A multi-modal solution method for mobile target position in dynamic radiation environment

Abstract

The application discloses a kind of multi-modal solution methods of mobile target position under dynamic radiation environment, belong to indoor positioning technical field.The application first collects various positioning and environmental data, compensates the phase lag caused by humidity by means of data-driven method, calculates phase position reliability factor relying on channel waveform similarity, adaptively adjusts filter measurement noise matrix, then realizes PDR step closed-loop calibration and timing smoothing update by using sliding window combined with high-precision positioning result, and finally completes two types of positioning information fusion output position result by extended Kalman filter.The application effectively solves the problems of systematic ranging deviation caused by humidity, low multi-path interference recognition accuracy, fixed filter weight and easy accumulation of pedestrian dead reckoning error in the prior art, can significantly improve the positioning accuracy, anti-interference ability and long-time running stability in complex indoor scene, has wide application range and high engineering practical value.

Description

Technical Field

[0001] This invention relates to a multimodal method for calculating the position of a moving target under dynamic radiation conditions, belonging to the field of indoor positioning technology. Background Technology

[0002] With the widespread adoption of the Internet of Things (IoT) and smart terminals, indoor location services are playing an increasingly crucial role in smart city operation monitoring and industrial internet scheduling. Bluetooth Low Energy (BLE), with its advantages of low power consumption, ease of deployment, and high device compatibility, has become the mainstream technology supporting indoor micro-positioning. BLE positioning technology mainly includes two mainstream schemes: fingerprint matching positioning based on Received Signal Strength Indicator (RSSI) and phase-based ranging (PBR) positioning based on Channel Sounding (CS). Fingerprint matching positioning achieves positioning by establishing a mapping relationship between signal characteristics and physical location, while PBR technology achieves centimeter-level ranging accuracy by measuring the phase difference of radio frequency signals.

[0003] The mainstream fusion architecture for indoor positioning is the BLE and PDR fusion positioning method based on EKF, such as the BtPDR system proposed by Yao et al. ([7]YAO Y, BAO Q, HAN Q, et al. BtPDR: Bluetooth and PDR-based indoor fusion localization using smartphones[J]. KSII Transactions on Internet and Information Systems, 2018, 12(8): 3657-3682.), which uses the inertial sensor (IMU) of the smart terminal to perform traditional pedestrian dead reckoning (PDR) to obtain continuous relative position prediction; at the same time, it uses the ranging information (such as RSSI or basic phase ranging) provided by BLE base stations in the area as external measurement, and realizes the open-loop fusion of the two through the standard EKF filtering framework. Guo et al. attempted to use residual chi-square test mechanism or macroscopic signal characteristics (such as signal-to-noise ratio SNR) to identify abnormal ranging values ​​under multipath interference, and tried to dynamically adjust the weight of abnormal observations through robust EKF in order to improve the robustness of the positioning system in occluded environments (Guo S, Zhang Y, Gui X, et al. An Improved PDR / UWB Integrated System for Indoor Navigation Applications[J].IEEE Sensors Journal, 2020, PP(99):1-1.DOI:10.1109 / JSEN.2020.2981635.).

[0004] However, existing solutions typically treat BLE phase ranging (PBR) as a standard geometric distance observation. In real-world dynamic radiation environments, fluctuations in medium conditions such as air humidity significantly alter the relative permittivity of the transmission medium, leading to electromagnetic wave propagation velocity attenuation and introducing additional phase lag at the physical level, independent of geometric distance. Existing solutions lack dynamic compensation mechanisms for such environmental radiation factors, resulting in severe systematic biases in the underlying ranging sources. While Guo et al.'s solution introduces robust filtering, its multipath identification often relies on macroscopic residual checks, exhibiting lag and failing to extract features from the underlying physical communication link. Furthermore, existing EKF fusion methods typically use a fixed measurement noise covariance matrix, which cannot accurately and in real-time respond to the time-varying characteristics of BLE ranging quality. When the signal suffers severe and sudden multipath fading, the system is highly susceptible to relying on contaminated ranging gross errors, leading to distortion of the positioning trajectory. Moreover, the PDR modules in existing fusion solutions mostly rely on fixed empirical models (such as the Weinberg model with fixed parameters) to estimate the step size. Due to the lack of high-precision absolute spatial constraints in reverse calibration, its step size parameter cannot adapt to the dynamic changes in pedestrian walking speed and gait patterns. This open-loop calculation mechanism causes the local error of PDR to accumulate monotonically with long-distance walking, ultimately dragging down the global accuracy of the entire fusion system. Summary of the Invention

[0005] To simultaneously improve the accuracy, robustness, and continuity of positioning in dynamic radiation environments, this invention provides a multimodal method for calculating the position of a moving target in dynamic radiation environments, including: Step 1: Construct a positioning parameter system and obtain BLE base station coordinates, pedestrian stride length, heading angle, PBR ranging value, and ambient relative humidity data; Step 2: Use a data-driven nonlinear phase dynamic compensation mechanism to correct the PBR phase lag caused by humidity and obtain the compensated ranging observation value; The nonlinear phase dynamic compensation mechanism includes: establishing a nonlinear mapping model between humidity and phase lag in the offline stage, collecting environmental humidity data in real time in the online stage, calculating the phase compensation amount based on the mapping model, and correcting the original PBR phase observation value. Step 3: Calculate the phase position confidence factor based on the similarity of the channel impulse response first path waveform; Step 4: Based on the phase position confidence factor, adaptively adjust the EKF measurement noise covariance matrix, as shown below:

[0006] in, Indicates the first i The base station in the t Phase position confidence factor at time, This represents the baseline variance of PBR ranging under ideal line-of-sight conditions. Step 5: Within the preset sliding window, calculate the actual physical translation distance within the window based on the posterior coordinates output by EKF, and simultaneously extract the sum of the range feature terms of each gait acceleration resolved by the IMU within the window; based on the spatial geometric equivalence relationship, back-calculate the instantaneous optimal PDR step length estimate of the current gait interval, and perform time-series smoothing update on the PDR step length estimate. Step 6: Combining the measurement noise covariance matrix adjusted in Step 4 and the PDR step size parameters corrected in Step 5, the EKF framework is used to perform PBR / PDR fusion localization.

[0007] Thus, to address the problem of systematic humidity phase lag in PBR under dynamic radiation environments, this invention employs a data-driven nonlinear dynamic phase compensation mechanism, which constructs humidity data offline. The phase lag mapping model and online real-time correction of phase observations effectively eliminate systematic ranging biases caused by changes in the dielectric constant of the medium due to air humidity, improving the accuracy and stability of PBR raw observations. This invention calculates the phase position confidence factor based on CIR first-path waveform similarity and adaptively adjusts the EKF measurement noise covariance matrix to achieve fine-grained physical layer channel quality assessment. It automatically reduces the weight of high-interference, non-line-of-sight ranging measurements, avoiding filter contamination by abnormal observations and significantly improving multipath and non-line-of-sight resistance capabilities. Within a sliding window, this invention uses EKF posterior coordinates to back-calculate the instantaneous optimal step size and performs time-series smooth updates to the step size parameters, achieving online adaptive calibration of the step size parameters, suppressing long-term cumulative drift in PDR, and improving long-endurance positioning continuity and global accuracy.

[0008] This invention utilizes phase compensation to provide a precise benchmark for observation, constructs confidence factors to provide reliable weights for filtering, and designs a step-size calibration mechanism to provide stable constraints for prediction. These three elements support each other and optimize in a closed loop. Experiments show that this method reduces the positioning RMSE to 0.467m in dynamic radiation environments, which is 28.5% and 19.6% higher than traditional PDR and pure PBR, respectively. Even under extreme conditions with only one available PBR base station, it maintains a high-precision positioning accuracy of 0.460m. It combines the comprehensive advantages of high positioning accuracy, strong anti-interference ability, continuous and smooth trajectory, and outstanding robustness in extreme environments, comprehensively solving the core pain points of indoor positioning in dynamic radiation environments.

[0009] Optionally, the phase position confidence factor is calculated as follows:

[0010]

[0011] in, Indicates the measured waveform Compared with ideal waveform Waveform structure similarity between and The measured waveform and the ideal waveform are captured in the respective capture window. W The mean amplitude within, Indicates the time sampling point. As a regulating factor, The center of the similarity threshold.

[0012] Thus, the similarity between the measured channel impulse response waveform and the ideal line-of-sight waveform is calculated using a normalized cross-correlation algorithm, and the phase position confidence factor is solved by combining the adjustment factor and the similarity threshold center. Compared with traditional macroscopic discrimination methods that rely on residuals and signal-to-noise ratios, this invention achieves fine-grained channel state identification at the physical layer waveform feature level, which can accurately distinguish between line-of-sight and non-line-of-sight propagation scenarios and identify abnormal ranging data in advance. This calculation method is not affected by signal energy attenuation and only relies on waveform structure to complete quality assessment. The discrimination results are more real-time and more accurate, and can provide a reliable quantitative basis for subsequent adaptive adjustment of measurement noise, effectively improving the accuracy of interference identification in complex indoor environments.

[0013] Optionally, the PDR step size estimate is expressed as:

[0014]

[0015] in, This represents the sum of the range features of acceleration for each gait as analyzed by the IMU. Represents the actual physical translation distance. This represents the instantaneous optimal PDR step size estimate for the current gait interval. This represents the smoothing learning rate factor.

[0016] Thus, by relying on the real displacement and gait acceleration features within the sliding window to solve for the instantaneous optimal step length, and combining this with a smooth learning rate to complete the temporal update, the shortcomings of the traditional Weinberg model, which uses a fixed step length parameter and cannot adapt to changes in walking speed and gait posture, are solved. Step length is calibrated through reverse closed-loop calibration using high-precision positioning results, freeing it from the constraints of empirical parameters and effectively correcting single-step estimation errors. Exponential moving average filtering is used to suppress step length oscillations caused by instantaneous observation noise, ensuring smooth and continuous changes in step length parameters. This fundamentally curbs the continuous accumulation of PDR dead reckoning errors, significantly improving the stability of relative trajectory estimation in long-distance walking scenarios and providing accurate and reliable motion constraints for fusion positioning.

[0017] Optionally, a state preservation strategy may also be included, the state preservation strategy comprising: (1) Set the physical boundary of the step size parameter as ; (2) Examine the estimated PDR step size within the current time window. If the current window's observation data is found to be subject to abnormal interference, the valid parameter values ​​from the previous time step will be used. ; (3) Monitor the external environment in real time. When the positioning system loses a BLE signal, it will... .

[0018] Thus, by adding physical boundary constraints and anomaly detection mechanisms for step size parameters, along with a state preservation strategy in case of signal loss, the problems of parameter out-of-bounds errors and step size distortion caused by abnormal observations during step size iteration updates are effectively solved. By reasonably defining the parameter value range, invalid values ​​that do not conform to human walking patterns can be avoided. When encountering strong environmental interference or distorted observation data, historical valid parameters are automatically used to avoid sudden changes in positioning results. In situations such as BLE signal interruption and base station disconnection, the operating status is locked in a timely manner to ensure that the PDR calculation process is not interrupted, greatly improving the system's fault tolerance, making the positioning trajectory smoother and more consistent, and further enhancing the operational stability and environmental adaptability of the positioning system under complex conditions.

[0019] Optionally, the Kalman gain in the fusion localization process in step 5 is expressed as:

[0020] in, This represents the prior state covariance prediction matrix. Represents the Jacobian matrix of the observation model transpose, This represents the measurement noise covariance matrix.

[0021] Thus, the standard calculation form of Kalman gain in the fusion positioning process is clearly defined, and the solution is completed based on the prior state covariance matrix, the observation Jacobian matrix, and the adaptive measurement noise matrix. This method can dynamically allocate correction weights according to the reliability of the observation data, giving full play to the correction effect of high-precision observation data, while mitigating the adverse effects of poor-quality ranging data. Compared with the fixed gain mode, it can accurately balance the weight relationship between the predicted state value and the measured observation value, optimize the filtering correction amplitude, effectively improve the rationality and scientificity of the state correction in the EKF fusion process, further improve the overall positioning convergence speed, ensure that the positioning results fit the actual walking trajectory, and enhance the overall accuracy and dynamic adaptation capability of fusion positioning.

[0022] Optionally, the formulas for calculating the posterior state vector and the posterior state covariance matrix in step 5 are as follows:

[0023]

[0024] in, Indicates based on The prior state vector for time-state prediction. Indicates Kalman gain, Indicates time t The observation vector, Represents the observation model function; The Jacobian matrix represents the observation model. This represents the prior state covariance prediction matrix.

[0025] Thus, by defining the update calculation formulas for the posterior state vector and the state covariance matrix, the complete correction process of EKF fusion positioning is improved. State measurement mapping is completed through the observation model, and accurate correction of the predicted state is achieved by combining Kalman gain, while simultaneously updating the state uncertainty matrix. This process can efficiently fuse PDR motion estimation results and PBR ranging observation information, reasonably correct walking position deviations, and quantify the positioning error range in real time. Compared with traditional fusion methods, it can steadily converge positioning errors, ensure continuous optimization of positioning results, make the output trajectory fit the actual movement path of pedestrians, effectively improve positioning continuity and data reliability, and further enhance the accuracy and dynamic adaptation performance of overall fusion positioning.

[0026] Optionally, the expression for the sum of the characteristic terms of the acceleration range of each gait resolved by the IMU is:

[0027] in, and The first j The maximum and minimum values ​​of acceleration within each gait cycle.

[0028] Thus, by statistically analyzing the extreme values ​​of acceleration within each gait cycle, obtaining the range feature, and summing them up, the gait motion features of pedestrians can be accurately extracted, conforming to the computational logic of the Weinberg stride length model. This feature can intuitively reflect the variation law of stride length, and the data extraction is simple and efficient, providing a stable and reliable feature basis for solving the instantaneous optimal stride length, ensuring that the stride length estimation closely matches the actual walking state.

[0029] Optionally, the state prediction model can be represented as:

[0030] in, This is the process noise vector. Represents the state vector. For pedestrians tTwo-dimensional position coordinates at time, for t The heading angle at any moment, for t The step size of time, This represents the integral change in heading of the gyroscope during the current gait cycle; The step size is the increment of the random walk.

[0031] Thus, by refining the pedestrian motion state prediction model, accurately constructing the state transfer relationship between position, heading, and stride length, and introducing the heading integral change and stride length random walk increment to fit actual walking characteristics, the model can accurately complete PDR state prior prediction, realistically restore the pedestrian motion trajectory change pattern, provide an accurate and reliable prediction basis for subsequent filtering and fusion, and effectively improve the rationality and accuracy of overall positioning inference.

[0032] In a second aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multimodal calculation method for the position of a moving target under dynamic radiation environment as described in any of the preceding claims.

[0033] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the multimodal solution method for the position of a moving target under dynamic radiation environment as described in any of the preceding claims. Attached Figure Description

[0034] 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 accompanying 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.

[0035] Figure 1 This is a diagram of the experimental scenario for this invention.

[0036] Figure 2 This is a single-step positioning error distribution diagram of the phase dynamic compensation of the present invention.

[0037] Figure 3 This is a comparison diagram of the phase screening optimization positioning results of the present invention.

[0038] Figure 4 This is a line graph showing the change in step error for each algorithm.

[0039] Figure 5 This is a comparison chart of two-dimensional localization results under different time length models.

[0040] Figure 6This is a comparison chart of the trajectory fitting effects of various algorithms.

[0041] Figure 7 This is a cumulative distribution diagram of trajectory errors for each algorithm.

[0042] Figure 8 This is a cumulative distribution diagram of system positioning errors for various algorithms under base station failure conditions.

[0043] Figure 9 It is a system positioning trajectory diagram of each algorithm under the condition of base station failure. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0045] Example 1: This embodiment provides a multimodal method for calculating the position of a moving target under dynamic radiation conditions, characterized in that the method includes: Step 1: Construct a positioning parameter system and obtain BLE base station coordinates, pedestrian stride length, heading angle, PBR ranging value, and ambient relative humidity data; Step 2: Use a data-driven nonlinear phase dynamic compensation mechanism to correct the PBR phase lag caused by humidity and obtain the compensated ranging observation value; The nonlinear phase dynamic compensation mechanism includes: establishing a nonlinear mapping model between humidity and phase lag in the offline stage, collecting environmental humidity data in real time in the online stage, calculating the phase compensation amount based on the mapping model, and correcting the original PBR phase observation values. Step 3: Calculate the phase position confidence factor based on the similarity of the channel impulse response first path waveform; Step 4: Based on the phase position confidence factor, adaptively adjust the EKF measurement noise covariance matrix, as shown below:

[0046] in, Indicates the first i The base station in the t The phase position confidence factor at time. This represents the baseline variance of PBR ranging under ideal line-of-sight conditions. Step 5: Within the preset sliding window, calculate the actual physical translation distance within the window based on the posterior coordinates output by EKF, and simultaneously extract the sum of the range feature terms of each gait acceleration resolved by the IMU within the window; based on the spatial geometric equivalence relationship, back-calculate the instantaneous optimal PDR step length estimate of the current gait interval, and perform time-series smoothing update of the PDR step length estimate. Step 6: Combining the measurement noise covariance matrix adjusted in Step 4 and the PDR step size parameters corrected in Step 5, the EKF framework is used to perform PBR / PDR fusion localization.

[0047] To address the issue of systematic humidity phase lag in PBR under dynamic radiation environments, this embodiment employs a data-driven nonlinear dynamic phase compensation mechanism, which constructs humidity data offline. The phase lag mapping model and online real-time correction of phase observations effectively eliminate systematic ranging biases caused by changes in the dielectric constant of the medium due to air humidity, improving the accuracy and stability of PBR raw observations. This embodiment calculates the phase position confidence factor based on CIR first-path waveform similarity and adaptively adjusts the EKF measurement noise covariance matrix to achieve fine-grained physical layer channel quality assessment. It automatically reduces the weight of high-interference and non-line-of-sight ranging measurements, avoiding filter contamination by abnormal observations and significantly improving multipath and non-line-of-sight resistance capabilities. Within a sliding window, this embodiment uses EKF posterior coordinates to back-calculate the instantaneous optimal step size and performs time-series smooth updates to the step size parameters, achieving online adaptive calibration of the step size parameters, suppressing long-term cumulative drift in PDR, and improving long-endurance positioning continuity and global accuracy.

[0048] This embodiment utilizes phase compensation to provide a precise benchmark for observation, constructs confidence factors to provide reliable weights for filtering, and designs a step-size calibration mechanism to provide stable constraints for prediction. These three elements support each other and optimize in a closed loop. Experiments show that this method reduces the positioning RMSE to 0.467m in a dynamic radiation environment, which is 28.5% and 19.6% higher than traditional PDR and pure PBR, respectively. Even under extreme conditions with only one available PBR base station, it maintains a high-precision positioning of 0.460m. It combines the comprehensive advantages of high positioning accuracy, strong anti-interference ability, continuous and smooth trajectory, and outstanding robustness in extreme environments, comprehensively solving the core pain points of indoor positioning in dynamic radiation environments.

[0049] Example 2: This embodiment provides a multimodal solution method for the position of a moving target under dynamic radiation environment. Based on Embodiment 1, the calculation method of the phase position confidence factor is defined as follows:

[0050]

[0051] in, Indicates the measured waveform Compared with ideal waveform Waveform structure similarity between and The measured waveform and the ideal waveform are captured in the respective capture window. W The mean amplitude within, Indicates the time sampling point. As a regulating factor, The center of the similarity threshold.

[0052] The similarity between the measured channel impulse response waveform and the ideal line-of-sight waveform is calculated by normalized cross-correlation algorithm. The phase position confidence factor is solved by combining the adjustment factor and the similarity threshold center. It is not affected by signal energy attenuation and completes the quality assessment based solely on the waveform structure. The discrimination results are more real-time and more accurate, and can provide a reliable quantitative basis for subsequent adaptive adjustment of measurement noise, effectively improving the accuracy of interference identification in complex indoor environments.

[0053] Example 3: This embodiment provides a multimodal method for calculating the position of a moving target under dynamic radiation conditions. Based on any of the above embodiments, the PDR step size estimate is defined as follows:

[0054]

[0055] in, This represents the sum of the range features of acceleration for each gait as analyzed by the IMU. Represents the actual physical translation distance. This represents the instantaneous optimal PDR step size estimate for the current gait interval. This represents the smoothing learning rate factor.

[0056] This embodiment solves for the instantaneous optimal step length by relying on the real displacement and gait acceleration features within the sliding window, and completes the temporal update by combining a smooth learning rate. This solves the problem that the traditional Weinberg model uses a fixed step length parameter and cannot adapt to changes in walking speed and gait posture. By using exponential moving average filtering to suppress step length oscillations caused by instantaneous observation noise, the step length parameter changes smoothly and continuously, fundamentally curbing the continuous accumulation of PDR dead reckoning errors, and significantly improving the stability of relative trajectory reckoning in long-distance walking scenarios, providing accurate and reliable motion constraints for fusion positioning.

[0057] Example 4: This embodiment provides a multimodal method for calculating the position of a moving target in a dynamic radiation environment. Based on embodiment 3, it adds a state preservation strategy, including: (1) Set the physical boundary of the step size parameter as ; (2) Examine the estimated PDR step size within the current time window. If the current window's observation data is found to be subject to abnormal interference, the valid parameter values ​​from the previous time step will be used. ; (3) Monitor the external environment in real time. When the positioning system loses a BLE signal, it will... .

[0058] This embodiment adds physical boundary constraints and anomaly detection mechanisms for step size parameters, along with a state preservation strategy in case of signal loss, effectively solving the problems of parameter out-of-bounds errors and step size distortion caused by abnormal observations during step size iteration updates. By reasonably defining the parameter value range, invalid values ​​that do not conform to human walking patterns can be avoided. When encountering strong environmental interference or distorted observation data, historical valid parameters are automatically used to avoid sudden changes in positioning results. In situations such as BLE signal interruption and base station disconnection, the operating status is locked in a timely manner to ensure that the PDR calculation process is not interrupted, greatly improving the system's fault tolerance, making the positioning trajectory smoother and more consistent, and further enhancing the operational stability and environmental adaptability of the positioning system under complex conditions.

[0059] Example 5: This embodiment provides a multimodal method for calculating the position of a moving target in a dynamic radiation environment. Based on any of the above embodiments, the Kalman gain in the fusion localization process in step 5 is defined as follows:

[0060] in, This represents the prior state covariance prediction matrix. Represents the Jacobian matrix of the observation model transpose, This represents the measurement noise covariance matrix.

[0061] This embodiment clarifies the standard calculation form of Kalman gain during the fusion positioning process, relying on the prior state covariance matrix, observation Jacobian matrix, and adaptive measurement noise matrix to complete the solution. This method can dynamically allocate correction weights according to the reliability of observation data, giving full play to the correction effect of high-precision observation data, while mitigating the adverse effects of poor-quality ranging data. Compared with the fixed gain mode, it can accurately balance the weight relationship between the predicted state value and the measured observation value, optimize the filtering correction amplitude, effectively improve the rationality and scientificity of state correction during EKF fusion, further improve the overall positioning convergence speed, ensure that the positioning result fits the actual walking trajectory, and enhance the overall accuracy and dynamic adaptation capability of fusion positioning.

[0062] Example 6: This embodiment provides a multimodal method for calculating the position of a moving target under dynamic radiation conditions. Based on any of the above embodiments, the calculation formulas for the posterior state vector and the posterior state covariance matrix in step 5 are defined as follows:

[0063]

[0064] in, Indicates based on The prior state vector for time-state prediction. Indicates Kalman gain, Indicates time t The observation vector, Represents the observation model function; The Jacobian matrix represents the observation model. This represents the prior state covariance prediction matrix.

[0065] This embodiment defines the update calculation formulas for the posterior state vector and the state covariance matrix, thus perfecting the complete correction process of EKF fusion positioning. State measurement mapping is completed through an observation model, and accurate correction of the predicted state is achieved by combining Kalman gain, while simultaneously updating the state uncertainty matrix. This process can efficiently fuse PDR motion estimation results and PBR ranging observation information, reasonably correct walking position deviations, and quantify the positioning error range in real time. Compared with traditional fusion methods, it can steadily converge positioning errors, ensure continuous optimization of positioning results, make the output trajectory fit the actual movement path of the pedestrian, effectively improve positioning continuity and data reliability, and further enhance the accuracy and dynamic adaptation performance of the overall fusion positioning.

[0066] Example 7: This embodiment provides a multimodal method for calculating the position of a moving target under dynamic radiation conditions, including: Step 1: Construct a positioning parameter system and obtain BLE base station coordinates, pedestrian step length, heading angle, PBR ranging value and ambient relative humidity data.

[0067] Step 2: Use a data-driven nonlinear phase dynamic compensation mechanism to correct the PBR phase lag caused by humidity and obtain the compensated ranging observation value.

[0068] According to electromagnetic wave propagation theory, changes in the environmental dielectric constant can cause a shift in PBR ranging results. If this systematic nonlinear deviation is not corrected, it will seriously affect the update steps of the Extended Kalman Filter (EKF) fusion positioning in dynamic environments. Given that it is extremely difficult to accurately construct a real-time physical model of the dielectric constant in complex industrial or indoor environments, this embodiment proposes a data-driven phase dynamic compensation strategy.

[0069] The compensation mechanism mainly comprises two stages: offline calibration and online correction. In the offline stage, the system constructs a phase shift database under different radiation conditions within a controlled experimental environment. Specifically, while maintaining a constant physical distance between the transmitting and receiving terminals, environmental parameters are adjusted using a set radiation condition step size. A large number of phase observations are collected at each radiation condition node, and statistical averaging and curve fitting methods are used to extract the phase reference characteristics under that radiation condition, thereby constructing a nonlinear mapping model between environmental radiation conditions and phase hysteresis. :

[0070] During the online positioning phase, the system acquires real-time radiation condition data from the hygrometer and applies a compensation factor to the original observations:

[0071] In the formula, This indicates the corrected phase after compensation. The raw measurement phase received by the base station. for Radiation conditions collected at all times.

[0072] By introducing this compensation item The system can eliminate the phase shift introduced by fluctuations in radiation conditions, thereby restoring the observed phase to a physical quantity that only represents geometric distance.

[0073] Step 3: Calculate the phase position confidence factor based on the similarity of the first path waveform of the Channel Impulse Response (CIR).

[0074] Although the aforementioned dynamic compensation mechanism effectively eliminates systematic phase lag, in multipath fading environments, the radio frequency signal is at risk of being blocked, causing the phase information calculated by the receiver to be dominated by the reflection path. Therefore, it is necessary to rigorously evaluate the ranging quality of a single communication link at the physical level.

[0075] This embodiment proposes a signal purity evaluation algorithm based on CIR first-path waveform matching. This algorithm does not require prior knowledge of the actual physical distance between the transmitting and receiving nodes and the channel attenuation coefficient. It can accurately quantify the degree of multipath interference solely based on waveform similarity. The algorithm's processing flow is as follows: First, in a relatively ideal, open environment free of multipath effects, CIR data under LOS conditions between the transmitting and receiving nodes are pre-acquired. Noise is filtered out through peak alignment and averaging, thereby extracting a reference pulse envelope that reflects the intrinsic response characteristics of the hardware system. Then, a length of [missing information] is truncated to both sides of the main peak value. Data fragments, constructing ideal waveform vectors .

[0076] In the online positioning solution During the step, targeting The CIR observation sequence is first used to lock the peak index of the first-path arrival time by adjusting the threshold detection. Then, using this index as the alignment reference, windows of the same length are dynamically truncated. The waveform segments constitute the measured first diameter feature vector. .

[0077] To eliminate the interference of absolute energy attenuation differences caused by distance variations on the matching results, this embodiment uses the normalized cross-correlation (NCC) function to calculate the online measured waveform. Compared with ideal waveform Waveform structure similarity between :

[0078] In the formula, , and These are the average amplitudes of the measured waveform and the ideal waveform within the selected window, respectively. Indicates the time sampling point.

[0079] The closer the value is to 1, the purer the received initial diameter waveform is, and the closer the ranging result is to the true geometric distance; conversely, it means that the signal is severely contaminated.

[0080] To enhance the algorithm's ability to suppress low-quality signals, this embodiment adjusts the similarity... A nonlinear mapping is performed to convert waveform similarity into weight parameters usable by the fusion algorithm. In this embodiment, the phase position confidence factor is defined as follows:

[0081] In the formula, As a modulating factor, it controls the sensitivity of confidence level to changes in similarity; The similarity threshold center is used. The CIR waveform similarity of the current link. Slightly below the threshold At that time, its phase position confidence factor It will attenuate; conversely, when the signal quality is excellent, It will quickly become saturated.

[0082] Step 4: Based on the phase position confidence factor, adaptively adjust the EKF measurement noise covariance matrix for BLE base stations. N For a system like this, the measurement noise covariance matrix is ​​expressed as:

[0083] in, The reference variance of PBR ranging under ideal line-of-sight conditions is given. When the communication link suffers severe multipath interference, its phase position confidence factor approaches 0, causing the corresponding observation noise variance to be nonlinearly amplified to infinity. Matrix inversion makes the weight of the infinite variance in the Kalman gain approach 0. The filter reduces the weight of this observation result, achieving adaptive isolation of gross errors in multipath ranging.

[0084] Step 5: Dynamically correct the PDR step size parameter using the high-confidence PBR ranging results through a sliding window feedback mechanism. This includes the following steps: In the sliding window Within, the actual physical translation distance calculated from the high-precision posterior coordinates output by EKF is: :

[0085] in, Indicates pedestrians t Two-dimensional position coordinates at time [time]. Indicates pedestrians W Two-dimensional position coordinates of the time step.

[0086] The sum of the characteristic terms of the acceleration range of each step resolved by the IMU within this window is extracted synchronously and denoted as . :

[0087] in, and The first j The maximum and minimum values ​​of acceleration within each gait cycle.

[0088] Based on spatial geometric equivalence, the estimated instantaneous optimal step length parameters within the current gait interval can be calculated by reverse deduction. :

[0089] because The solution is inevitably affected by local observation noise. Directly updating the PDR model with this noise would lead to oscillations in the step size estimation. Therefore, this embodiment uses a first-order exponential moving average (EMA) filter to perform time-series smoothing updates of the parameters:

[0090] In the formula, This is a smoothing learning rate factor used to balance the inertia of historical parameters with the corrected weights of new observations.

[0091] Furthermore, to prevent extreme data jumps from compromising filtering convergence and to ensure the algorithm can continue operating with the latest, calibrated user gait parameters when moving from an open region to an occluded region, the state-preservation strategy steps are as follows: (1) Set the physical boundary of the Weinberg step size parameter as This avoids the step size calculated by the system being too large or too small; (2) Verify the step size parameters calculated online within the current time window. .like If the current window's observation data is found to be subject to abnormal interference, the valid parameter values ​​from the previous time step will be used. ; (3) Real-time monitoring of the external environment. When the positioning system loses a BLE signal, it will... .

[0092] Step 6: Combining the adjusted measurement noise covariance and the corrected step size parameters, the EKF framework is used for PBR / PDR fusion localization. The specific implementation steps are as follows: Select the two-dimensional plane position of the pedestrian Heading angle and step length As the system state vector, defined State vector at time step for:

[0093] in, For pedestrians t Two-dimensional position coordinates at time, for t The heading angle at any moment, for t The step size of time.

[0094] The system's state transition equations are constructed using the PDR kinematic derivation model, assuming... to If walking events are detected at all times, the state prediction model can be expressed as:

[0095] In the formula, This represents the integral change in heading of the gyroscope during the current gait cycle; The step size random walk increment is set to a zero-mean Gaussian process without loss of generality; This is the process noise vector, representing sensor drift during the estimation process.

[0096] Regarding the aforementioned nonlinear state transition function Calculate its Jacobian moment:

[0097] The corresponding prior state covariance prediction matrix is ​​updated as follows:

[0098] In the formula, The process noise covariance matrix mainly originates from the accumulation of errors in step size and heading estimation; for t The posterior state covariance matrix at time -1.

[0099] The prior deployment location is System observation vector Composed of the ranging values ​​of each base station after environmental feature compensation, and expressed as:

[0100] in, Indicates the first n Each base station at time t The PBR ranging value, To measure noise. Due to the heading angle With step size It cannot be directly observed via the BLE distance, as its derivative is 0. Therefore, the Jacobian matrix of the observation is... The OK It can be represented as:

[0101] in, for time The geometric distance to the pedestrian's a priori estimated position.

[0102] Calculate Kalman gain :

[0103] When a communication link is subjected to severe multipath interference, its reliability factor The numerical value approaches 0, causing the corresponding observation noise variance to be nonlinearly amplified to infinity. Matrix inversion results in the infinite variance corresponding to... When the weights approach 0, the filter reduces the weight of the current observation; when When within the normal range, Will be corrected And adjust in the opposite direction and This enables dynamic calibration of PDR cumulative drift.

[0104] Finally, the formulas for calculating the posterior state vector and the posterior state covariance matrix are as follows:

[0105]

[0106] in, Indicates based on The prior state vector for time-state prediction. Indicates Kalman gain, Indicates time t The observation vector, Represents the observation model function; The Jacobian matrix represents the observation model. This represents the prior state covariance prediction matrix.

[0107] Some steps in the embodiments of the present invention can be implemented using software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0109] To comprehensively evaluate the actual performance of the proposed fusion positioning algorithm in a dynamic radiation environment, a hall in a college of a university was selected as the actual physical experimental site. Figure 1 As shown, the experiment planned a rectangular test area of ​​7m×5m in the hall, and set up four reference coordinate points. (0,9), (0,7), (0,7), At locations (0,7), BLE base stations based on the nRF54L15 chip were deployed as ranging beacons. To best replicate the actual indoor positioning base station topology, the installation height of all beacons was uniformly set to 1.6m above the ground.

[0110] During the dynamic data acquisition phase, the tester simultaneously held a smartphone terminal and a BLE receiver module, moving along a preset rectangular closed trajectory (1,1). (1,7) (5,7) (5,1) (1,1) Move clockwise at a constant speed for a total of 10 revolutions. To simulate different levels of dynamic radiation interference, the experiment specifically selected... =72%, =77% and =80% as the test radiation condition.

[0111] For underlying data acquisition, the RF observation data parsed by the BLE receiving module is transmitted to the host computer via serial port for real-time acquisition and offline storage. The smartphone serves as the inertial data source for the PDR algorithm, using Phyphox software to record real-time data from its built-in IMU sensors, including the accelerometer, gyroscope, and magnetometer. To effectively suppress accidental noise from single measurements, all raw data under different radiation conditions were acquired through multiple independent random experiments and pre-processed with mean smoothing to ensure high confidence in the final test sample set.

[0112] (1) Phase compensation performance analysis In real-world indoor positioning scenarios, the dynamic time-varying characteristics of the propagation environment, including radiation factors, cause microscopic disturbances in the propagation rate and path of radio signals, thereby inducing systematic phase lag at the physical layer. Without rigorous error calibration, this phase lag bias will be deeply coupled with NLOS and multipath fading effects, leading to a significant positive nonlinear drift in the underlying ranging observations. To verify the effectiveness of the data-driven phase dynamic compensation strategy constructed in this invention, experiments were conducted under three different radiation conditions. , , A comparative experiment was conducted below.

[0113] In the EKF fusion framework, ranging data affected by dynamic radiation interference exhibits a long-tailed distribution bias, introducing gain error during the state update phase. This leads to topological distortion and global yaw in the positioning trajectory, and the error significantly worsens with increasing relative humidity. Figure 2 The temporal distribution of single-step positioning errors and measured data show that the impact of radiated interference on positioning accuracy exhibits a significant gradient dependence. When the radiation condition φ=80%, the uncompensated system shows a divergence trend in error from the 5th to the 10th step, with a single-step peak error of 2.740m and an RMSE of 1.345m. When φ=77%, the uncompensated single-step error peak is 1.719m with an RMSE of 0.725m; and when φ=72%, the uncompensated single-step peak is 2.020m with an RMSE of 0.854m. It is evident that increased radiation conditions exacerbate BLE signal phase distortion, leading to increased ranging deviation and reduced positioning accuracy and stability.

[0114] After adopting the phase dynamic compensation mechanism, the system error was effectively suppressed, with particularly significant optimization effects under strong radiation environments. At φ=80%, the compensated RMSE decreased to 0.366m, an optimization margin of 72.8%, and the maximum single-step error was compressed to 0.682m; at φ=77%, the RMSE decreased to 0.601m, an optimization margin of 17.1%, and the maximum single-step error was 0.892m; at φ=72%, the RMSE decreased to 0.467m, an optimization margin of 45.3%, and the maximum single-step error was 0.881m. The results demonstrate that this compensation mechanism can effectively correct ranging distortion under different radiation intensities.

[0115] The quality of the underlying channel observations determines the upper limit of the fusion filter's accuracy. Dynamic phase compensation methods can correct phase lag and ranging bias caused by radiation conditions. Combining the PBR reliability factor with adaptive adjustment of the measurement noise covariance can balance the observation weights of heterogeneous sensors, reducing the interference of multipath effects and radiation phase distortion on the fusion architecture.

[0116] (2) Performance analysis of phase position reliability factor To verify the anti-interference performance of the PBR optimization algorithm based on phase position confidence factor weighting in complex indoor multipath environments, an ablation experiment was designed. Under unified radiation conditions, test scenarios, and base station topology, the weighted centroid localization method, the traditional trilateration algorithm, and the improved weighted algorithm of this invention were used to test the radiation conditions. Position calculation was performed using PBR ranging results at 72%.

[0117] Global comparison of the 2D positioning trajectory output by the three algorithms with the actual reference path Figure 3 As shown, the calculated trajectory of the weighted centroid method deviates significantly from the actual path. Due to its over-reliance on the absolute reciprocal of the ranging value, when the ranging error of a certain base station is drastically amplified by multipath propagation, the entire trajectory loses its ability to track indoor motion trends. While the calculated trajectory of the traditional trilateration algorithm can roughly reflect the motion trend, it exhibits significant high-frequency nonlinear jumps at multiple coordinate points, resulting in a certain degree of spatial divergence. This phenomenon is mainly attributed to the multipath fading effect caused by indoor obstacles, leading to significant gross errors in the PBR ranging values ​​of some physical links. Because traditional algorithms lack an adaptive sensing mechanism for the underlying channel quality, poor-quality measurement data containing gross errors is directly absorbed, thus affecting the convergence process of the location calculation system.

[0118] In contrast, introducing a phase position confidence factor can identify damaged channels and adaptively reduce the weight contribution of contaminated ranging links in equation solving based on the degree of CIR waveform distortion, effectively suppressing the adverse effects of multipath interference on spatial geometric constraints from a physical level. The optimized positioning trajectory filters out some abnormal distortion points, achieving smooth and high-precision positioning under complex conditions.

[0119] Table 1 Comparison of Position Calculation Algorithm Results

[0120] Combining Table 1 and Figure 4 Quantitative data shows that the localization output of traditional equal-weighted solution models exhibits certain fluctuations. The weighted centroid method, constrained by its algorithmic mechanism, suffers from a global RMSE as high as 1.442m, with the maximum error even soaring to 1.929m, resulting in poor performance in complex dynamic testing scenarios. The traditional trilateration equal-weighted solution model is also affected by the local multipath superposition effect, with a maximum single-step localization error reaching 1.184m and a global RMSE of 0.633m. Figure 4 It is evident that the error time-series curve of traditional algorithms exhibits significant oscillations and error spikes, indicating their susceptibility to changes in the underlying channel environment. This invention introduces a physical layer quality assessment mechanism based on CIR underlying features, which online senses and reduces the interference weight of inferior links. The error fluctuation range of this invention is narrowed, with the overall positioning RMSE converging to 0.572m and the average error reduced to 0.498m, achieving approximately 9.64% improvement in global accuracy compared to the traditional three-sided algorithm. Furthermore, this weighting strategy effectively suppresses extreme value errors, compressing the maximum error from 1.184m to 0.993m, keeping the error of most sampling points within 1.0m.

[0121] It is worth noting that at a very small number of discrete sampling points in the experimental data, the error of the weighted algorithm was slightly greater than that of the traditional equal-weighted algorithm, resulting in local accuracy degradation. This mainly occurred in local areas with harsh electromagnetic environments. Under such conditions, all available base station links detected by the system suffered from varying degrees of fading and multipath distortion. Even though the algorithm assigned higher weights to relatively better links based on CIR characteristics, the underlying ranging values ​​still retained some nonlinear multipath bias that was not completely removed, thus introducing a slight estimation bias into the weighted network. However, considering the convergence trend of the global error distribution and the decrease in the RMSE index, the dynamic quality weighting strategy proposed in this invention can still improve the accuracy and smoothness of PBR independent positioning under complex multipath conditions, thereby providing a relatively reliable position measurement benchmark for subsequent fusion positioning.

[0122] (3) Analysis of the effect of adaptive PDR step size optimization To verify the effectiveness of the proposed adaptive step size calibration algorithm based on EKF closed-loop feedback and sliding window in a real multi-source fusion positioning system, fixed step size calculation and Weinberg model calculation were set as comparison algorithms under the same observation conditions and EKF filtering framework.

[0123] As shown in Table 2, the quantitative statistical results indicate that the adaptive multi-source fusion localization algorithm constructed in this invention improves the global solution accuracy and stability. Its average error and root mean square error are lower than those of the traditional fixed step size and pure Weinberg extrapolation models, which to some extent suppresses the tendency of inertial dead reckoning to diverge over time. Although the algorithm produces a slightly higher transient extreme value error than the Weinberg model when encountering large physical disturbances and triggering the reconstruction of the system's underlying noise matrix and step size parameters, the convergence of the root mean square error confirms that the system has good observation feedback and state adjustment capabilities. This closed-loop fusion architecture improves the smoothness of the system throughout the global operating cycle while accommodating local errors at a very small number of extreme sampling points, realizing the complementary advantages of multi-source heterogeneous sensors.

[0124] Table 2. Error Table for Three Step Size Algorithms

[0125] observe Figure 5 The two-dimensional trajectory spatial distribution reveals that the traditional fixed-step-size model cannot perceive the dynamic evolution of stride length during pedestrian starts, turns, accelerations, and decelerations, leading to prior position drift introduced by the EKF during the state prediction phase. Under this condition, even with the system incorporating PBR measurements for posterior updates, the trajectory still exhibits a certain degree of divergence and geometric topological deformation at turning points. After introducing the Weinberg model for step-size adjustment, the system gains preliminary dynamic response capabilities, enabling it to perceive instantaneous stride length using the vertical axis acceleration characteristics of the mobile device, thus improving trajectory following performance.

[0126] Compared to the two algorithms mentioned above that rely on open-loop priors, the step-size adaptive algorithm based on high-confidence feedback of PBR proposed in this invention exhibits better dynamic gait tracking capabilities. This mechanism constructs a sliding time window by extracting EKF posterior coordinates, achieving online closed-loop identification and time-varying drift correction of the step-size ratio coefficient. Comprehensive comparative analysis shows that the positioning trajectory calculated by the algorithm of this invention has a high degree of fit with the set real rectangular reference path. It not only achieves a smoother transition at turning nodes and suppresses trajectory overshoot, but also reduces the global average error of the system, improving the accuracy and robustness of continuous trajectory calculation in complex indoor dynamic scenes.

[0127] (4) Comparison of fusion performance To evaluate the robustness and positioning accuracy of the proposed adaptive multi-source fusion positioning algorithm in indoor dynamic environments, a comparative experiment was designed between single-source and fusion architectures. The positioning performance of each algorithm was quantitatively statistically analyzed by extracting and comparing the absolute positioning errors of PDR dead reckoning based on the Weinberg model, PBR independent positioning based on traditional trilateration, and the adaptive EKF fusion architecture presented in this paper under the same conditions.

[0128] Table 3. Statistical table of the impact of fusion architecture on algorithm error

[0129] The quantitative statistical results in Table 3 show that single-source positioning systems exhibit certain limitations in dynamic radiation environments. The PDR estimation system, due to its heavy reliance on prior gait models, experiences sensor drift-induced errors that accumulate over time, resulting in a maximum error of 1.432 m and a global RMSE of 0.653 m, exhibiting a divergence trend. In contrast, PBR positioning based on phase reliability factor weighting utilizes absolute spatial ranging constraints to control the maximum error to 0.993 m, limiting the continued divergence of errors to some extent. However, when facing indoor multipath interference, the fading of radio frequency signals easily causes nonlinear jumps in local trajectories, resulting in an RMSE of 0.581 m and limiting its ability to independently output smooth tracks.

[0130] The fusion algorithm, through a tightly coupled architecture, effectively leverages the complementary advantages of the two heterogeneous systems. Data shows that the algorithm's average error is 0.415m, with an RMSE convergence to 0.467m, representing performance improvements of 28.5% and 19.6% compared to the PDR and PBR models, respectively. This algorithm not only mitigates local observation jumps caused by radio frequency multipath propagation but also calibrates the long-term cumulative drift of inertial estimation using ranging data. Test results demonstrate that this system compensates for the limitations of a single sensor, effectively balancing trajectory smoothness and positioning accuracy in complex dynamic scenarios.

[0131] Combination Figure 6 It can be seen that PDR exhibits error characteristics of short-term local smoothing and long-term global divergence. In the initial stage of positioning, its calculated trajectory has a good fit with the actual path; however, in subsequent turns and long straight-line travel, due to the difficulty of the fixed parameters of the traditional Weinberg model in matching the changes in pedestrian gait in real time, the deviation in step length estimation accumulates over time, resulting in a relatively low global solution accuracy in the three sets of control experiments. This phenomenon indicates that, in the absence of external absolute spatial reference constraints, the relative calculation mechanism relying solely on inertial sensors is prone to error accumulation and divergence, making it difficult to meet the requirements of long-term continuous positioning.

[0132] On the other hand, while the PBR positioning algorithm based on the credibility factor weighting eliminates some measurement gross errors caused by multipath propagation through physical layer quality sensing, in complex indoor electromagnetic environments, when the target enters a densely overlapping area of ​​multipath propagation, multiple radio frequency links may experience signal degradation and fading. Due to the lack of prior smoothing constraints on kinematic states, fluctuations in the underlying measurements affect the solution results, leading to certain nonlinear jumps and position jitters in the PBR positioning trajectory in some areas, thus limiting the smoothness of the overall trajectory.

[0133] According to Figure 7 As shown, the error distribution curve of the EKF fusion algorithm proposed in this invention is relatively close to the Y-axis, exhibiting good convergence and robustness. This algorithm achieves complementary advantages between inertial recursion and radio frequency ranging through data fusion. On the one hand, the absolute coordinate observations provided by PBR suppress the cumulative divergence of the PDR system to a certain extent; on the other hand, the high-frequency continuous displacement characteristics of PDR smooth out the local jumps and jitter of PBR. Furthermore, this invention designs a dual adaptive mechanism within the EKF framework: the observation noise matrix is ​​dynamically adjusted according to the PBR confidence factor, and the predicted state model slides back to deduce the step size coefficient based on the EKF posterior coordinates. Under this dual optimization, the error index of the fusion algorithm is improved; within an 80% confidence interval, the error of the algorithm is controlled at approximately 0.6m. In summary, the adaptive fusion positioning system constructed in this invention compensates for the shortcomings of a single sensor and, to a certain extent, reduces the impact of multipath interference and gait drift caused by complex dynamic environments.

[0134] (5) Robustness analysis In complex and ever-changing indoor environments, dynamic NLOS obstruction or base station node hardware failures can easily occur, leading to sudden signal attenuation or even complete loss of signals from one or more radio frequency base stations. To verify the robustness and continuous positioning capability of the proposed EKF multi-source fusion positioning architecture under extreme conditions such as missing measurement information, a cross-comparison experiment based on controlled base station failure simulation was conducted. Two single-source algorithms were selected as comparison benchmarks: one is a pure PDR algorithm completely unaffected by external radio frequency base stations; the other is a pure PBR algorithm under an ideal state of no obstruction (i.e., 0 base station failures). Based on this benchmark, by randomly blocking the observation data of 1, 2, and 3 base stations at time steps in the measured PBR data, the fusion system was forced into an extreme state of severely incomplete measurement information, thereby comprehensively evaluating the positioning accuracy and fault tolerance performance of the system under different degrees of obstruction.

[0135] Table 4. Algorithm positioning error under base station interference environment

[0136] As shown in Table 4, the quantitative data reveals significant physical limitations in single-source localization. The pure PDR algorithm, lacking global absolute coordinate correction, experiences an accumulation of error with each step, resulting in an RMSE as high as 0.653m, reflecting the upper bound of error divergence in pure inertial calculations. While the pure PBR model can limit the maximum error to 0.993m under ideal conditions with zero failures, it is highly susceptible to physical layer multipath effects, with an RMSE still reaching 0.581m. When two or more base stations are missing, the model becomes completely incapable of localization due to rank deficiency in the observation matrix. In contrast, the fusion algorithm proposed in this invention demonstrates strong fault tolerance. Even under severe obstruction conditions with two base stations simultaneously failing, the RMSE of the fusion algorithm remains at 0.400m. This not only ensures stable system operation but also surpasses the accuracy of the pure PBR model under ideal unobstructed conditions. Under extremely adverse conditions with three base stations failing, the RMSE of the fusion algorithm remains stable at 0.460m, significantly lower than the divergence benchmark of pure PDR. This fully demonstrates that the algorithm can maximize the transformation of fragmented residual measurement information into high-value absolute position constraints under extremely limited observation conditions, thereby achieving highly robust continuous positioning in dynamic and harsh environments.

[0137] To further and more intuitively evaluate the overall error convergence of the algorithm, Figure 8 The error cumulative distribution curves of six sets of comparative experiments are shown. It is clearly observed from the figures that the CDF curve of the fusion algorithm of this invention still maintains a significant overall leftward shift when BLE base stations fail, demonstrating a high degree of error convergence. Specifically, the curves of PDR and PBR rise relatively slowly; at the 80% cumulative probability confidence interval, their positioning errors are approximately 0.822m and 0.773m, respectively. In contrast, the errors corresponding to the 80% cumulative probability of the failure of this invention for 0-3 base stations are 0.527m, 0.603m, 0.551m, and 0.528m, respectively, all controlled within 0.65m. This indicates that the fusion positioning system statistically compresses the error distribution interval; even when only one usable BLE base station remains, the overall error distribution is still better than the pure PDR benchmark, demonstrating high system reliability.

[0138] Combination Figure 9The spatial trajectory projection of this invention can further analyze the role of the underlying geometric constraints in the fusion system under specific operating conditions. Traditional PBR trilateration algorithms rely on distance intersections from at least three base stations; if the number of effective observation base stations is less than three, the geometric circle usually cannot form an effective intersection point, easily leading to interruption of position calculation or large output errors. The EKF multi-source fusion architecture proposed in this invention is characterized by not using traditional trilateration geometric analytical solutions, but instead directly incorporating a single ranging link as an independent measurement constraint into the state update process. During system operation, PDR provides high-frequency relative motion priors, enhancing system fault tolerance while also improving the solution efficiency of the fusion algorithm; while the measurement update stage of EKF transforms the ranging information of each effective base station into constraints in the state space. Therefore, even under occlusion conditions where one or two base stations fail, the fused trajectory can still follow the actual motion route well, mitigating the high-frequency jump phenomenon that occurs in the PBR algorithm.

[0139] With only one available base station remaining, a single base station can only provide a single spherical distance constraint, making it difficult to determine the absolute spatial orientation of the target. However, the fusion algorithm of this invention did not experience computational interruption due to the reduction in measurement information. The system triggered a smooth degradation mechanism, using PDR dead reckoning to calculate the dominant trajectory shape, while simultaneously utilizing the absolute distance information from this single base station to, to some extent, guide and suppress the cumulative divergence of PDR. From the perspective of the spatial trajectory morphology, the trajectory maintains a relatively complete walking topology, avoiding the typical unbounded divergence problem of PDR trajectories. The above experimental results show that directly performing EKF fusion based on the original one-dimensional ranging information can, to some extent, avoid the defect of traditional trilateration algorithms being prone to failure when base stations are missing. This mechanism realizes the fusion of PBR / PDR in the environment of base station failure, thus providing better robustness for the overall positioning system.

Claims

1. A multimodal method for calculating the position of a moving target under dynamic radiation conditions, characterized in that, The method includes: Step 1: Construct a positioning parameter system and obtain BLE base station coordinates, pedestrian stride length, heading angle, PBR ranging value, and ambient relative humidity data; Step 2: Use a data-driven nonlinear phase dynamic compensation mechanism to correct the PBR phase lag caused by humidity and obtain the compensated ranging observation value; The nonlinear phase dynamic compensation mechanism includes: establishing a nonlinear mapping model between humidity and phase lag in the offline stage, collecting environmental humidity data in real time in the online stage, calculating the phase compensation amount based on the mapping model, and correcting the original PBR phase observation value. Step 3: Calculate the phase position confidence factor based on the similarity of the channel impulse response first path waveform; Step 4: Based on the phase position confidence factor, adaptively adjust the EKF measurement noise covariance matrix, as shown below: in, Indicates the first i The base station in the t The phase position confidence factor at time. This represents the baseline variance of PBR ranging under ideal line-of-sight conditions. Step 5: Within the preset sliding window, calculate the actual physical translation distance within the window based on the posterior coordinates output by EKF, and simultaneously extract the sum of the range feature terms of each gait acceleration resolved by the IMU within the window; based on the spatial geometric equivalence relationship, back-calculate the instantaneous optimal PDR step length estimate of the current gait interval, and perform time-series smoothing update of the PDR step length estimate. Step 6: Combining the measurement noise covariance matrix adjusted in Step 4 and the PDR step size parameters corrected in Step 5, the EKF framework is used to perform PBR / PDR fusion localization.

2. The multi-modal calculation method for the position of a moving target under dynamic radiation environment according to claim 1, characterized in that, The phase position confidence factor is calculated as follows: in, Indicates the measured waveform Compared with ideal waveform Waveform structure similarity between and The measured waveform and the ideal waveform are captured in the respective capture window. W The mean amplitude within, Indicates the time sampling point. As a regulating factor, The center of the similarity threshold.

3. The multimodal calculation method for the position of a moving target under dynamic radiation environment according to claim 1, characterized in that, The estimated PDR step size is expressed as follows: in, This represents the sum of the range features of acceleration for each gait as analyzed by the IMU. Represents the actual physical translation distance. This represents the instantaneous optimal PDR step size estimate for the current gait interval. This represents the smoothing learning rate factor.

4. The multimodal calculation method for the position of a moving target under dynamic radiation environment according to claim 1 or 3, characterized in that, It also includes a state preservation strategy, which includes: (1) Set the physical boundary of the step size parameter as ; (2) Examine the estimated PDR step size within the current time window. If the current window's observation data is found to be subject to abnormal interference, the valid parameter values ​​from the previous time step will be used. ; (3) Monitor the external environment in real time. When the positioning system loses a BLE signal, it will... .

5. The multimodal calculation method for the position of a moving target under dynamic radiation environment according to claim 1, characterized in that, The Kalman gain in the fusion localization process in step 5 is expressed as: in, This represents the prior state covariance prediction matrix. Represents the Jacobian matrix of the observation model transpose, This represents the measurement noise covariance matrix.

6. The multimodal calculation method for the position of a moving target under dynamic radiation environment according to claim 1, characterized in that, The formulas for calculating the posterior state vector and the posterior state covariance matrix in step 5 are as follows: in, Indicates based on The prior state vector for time-state prediction. Indicates Kalman gain, Indicates time t The observation vector, Represents the observation model function; The Jacobian matrix represents the observation model. This represents the prior state covariance prediction matrix.

7. The multimodal calculation method for the position of a moving target under dynamic radiation environment according to claim 1, characterized in that, The expression for the sum of the characteristic terms of the gait acceleration ranges resolved by the IMU is: in, and The first j The maximum and minimum values ​​of acceleration within each gait cycle.

8. The multimodal calculation method for the position of a moving target under dynamic radiation environment according to claim 6, characterized in that, The state prediction model is represented as: in, This is the process noise vector. Represents the state vector. For pedestrians t Two-dimensional position coordinates at time, for t The heading angle at any moment, for t Step size of time, This represents the integral change in heading of the gyroscope during the current gait cycle; The step size is the increment of the random walk.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multimodal calculation method for the position of a moving target under dynamic radiation environment as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the multimodal solution method for the position of a moving target under dynamic radiation environment as described in any one of claims 1 to 8.

Images