions

[0060]

[0061] (2) Positioning accuracy comparison experimental data

[0062] In a 200m×100m shopping mall scenario, 10 test points were selected, and 50 sets of data were collected at each point to compare the positioning accuracy of the present invention with that of the prior art:

[0063]

[0064]

[0065] (3) Comparison of environmental adaptability

[0066] Type of environmental change Changes in positioning accuracy of existing technologies Positioning accuracy changes of the present invention <![CDATA[New 10m 2 Obstacle]]> RMSE increases by 40%-60% RMSE increases by 5%-10% AP position adjustment 1m RMSE increases by 30%-50% RMSE increases by 8%-15% Changes in population density RMSE fluctuation ±0.8m RMSE fluctuation ±0.2m Long-term continuous positioning (1 hour) The error accumulates to 5-8m The error is controlled within 1.5m

[0067] (IV) Summary of advantages

[0068] Significantly improved accuracy: Compared with existing technologies, the root mean square error of positioning is reduced from 2.95m to 0.99m, and the accuracy is improved by 66.4%, meeting the high-precision requirements of indoor navigation, asset tracking, etc.

[0069] Strong environmental adaptability: The dynamic update mechanism reduces the fluctuation of positioning accuracy by more than 80% in scenarios such as obstacle changes and AP adjustments, and controls the long-term positioning error within 1.5m, while the accumulated error of existing technologies can reach 8m.

[0070] Outstanding anti-interference capability: Dual noise reduction processing and scene constraint mechanism reduce the impact of multipath and crowd interference on positioning by 75%, and the RSSI signal fluctuation suppression rate reaches 70%.

[0071] 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 in the scope of protection of the present invention.

Claims

1. A WiFi positioning accuracy optimization method suitable for fixed scenarios, characterized in that: Step 1: Perform Kalman filter preprocessing on the collected WiFi signal strength (RSSI) data to remove noise and multipath interference; Step 2: Match the pre-established location-RSSI fingerprint database to obtain the target's preliminary location. Combine the reference path loss coefficient and the ITU model to calculate the weight coefficient of each AP, and then weight them to obtain the target's rough location. Step 3: Collect the heading angle and acceleration values ​​of the target, smooth the original acceleration values, dynamically set the number of steps for calculating the state transition parameters, and calculate the single step length through a nonlinear step length estimation method; Step 4: Combine the fixed scene map information and obstacle distribution to build a priori knowledge fingerprint library containing the obstacle attenuation model and set up a dynamic update mechanism; Step 5: Using the weighted K-nearest neighbor (WKNN) algorithm, combined with the scene's traversable area constraints, the adaptive unscented particle filter algorithm is used to fuse the coarse position, number of steps, and step length to obtain the precise position of the target.

2. The WiFi positioning accuracy optimization method applicable to fixed scenarios according to claim 1 is characterized in that: The calculation formula of the reference path loss coefficient PL0 is: PL0 = 20log(f) + c(k, f) - 28, where f is the transmission signal frequency, c(k, f) is the ground loss permeability coefficient, and k is the number of floors between the transmitting end and the receiving end.

3. The WiFi positioning accuracy optimization method applicable to fixed scenes according to claim 1 is characterized in that: Calculate the distance between the target and AP based on the ITU model The formula is: in is the RSSI value of the i-th AP at time t, and α is the path loss index, which ranges from 2.5 to 4.

0.

4. The WiFi positioning accuracy optimization method applicable to fixed scenarios according to claim 1 is characterized in that: AP weight coefficient The calculation formula is: The formula for calculating the target rough position (x, y) is: Where N is the number of APs, (x i ,y i ) is the initial position.

5. The WiFi positioning accuracy optimization method applicable to fixed scenarios according to claim 1, characterized in that: The dynamic update mechanism includes: regular patrol updates: every 7 days, detect obstacles or AP position changes, and when the area of ​​new or removed obstacles exceeds 10m 2 The update is triggered when there is an obstacle or when the position adjustment of at least three APs exceeds 0.5m. Accuracy trigger update: When the root mean square error (RMSE) of 10 consecutive positioning times exceeds 1.5m, the global fingerprint library is updated. The formula for calculating the root mean square error is: n=10, is the real coordinate of the i-th positioning, is the estimated coordinate of the i-th positioning.

6. The WiFi positioning accuracy optimization method applicable to fixed scenarios according to claim 5 is characterized in that: In the accuracy trigger update, when the RMSE exceeds the threshold, the full-scene RSSI data is re-collected according to the 1m×1m grid, and 20 groups of signal strength values ​​at different time periods are collected at each grid point.

7. The WiFi positioning accuracy optimization method applicable to fixed scenarios according to claim 5, characterized in that: The method for constructing the prior knowledge fingerprint library includes: dividing the scene into 1m×1m grids, measuring the coordinates of each grid point, collecting RSSI data for multiple periods, and performing attenuation correction on the signal strength according to the material and thickness of the obstacle. The correction formula is: where β j is the attenuation coefficient of the jth type of obstacle, l j is the length of the signal passing through the jth type of obstacle.

8. The WiFi positioning accuracy optimization method applicable to fixed scenarios according to claim 5, characterized in that: The implementation method of the passable area constraint of the scene is: based on the electronic map, the obstacle position is marked, the positioning result is spatially intersected with the obstacle polygon, the coordinates of the non-passable area are excluded, and the height dimension constraint is performed.

9. The WiFi positioning accuracy optimization method applicable to fixed scenarios according to claim 5, characterized in that: The calculation formula of the nonlinear step length estimation method is: Where k is the step size estimation parameter, a max and a min The maximum and minimum acceleration values ​​for the target to take one step.

10. The WiFi positioning accuracy optimization method applicable to fixed scenes according to claim 1, characterized in that: The fusion calculation steps of the adaptive unscented particle filter algorithm include: establishing a target motion model and state equations, observation equations, iteratively calculating the prior probability density function, likelihood function and posterior probability density function until a preset number of iterations is reached to extract accurate motion state information.

Images