An indoor positioning method of LTE signal fingerprint matching and neighbor observation constraint PDR

Abstract

This invention discloses an indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observations constrained by PDR (Progressive Directional Recognition), belonging to the field of indoor navigation and positioning technology. The method first performs step count detection, step length estimation, and heading estimation based on MEMS sensor output to obtain PDR positioning results. Then, it uses the PDR positioning results and an indoor map to estimate LTE signal propagation model parameters online, quickly constructing a fingerprint database, and obtaining location observations through fingerprint matching. Simultaneously, it utilizes the peak variation pattern of the LTE signal RSRP (Regressive Range Recognition Point) to identify antenna positions using peak detection and threshold constraints, extracting nearest neighbor position, step length, and heading observations. Finally, it employs unscented Kalman filtering to fuse the PDR results and the aforementioned observations in different scenarios. This invention requires no prior data, can adapt to changes in environment and equipment, and combines low cost with high accuracy, making it suitable for indoor pedestrian positioning.

Description

Technical Field

[0001] This invention belongs to the field of indoor navigation and positioning technology, specifically relating to an indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR (Pedestrian Dead Reckoning). Background Technology

[0002] In recent years, with the rapid development of mobile internet and smart terminal technologies, the demand for indoor positioning services has been increasing, and they are widely used in scenarios such as shopping mall guidance, museum navigation, emergency rescue, and social networks. Unlike outdoor environments, indoor environments present challenges such as signal obstruction, complex multipath effects, and diverse spatial structures, making high-precision and low-cost indoor positioning a continued challenge.

[0003] Currently, mainstream indoor positioning methods include Radio Frequency Identification (RFID), Bluetooth, Wi-Fi fingerprint matching, Ultra Wideband (UWB), and pedestrian trajectory estimation. Among these, PDR (Pedestrian Tracking) uses microelectromechanical systems (MEMS) accelerometers, gyroscopes, and magnetometers to detect steps, estimate step length, and calculate heading. It offers advantages such as low equipment cost and no need for external infrastructure deployment. However, PDR suffers from unavoidable error accumulation, making it difficult to maintain high-precision positioning for extended periods when used alone.

[0004] To suppress the cumulative error of PDR, it is often combined with wireless signal positioning technology. Among many wireless signals, Long Term Evolution (LTE) signals have advantages such as a large number of transmission towers, wide signal coverage, strong transmission power, and stable indoor reception. Moreover, parameters such as Reference Signal Receiving Power (RSRP) can be directly obtained using smartphones without the need for additional hardware.

[0005] Existing positioning systems based on LTE signal RSRP mostly employ fingerprint matching. This method typically requires offline collection of signal features from a large number of reference points to construct a fingerprint database. This involves a large workload of field surveys, is time-consuming and labor-intensive, and the applicability of the fingerprint database decreases significantly when equipment is replaced or the environment changes. In addition, some studies have attempted to use methods based on Time of Arrival (TOA), Time Difference of Arrival (TDOA), or Angle of Arrival (AOA), but these methods often require dedicated receiving equipment or precise time synchronization, making them difficult to widely apply on ordinary smartphones.

[0006] Therefore, how to fully utilize the available information of LTE signals to build a highly adaptable fingerprint database without relying on a large amount of prior data, and effectively integrate it with PDR to suppress error accumulation, while ensuring low cost and high accuracy, is a problem that current indoor positioning technology urgently needs to solve. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides an indoor positioning method for PDR constrained by LTE signal fingerprint matching and nearest neighbor observations. It utilizes the peak variation pattern of LTE signals during pedestrian walking to identify antenna positions in real time as nearest neighbor observations. Simultaneously, it dynamically constructs a fingerprint database by estimating propagation model parameters online and fuses multiple observations through unscented Kalman filtering to constrain the cumulative error of PDR.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] An indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR is applied to an indoor pedestrian positioning system. The method uses the UKF method to combine multiple observations extracted through LTE signal fingerprint matching and nearest neighbor methods with PDR to achieve indoor positioning. The method includes the following steps:

[0010] Step 1: Based on the output of the micro-electro-mechanical system (MEMS) accelerometer, gyroscope and magnetometer, the PDR algorithm is used to perform step count detection, step length estimation and heading estimation to obtain the PDR positioning result;

[0011] Step 2: Based on the PDR positioning results of the pedestrian at each step in Step 1, calculate the geometric distance between the pedestrian and the indoor antenna of the LTE signal, select the antenna with the smallest geometric distance as the main signal source for the current step, and determine whether there is a wall between the current position and the main signal source; based on the measured RSRP data of the LTE signal, use the Gauss-Newton method to estimate the parameters of the LTE signal propagation model, including path loss factor, transmit power, and wall penetration loss factor; use the propagation model to predict the RSRP of the test point (TP) in the target area, and realize the online construction of the fingerprint database;

[0012] Step 3: Based on the fingerprint database constructed in Step 2, first use the PDR positioning result of the current step to constrain the matching range, and then use the weighted K-Nearest Neighbors (WKNN) algorithm to match with the sub-fingerprint database in the selected area to obtain the location observation based on the fingerprint.

[0013] Step 4: Utilize the LTE signal RSRP's pattern of increasing and then decreasing as the user approaches and moves away from the antenna, and combine peak detection and threshold constraints to achieve real-time identification of the TP at the antenna; Based on the PDR positioning results of each step of the pedestrian in Step 1, calculate the geometric distance between the current step and all antennas, and select the antenna with the smallest geometric distance. The position of this antenna is the position observation obtained based on the nearest neighbor method; Based on the number of LTE signal base stations passed, further extract the step length parameter and heading observation.

[0014] Step 5: Use the UKF method to fuse the PDR positioning results obtained in Step 1, the fingerprint position observations extracted in Step 2, and the nearest neighbor method position / step size parameters / heading observations obtained in Step 3 to complete indoor pedestrian positioning.

[0015] Furthermore, step two involves online construction of the fingerprint database through main signal source identification, wall penetration detection, and propagation model parameter estimation.

[0016] For each step, this paper first uses the PDR positioning result of that step to calculate the geometric distance between the current location and each indoor LTE signal antenna, and selects the antenna with the smallest geometric distance as its corresponding main signal source. Second, it determines whether there is a wall between the current location and the main signal source. If there is, the current TP is a non-line-of-sight (NLOS) point and is added to the NLOS TP set; otherwise, it is a line-of-sight (LOS) point and is added to the LOS TP set. Then, using the location information of the TP set and the measured LTE signal RSRP data, the Gauss-Newton method is used to estimate the propagation model parameters of the main signal source, including path loss factor, transmit power, and wall penetration loss factor.

[0017] Subsequently, the RSRP of all TPs is predicted using a propagation model to achieve online construction of the fingerprint database. In the initial stage, the number of antennas traversed by the user is small, making it impossible to obtain the optimal model parameters for all antennas. Since the propagation model parameters of neighboring LTE signal sources are similar, the model parameters corresponding to the neighboring antennas that have been passed are assigned to each antenna that has not yet been passed. As the number of antennas passed increases, the propagation model parameters are optimized using the measured RSRP data and positioning results during the walking process, and the fingerprint database is updated.

[0018] Furthermore, step three involves obtaining fingerprint-based location observations through available cell selection and matching:

[0019] In terms of available cell selection, a mobile overlap window is added to the RSRP of all LTE cells that can be received at the current time. The mean and standard deviation of RSRP within the window are calculated. Then, the two standardized variables are weighted and averaged and sorted. The cells with the largest values ​​are selected as available cells.

[0020] In terms of matching and localization, a matching area within a certain range is selected based on the PDR localization result of the current step to achieve coarse localization; then, the WKNN algorithm is used to match with the sub-fingerprint database within the selected area to achieve precise localization, thereby obtaining fingerprint-based location observations.

[0021] Furthermore, step four combines peak detection and threshold constraint to achieve real-time identification of TP at the antenna, thereby obtaining observations based on the nearest neighbor method:

[0022] First, the raw RSRP data of the LTE signal contains noise. A two-stage filter, which includes mean filtering and median filtering, is used to filter out the noise and obtain the effective information.

[0023] Secondly, measured data is collected near the antenna to obtain the RSRP attenuation pattern template. Taking advantage of the change trend of the LTE signal RSRP first increasing and then decreasing as the user approaches and moves away from the antenna, peak detection is performed by adding an overlapping window to the filtered RSRP data and calculating the Pearson correlation coefficient between the RSRP sequence in the window and the attenuation pattern template. The correlation coefficient needs to be greater than the set threshold.

[0024] Then, the RSRP range within the overlapping window is calculated and compared with the set threshold to perform threshold constraint; the range must be greater than the set threshold.

[0025] If both peak detection and threshold constraints are met, the current step is considered to be at a certain antenna. The geometric distance is calculated using the PDR positioning result of this step and the positions of all antennas. The antenna with the smallest geometric distance is selected and its position is used as the position observation obtained based on the nearest neighbor method. If multiple antennas have been passed and the current antenna is in a straight-line walking phase with its adjacent antennas, the step size parameter and heading observation are further extracted to obtain multiple observations based on the LTE signal nearest neighbor method.

[0026] Furthermore, step five uses the UKF method to combine various observations of the PDR and LTE signals:

[0027] The state variables of the UKF method are the two-dimensional position, step size model coefficients, and heading angle obtained from PDR, and the state variable dimension is 4-dimensional.

[0028] Based on the straight-line / turn recognition results and the nearest neighbor detection results, the observations are obtained in different cases: If no antenna has been passed, the observation is the two-dimensional position provided by fingerprint matching positioning; if the antenna has been passed for the first time or there is a turning phase between the current antenna and the adjacent antenna that has already been passed, the observations include the two-dimensional position provided by fingerprint matching positioning and the two-dimensional position provided by the nearest neighbor method, and the two types of observations are combined through sequential processing; if an antenna has been passed and there is a straight-line phase between the current antenna and the adjacent antenna that has already been passed, the observations include the two-dimensional position provided by fingerprint matching positioning, as well as the two-dimensional position, step size coefficient, and heading provided by the nearest neighbor method, and the two types of observations are combined through sequential processing to finally obtain the positioning result.

[0029] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned indoor positioning method of LTE signal fingerprint matching and nearest neighbor observation constraint PDR.

[0030] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned indoor positioning method of LTE signal fingerprint matching and nearest neighbor observation constraint PDR.

[0031] The beneficial effects of this invention are as follows:

[0032] No prior data is required, and the system is highly adaptable to different environments. This invention utilizes PDR positioning results and indoor maps to estimate LTE signal propagation model parameters online and dynamically construct a fingerprint database. This avoids the problems of large offline data collection workload and database failure after environmental changes in traditional fingerprint positioning methods, significantly improving the system's practicality and robustness.

[0033] Multi-observation fusion effectively suppresses PDR cumulative error. This invention not only obtains location observations through fingerprint matching, but also utilizes the LTE signal RSRP's pattern of first increasing and then decreasing during the user's approach-away from the antenna process. Combined with peak detection and threshold constraints, it identifies the antenna position in real time, and then extracts position, step size, and heading observations. By constraining PDR error from multiple dimensions, it effectively solves the problem of single PDR positioning error accumulating over time.

[0034] Low cost, high precision, and easy to promote. This invention can be implemented using only commercial devices such as smartphones, without the need for additional dedicated base stations or hardware. It ensures high-precision positioning while meeting the requirement of low cost, and is suitable for various indoor scenarios such as shopping malls, museums, and underground parking lots. Attached Figure Description

[0035] Figure 1This is a flowchart illustrating the implementation of an indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR according to the present invention.

[0036] Figure 2 This is a schematic diagram illustrating the online construction of the fingerprint database according to the present invention;

[0037] Figure 3 This is a schematic diagram of the nearest neighbor method of the present invention. Detailed Implementation

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0039] like Figure 1 As shown, this invention provides an indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation-constrained PDR. It uses unscented Kalman filtering (UKF) to combine multiple observations extracted through LTE signal fingerprint matching and nearest neighbor methods with PDR to achieve indoor positioning. The method includes the following steps:

[0040] Step 1: Based on the output of the microelectromechanical system (MEMS) accelerometer, gyroscope and magnetometer, the pedestrian trajectory estimation (PDR) algorithm is used to detect the number of steps, estimate the step length and the heading, and obtain the PDR positioning result.

[0041] Step 2: Based on the PDR positioning results of the pedestrian at each step in Step 1, calculate the geometric distance between the pedestrian and the indoor antenna of the LTE signal, select the antenna with the smallest geometric distance as the main signal source for the current step, and determine whether there is a wall between the current location and the main signal source; based on the measured data of the reference received power (RSRP) of the LTE signal, use the Gauss-Newton method to estimate the parameters of the LTE signal propagation model, including the path loss factor, transmit power, and wall penetration loss factor; use the propagation model to predict the RSRP of the test point (TP) in the target area, and realize the online construction of the fingerprint database;

[0042] Step 3: Based on the fingerprint database constructed in Step 2, first use the PDR positioning result of the current step to constrain the matching range, and then use the weighted K-nearest neighbor (WKNN) algorithm to match with the sub-fingerprint database in the selected area to obtain the location observation based on the fingerprint.

[0043] Step 4: Utilize the LTE signal RSRP's pattern of increasing and then decreasing as the user approaches and moves away from the antenna, and combine peak detection and threshold constraints to achieve real-time identification of the TP at the antenna; Based on the PDR positioning results of each step of the pedestrian in Step 1, calculate the geometric distance between the current step and all antennas, and select the antenna with the smallest geometric distance. The position of this antenna is the position observation obtained based on the nearest neighbor method; Based on the number of LTE signal base stations passed, further extract the step length parameter and heading observation.

[0044] Step 5: The UKF method is used again to fuse the PDR positioning results obtained in Step 1, the fingerprint position observations extracted in Step 3, and the nearest neighbor method position / step size parameters / heading observations obtained in Step 4 to complete the indoor pedestrian positioning.

[0045] Step one specifically involves: based on the output of MEMS accelerometer, gyroscope and magnetometer, noise reduction and filtering are first performed, and then the PDR positioning result is obtained through step count detection, step length estimation and heading estimation.

[0046] The data input for step counting is triaxial acceleration. The principle is that during walking, the acceleration exhibits periodic peaks and troughs. This periodicity is reflected in the periodic peaks of acceleration. Therefore, step counting can be achieved by detecting these acceleration peaks. Here, acceleration amplitudes with gravitational acceleration removed are used. The acceleration peaks start at 0, rise, and then return to 0. First, zero-crossing detection is used to initially locate the acceleration peaks. Second, the duration of the acceleration peak must meet certain conditions to be considered a valid peak. Using these methods, the various acceleration peaks can be identified, thus detecting the number of steps.

[0047] The data input for step size estimation is triaxial acceleration, implemented using the Weinberg formula based on the range of acceleration magnitudes:

[0048] ,

[0049] in, For the first Step length, These are the step-size model coefficients. and These represent the maximum and minimum values ​​of the z-axis acceleration of the navigation system within one step.

[0050] The data inputs for heading estimation are three-axis acceleration, three-axis angular velocity, and three-axis magnetic field strength vectors. A UKF model is designed to combine acceleration, angular velocity, and magnetic field strength vectors to estimate the heading, and observations are selected according to different cases to constrain the cumulative heading error: For the straight-line phase, the pedestrian heading angle is almost stable and constant, so the straight-line characteristic can be used to constrain the heading error; for the stationary phase, the three-axis angular velocity is approximately assumed to be 0, and the three-axis output of the gyroscope is an offset, so zero angular velocity correction is used to constrain the heading; for the non-straight and non-stationary phase, acceleration and magnetic field strength are used as observations to constrain the error.

[0051] like Figure 2As shown, step two specifically involves: online construction of the fingerprint database through main signal source identification, wall penetration detection, and propagation model parameter estimation. Each cell's LTE signal may originate from multiple antennas; therefore, it is necessary to identify the antenna that has the primary impact on the current TP point, i.e., the main signal source. For each step, firstly, the geometric distance between the current TP and each indoor LTE signal antenna is calculated using the PDR positioning result of this step, and the antenna with the smallest geometric distance is selected as its corresponding main signal source. Secondly, it is determined whether there is a wall between the current location and the main signal source. If there is, the current TP is an NLOS point and is added to the NLOSTTP set; otherwise, it is a LOS point and is added to the LOS TP set. Then, using the location information of the TP set and the measured LTE signal RSRP data, the propagation model parameters of the main signal source, including path loss factor, transmit power, and wall penetration loss factor, are estimated using the Gauss-Newton method. The average wall model (AWM) is selected as the propagation model, as shown in the following equation:

[0052] ,

[0053] in, , These are the RSRP values ​​at the TP point and the antenna, respectively, in dB. The distance between TP point and antenna. This is the path loss factor. For the first The penetration loss of an obstacle is expressed in dB. This represents the number of obstacles.

[0054] The Gauss-Newton method is used to estimate the model parameters. Parameter estimation requires solving the following problem:

[0055] ,

[0056] in, This refers to the number of TPs with each antenna as the primary signal source. For TP two-dimensional position, For the two-dimensional position of the antenna, Let RSRP represent the TP point where the i-th antenna is the main signal source.

[0057] The error is defined as:

[0058] ,

[0059] For LOS TP, the estimated variable include Jacobian matrix For NLOSTP, the estimated variable is... for Jacobian matrix The incremental equation for Gauss-Newton's method is:

[0060] ,

[0061] in, , , Given the initial values ​​of the parameters to be estimated, the following can be obtained: ,like If it is small enough, stop; otherwise, let Repeat the above process until the condition is met.

[0062] Subsequently, the RSRP of all TPs is predicted using a propagation model to achieve online construction of the fingerprint database. In the initial stage, the number of antennas traversed by the user is small, making it impossible to obtain the optimal model parameters for all antennas. Since the propagation model parameters of neighboring LTE signal sources are similar, the model parameters corresponding to the neighboring antennas that have been passed are assigned to each antenna that has not yet been passed. As the number of antennas passed increases, the propagation model parameters are optimized using the measured RSRP data and positioning results during the walking process, and the fingerprint database is updated.

[0063] Step three specifically involves: based on the fingerprint database constructed in step two, firstly, using the PDR positioning result from the current step to constrain the matching range, and then using the weighted K-nearest neighbor (WKNN) algorithm to match it with the sub-fingerprint database within the selected area to obtain fingerprint-based location observations. The WKNN algorithm is as follows:

[0064] / ,

[0065] Where, in the formula, This represents the obtained fingerprint-based location observations; It is the nearest neighbor. The coordinates of the reference points; This indicates that the real-time measured RSRP value is compared with the value in the fingerprint database constructed using step two. RSRP of a reference point Euclidean distance between them To estimate the RSRP using AWM, the AWM model parameters are... Obtained using the Gauss-Newton method in step two; These are the normalized weighted coefficients; It is a small positive constant defined as a weighting coefficient to avoid zero in the denominator.

[0066] The available cells for LTE signals vary across different areas, and fingerprint ambiguity is common in areas with similar building structures. To address these issues, this invention selects available cells in real time and employs a two-step method involving coarse and precise positioning stages to obtain fingerprint matching positioning results. For available cell selection, a moving overlap window is applied to the RSRP of the LTE signals from all currently receivable cells. The mean and standard deviation of the RSRP within the window are calculated, and then the two standardized variables are weighted, ranked, and the cells with the highest values ​​are selected as available cells. For fingerprint matching positioning, a matching area within a certain range is selected based on the fusion positioning result of the current step to achieve coarse positioning. Based on this, the WKNN algorithm is used to match the fingerprints with a sub-fingerprint database within the selected area to achieve precise positioning. Limiting the matching area effectively improves matching efficiency while reducing the impact of fingerprint ambiguity on matching positioning accuracy.

[0067] Step four specifically involves: (e.g.) Figure 3 As shown, the LTE signal RSRP first increases and then decreases as the user approaches and moves away from the antenna. Combined with peak detection and threshold constraints, real-time identification of the TP at the antenna is achieved. Based on the PDR positioning results of the pedestrian at each step in step one, the geometric distance between the current step and all antennas is calculated and the antenna with the smallest geometric distance is selected. The position of this antenna is the position observation obtained based on the nearest neighbor method. According to the number of LTE signal base stations passed, the step length parameter and heading observation are further extracted.

[0068] Because the original RSRP data contains noise, a two-stage filter consisting of mean filtering and median filtering is used to remove the noise. Testing time, equipment, and environment are all key factors affecting RSRP. To improve the adaptability of the proposed algorithm, this invention does not use absolute RSRP values, but instead evaluates the similarity between the RSRP sequence and a pre-extracted RSRP attenuation pattern to achieve peak detection, thereby identifying the TP point at the antenna. The RSRP attenuation pattern template is extracted from measured data collected near the antenna.

[0069] An overlapping window is applied to the filtered RSRP data, and the Pearson correlation coefficient between the in-window RSRP sequence of the selected cell and the attenuation pattern template is calculated. Peak detection is performed, and the correlation coefficient is calculated as follows:

[0070] ,

[0071] in, It is the covariance between the measured RSRP sequence and the RSRP template sequence. and These are the variances of the two sequences, respectively.

[0072] If the correlation coefficient condition is met, that is... , This represents the correlation coefficient threshold. To ensure detection accuracy, a threshold constraint is added, calculated by setting the intra-window RSRP range corresponding to this TP. and the set range threshold Compare the results; if both conditions are met, then... If so, then the current step is considered to be located at a certain antenna.

[0073] Since there may be multiple antennas in the scene, it is necessary to further determine which antenna the current TP is located at. The location information of the current TP is obtained using the PDR positioning results. The geometric distance between the TP and all antennas is calculated, and the antenna with the smallest geometric distance is selected. Using the location information of this antenna, a location observation based on the nearest neighbor method can be obtained. Furthermore, if multiple antennas have already been traversed and the current traversed antenna is in a straight-line walking phase with its adjacent antennas, the step size parameter and heading observation are further extracted to obtain a multi-observation measurement based on the LTE signal nearest neighbor method.

[0074] Step five specifically involves using the UKF method to fuse the PDR positioning results obtained in step one, the fingerprint location observations extracted in step three, and the nearest neighbor method location / step size parameters / heading observations obtained in step four to complete indoor pedestrian positioning.

[0075] The two-dimensional position, step size model coefficients, and heading angle are used as state variables:

[0076] ,

[0077] in, , and The first The two-dimensional position, step size model coefficients, and heading are obtained using PDR.

[0078] The state transition equation is as follows:

[0079] ,

[0080] in, For the calculation using PDR The change in heading within a step.

[0081] The observations are obtained based on the straight / turn recognition results and the nearest neighbor detection results, with the straight / turn motion pattern recognition based on the PDR heading. Specifically, a moving overlapping window is added to the PDR heading angle and the difference between the beginning and end of the heading angle within the window is calculated, as shown in the following formula. If the difference between the beginning and end of the heading angle is greater than a threshold, it is considered to be a turning phase; otherwise, it is considered to be a straight-line walking phase.

[0082] ,

[0083] in, , These are the heading angles at the rear and front of the window, respectively. The threshold value set.

[0084] Scenario 1: The antenna is not currently in use, and the observed measurement is a two-dimensional location provided by fingerprint matching and positioning.

[0085] ,

[0086] The observation equation is:

[0087] ,

[0088] in, for Observation noise for LTE signal fingerprint matching at any given time.

[0089] Case 2: There is a turning and walking phase between the current antenna and the first antenna passed or the adjacent antenna that has been passed. The observations include the two-dimensional position provided by fingerprint matching and the two-dimensional position provided by the nearest neighbor method. This paper combines the two types of observations through sequential processing.

[0090] First, the nearest neighbor method provides the following observations:

[0091] ,

[0092] in, This represents the two-dimensional location of the currently passed base station.

[0093] The observation equation is:

[0094] ,

[0095] in, for Observation noise of LTE signal nearest neighbor location observations at any given time.

[0096] The updated state variables can be obtained using the above formula. Then, As a prediction, it is fused with the fingerprint matching location observation to obtain the final updated state.

[0097] Scenario 3: The antenna has passed, and the current antenna is in a straight-line walking phase. In this case, the observations include the two-dimensional position provided by fingerprint matching positioning, and the two-dimensional position, step size coefficient, and heading provided by the nearest neighbor method. Similarly, this paper combines the two types of observations through sequential processing.

[0098] First, the nearest neighbor method provides the following observations:

[0099] ,

[0100] They are obtained from the following formulas:

[0101] ,

[0102] ,

[0103] in, , These are the two-dimensional positions of the currently passed antenna and the previously passed antenna, respectively. This represents the number of steps taken between the last antenna visited and the currently visited antenna.

[0104] The observation equation is:

[0105] ,

[0106] in, for Observation noise of LTE signal nearest neighbor method position / step / heading observations at any given time.

[0107] The updated state variables can be obtained using the above formula. Then, As a prediction, it is fused with the fingerprint matching location observation to obtain the final updated state.

[0108] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned indoor positioning method of LTE signal fingerprint matching and nearest neighbor observation constraint PDR.

[0109] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned indoor positioning method of LTE signal fingerprint matching and nearest neighbor observation constraint PDR.

[0110] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are 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. An indoor positioning method using LTE signal fingerprint matching and nearest neighbor observation constraint PDR, characterized in that, include: Step 1: Based on the outputs of the microelectromechanical system accelerometer, gyroscope and magnetometer, the pedestrian trajectory estimation algorithm (PDR) is used to obtain the pedestrian trajectory estimation and positioning results; Step 2: Based on the pedestrian trajectory estimation and positioning results, select the indoor LTE signal antenna with the smallest geometric distance as the main signal source. Combined with wall penetration detection, use the Gauss-Newton method to estimate the propagation model parameters including path loss factor, transmit power and wall penetration loss factor. Use the propagation model parameters to predict the reference signal received power at the test point, so as to realize the online construction of fingerprint database without prior data. Step 3: Based on the online fingerprint database, use the pedestrian trajectory to calculate the positioning results to constrain the matching range, and obtain the location observation based on fingerprint matching through weighted K-nearest neighbor algorithm matching; Step 4: Utilize the variation law of the received power of the LTE signal reference signal increasing and then decreasing as the user passes through the antenna, and combine peak detection and threshold constraint to identify the antenna position, and take the position of the antenna with the closest geometric distance as the position observation based on the nearest neighbor method. Furthermore, when multiple antennas have been passed and the current antenna is in a straight-line walking phase with the adjacent antenna, the step length parameter is extracted using the geometric distance between the two antennas and the number of walking steps, and the heading observation is extracted using the direction of the line connecting the two antennas. Step 5: Using the unscented Kalman filter method, the pedestrian trajectory estimation and positioning results are used as state variables. Observations are adaptively selected based on three cases: whether the pedestrian has passed an antenna and whether the pedestrian is in a straight-line walking phase. The selected observations are then fused through a sequential processing method. The selected observations include: the fingerprint-based position observation, the nearest neighbor-based position observation, step size parameters, and heading observations. The step size parameters and heading observations are only selected and used during the straight-line walking phase.

2. The indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR according to claim 1, characterized in that, Step 2 specifically includes: For each step, the pedestrian trajectory is used to calculate the positioning result of the step and calculate the geometric distance between the pedestrian and each LTE signal indoor antenna. The antenna with the smallest geometric distance is selected as the corresponding main signal source. Determine whether there is a wall between the current location and the main signal source. If there is, add the current test point to the non-line-of-sight test point set; otherwise, add it to the line-of-sight test point set. Using the location information of the test point set and the measured reference signal received power data, the path loss factor, transmit power and wall penetration loss factor of the main signal source are estimated by the Gauss-Newton method. In the initial stage, the model parameters of the neighboring antennas that have been passed are assigned to the antennas that have not yet been passed. As the number of passed antennas increases, the propagation model parameters are optimized using the measured reference signal received power data and positioning results during the walking process, and the fingerprint database is updated.

3. The indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR according to claim 1, characterized in that, Step 3 specifically includes: For the reference signal received power of all LTE cells that can be received at the current time, a mobile overlap window is added, the mean and standard deviation of the reference signal received power within the window are calculated, the standardized mean and standard deviation are weighted and averaged, and the cells with the largest weighted average are selected as available cells. Based on the pedestrian trajectory of the current step, the positioning result is calculated, and a matching area within a certain range is selected to achieve rough positioning; Within the selected matching area, the weighted K-nearest neighbor algorithm is used to match with the sub-fingerprint database to obtain location observations based on fingerprint matching; The weighting coefficients of the weighted K-nearest neighbor algorithm are determined based on the Euclidean distance between the real-time measured reference signal received power and the predicted reference signal received power in the fingerprint database.

4. The indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR according to claim 1, characterized in that, Step 4, identifying the antenna location, specifically includes: Noise in the raw data of the received power of the reference signal is filtered out using a two-stage filter that includes mean filtering and median filtering. Based on the measured data collected near the antenna, a reference signal received power attenuation pattern template is extracted. An overlapping window is added to the filtered reference signal received power data, and the Pearson correlation coefficient between the reference signal received power sequence within the window and the attenuation pattern template is calculated. Calculate the range of the received power of the reference signal within the overlapping window; If the conditions of correlation coefficient being greater than a preset threshold and range being greater than their respective preset thresholds are met simultaneously, then the current step is considered to be located at the antenna.

5. The indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR according to claim 1, characterized in that, The three scenarios in step 5 are as follows: Scenario 1: When the antenna is not currently in use, the selected observations are only location observations based on fingerprint matching; Scenario 2: When passing an antenna for the first time, or when there is a turning phase between the current antenna and an adjacent antenna that has already been passed, the selected observations include location observations based on fingerprint matching and location observations based on the nearest neighbor method; Scenario 3: When the current antenna is in a straight-line walking phase after passing an antenna and the adjacent antennas that have been passed are gone, the selected observations include position observations based on fingerprint matching, position observations based on the nearest neighbor method, step size parameters, and heading observations.

6. The indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR according to claim 5, characterized in that, The identification of the turning and straight walking phases is based on the heading angle calculated from the pedestrian's trajectory: a moving overlapping window is added to the heading angle calculated from the pedestrian's trajectory, and the difference between the beginning and end of the heading angle within the window is calculated. If the difference between the beginning and end is greater than a preset threshold, it is identified as a turning phase; otherwise, it is identified as a straight walking phase.

7. The indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR according to claim 1, characterized in that, In step 4, the step size parameter is extracted in the following way: The geometric distance between the previously passed antenna and the currently passed antenna is divided by the total number of steps taken between the previously passed antenna and the currently passed antenna. The total number of steps is determined based on the gait detection results.

8. The indoor positioning method based on LTE signal fingerprint matching and nearest neighbor observation constraint PDR according to claim 1, characterized in that, In step 4, the heading observation is extracted in the following way: calculate the direction angle of the line connecting the previously passed antenna and the currently passed antenna, and use this direction angle as the heading observation.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the indoor positioning method of LTE signal fingerprint matching and nearest neighbor observation constraint PDR as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the indoor positioning method of LTE signal fingerprint matching and nearest neighbor observation constraint PDR as described in any one of claims 1-8.

Images