UWB (ultra wide band) real-time positioning method and system for obstacle shielding scene

Abstract

The invention discloses an ultra wide band (UWB) real-time positioning method and system for an obstacle shielding scene. The UWB real-time positioning method comprises the steps of initializing a positioning system, collecting UWB ranging and IMU data, executing multilateral positioning and preliminary positioning calculation, constructing a Jacobian matrix, iteratively optimizing position parameters, updating the posture of an unmanned vehicle, performing state prediction and error correction, performing robustness fusion on UWB and IMU data, and comprehensively processing a result to output an accurate position. According to the method, IMU data and UWB ranging data are fused, position parameters are optimized by using an error state Kalman filter and a Levenberg-Marquardt algorithm, high-precision positioning is realized, and meanwhile, the method has robustness to UWB ranging errors and IMU measurement noise.

Description

Technical Field

[0001] The invention relates to wireless positioning technology, and in particular to an ultra-wideband (UWB) real-time positioning method and system for obstacle-occluded scenes. Background Art

[0002] Ultra-Wideband (UWB) technology has the advantages of low system complexity, high information security, and strong anti-multipath fading capabilities, becoming a bright spot in the field of wireless positioning. Ultra-wideband positioning technology is widely used in indoor positioning, drone positioning systems and other fields. However, when there are obstacles in the scene, the ultra-wideband signal will be affected by obstacles such as metal plates, causing large errors in the ranging signal. Designing a UWB positioning system that can maintain high positioning accuracy in obstacle scenes can greatly improve the versatility of the UWB positioning system.

[0003] The UWB RF module with known actual coordinates is called a UWB base station, and the UWB RF module with unknown coordinates is called a UWB tag. The UWB positioning system is based on the principle of multilateral positioning and calculates the position of the UWB tag according to the ranging signal between the UWB RF modules and the coordinates of the UWB base station.

[0004] The situation where there are no obstacles between UWB RF modules is called line-of-sight (LOS) situation, and the situation where there are obstacles between UWB RF modules is called non-line-of-sight (NLOS) situation. The NLOS situation formed by obstacles such as metal that have a strong impact on electromagnetic wave signals will have a greater impact on the signal transmission between UWB RF modules, thereby affecting the positioning accuracy of the UWB positioning system for UWB tags.

[0005] There is an indoor fingerprint positioning method based on UWB, including offline and online stages; in the offline stage, the DNN classification model, LOS ranging error regression model and slight NLOS ranging error regression model are trained, and the mean value of LOS ranging error is obtained at the same time; in the online stage, the DNN classification model is used to classify LOS and NLOS, and the PD value is further used to classify NLOS to obtain slight NLOS, and the LOS ranging error regression model is used to estimate the ranging error of the LOS link and correct the LOS ranging, and the slight NLOS ranging error regression model is used to estimate the ranging error of the slight NLOS link and correct the slight NLOS ranging; then the MLE algorithm is used to estimate the position of the positioning target. This fingerprint positioning method is highly dependent on the environment and requires offline training of a large amount of scene information. When the positioning scene changes significantly, the positioning method fails.

[0006] There is another robot positioning method that integrates UWB and laser in NLOS scenario, which is applied to the robot and includes: mapping the scene to obtain a laser map, which is a grayscale map; marking the location of the UWB base station on the laser map; broadcasting a time data packet with its own timestamp to each UWB base station through a UWB positioning tag; receiving UWB positioning data, which includes the distance and time information between the UWB base station and the UWB positioning tag, and the time information is the time data packet; calculating whether there is an occlusion between the UWB positioning tag and the UWB base station according to the positioning result at the last moment, the laser map and the location of the UWB base station, and calculating the occlusion weight of each UWB base station; extracting the positioning data of the UWB base station based on the occlusion weight, and calculating the UWB positioning information of the robot in the map; performing EKF filtering fusion according to the UWB positioning information, the odometer and the current laser positioning result to obtain the current positioning result of the robot. However, this method requires mapping the scene based on laser radar, which increases the cost, occupies a large amount of computing resources, and has poor versatility.

[0007] In the presence of obstacles, the UWB radio module is affected by the obstacles and produces large positioning errors and lacks stability. For actual indoor positioning scenarios, due to the NLOS influence of obstacles in complex scenes, it is difficult for the UWB positioning system to obtain high positioning accuracy.

[0008] It should be noted that the information disclosed in the above background technology section is only used for understanding the background of the present application, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the invention

[0009] The main purpose of the present invention is to solve the problems existing in the above-mentioned background technology and to provide an ultra-wideband (UWB) real-time positioning method and system.

[0010] To achieve the above object, the present invention adopts the following technical solutions:

[0011] The first aspect of the present invention is an ultra-wideband (UWB) real-time positioning method, comprising the following steps:

[0012] S1: Initialize the UWB positioning system, configure the parameters of the error state Kalman filter ESKF and the Levenberg-Marquardt (LM) nonlinear optimization algorithm, and prepare to fuse IMU data and UWB ranging data;

[0013] S2: Start the UWB radio frequency module to collect the ranging signals between multiple UWB base stations and the UWB tags on the mobile platform to provide raw data for multilateral positioning; at the same time, use the IMU sensor to collect the acceleration and angular velocity data of the mobile platform to provide the motion model input for the state Kalman filter ESKF;

[0014] S3: Combining the UWB ranging data and IMU data obtained in step S2, executing the UWB multi-lateration positioning process, starting the LM algorithm for preliminary real-time positioning solution, and thus estimating the relative position error between the UWB base station and the UWB tag;

[0015] S4: Construct a Jacobian matrix based on the LM algorithm, which contains the relationship between the measured distance from the UWB base station to the tag and the predicted position; solve the descent vector of the position parameter of the UWB tag relative to the observation error, and update the position parameter of the UWB tag by the least squares method, and prepare the observation data of the UWB tag for the prediction step of the state Kalman filter ESKF;

[0016] S5: Iterative optimization of the LM algorithm is performed to solve the UWB multi-lateration positioning problem until the termination condition is met: the sum of squares of the error vector is less than the preset tolerance or the maximum number of iterations is reached, and the fine adjustment of the position parameters is completed;

[0017] S6: Using the IMU data collected in step S2, the mobile platform posture information is updated through dead reckoning to provide the latest motion state for the state update step of the ESKF;

[0018] S7: Input the position parameters of the UWB tag optimized in step S5 as observation data into the state Kalman filter ESKF, and perform state prediction and error correction in combination with the motion state updated in step S6;

[0019] S8: In the state Kalman filter ESKF, the position and velocity estimates of the mobile platform are updated based on the predicted and observed data, and the fusion is used to achieve robust processing of UWB ranging errors and IMU measurement noise;

[0020] S9: Combining the results of the state Kalman filter ESKF fusion processing in step S8 and the position parameters optimized by the LM algorithm in step S5, the precise real-time position of the UWB tag is output to complete high-precision positioning.

[0021] A second aspect of the present invention is an ultra-wideband (UWB) positioning system for obstacle-occluded scenes, the system comprising:

[0022] A plurality of UWB radio frequency modules, including a plurality of UWB base stations distributed at predetermined locations and a UWB tag carried on a mobile platform, for measuring the distance between the plurality of UWB base stations and the UWB tag;

[0023] Inertial measurement unit (IMU), mounted on the mobile platform, is used to provide information on movement speed and direction to assist UWB positioning;

[0024] The mobile platform, as a mobile carrier, carries the above-mentioned UWB tag and performs actual movement;

[0025] The main control machine communicates with the multiple UWB radio frequency modules and the inertial measurement unit IMU, and is used to execute the ultra-wideband UWB real-time positioning method.

[0026] A third aspect of the present invention is a computer program product, comprising a computer program, wherein the computer program implements the ultra-wideband (UWB) real-time positioning method when executed by a processor.

[0027] The present invention has the following beneficial effects:

[0028] The ultra-wideband UWB real-time positioning method of the present invention achieves high-precision positioning capability in the presence of obstacles through a series of innovations. First, the method adopts a lightweight positioning solution, and realizes effective fusion processing of IMU data and UWB ranging data by initializing the UWB system and configuring advanced error state Kalman filter (ESKF) and Levenberg-Marquardt (LM) nonlinear optimization algorithm. This method is not only applicable to a variety of scenarios, but also shows high robustness to environmental changes, and does not require a lot of offline training even when the obstacle scene changes.

[0029] In the multi-lateration positioning process, by constructing the Jacobian matrix and performing the least squares method, the method can accurately calculate and update the location parameters of the UWB tag, and iteratively optimize until strict termination conditions are met, ensuring the fine adjustment of the location parameters. In addition, the method also includes real-time detection of obstacles and elimination of the influence of non-line-of-sight (NLOS) channels, which significantly improves the positioning accuracy.

[0030] In addition, the present invention also specially designs a sensor data corrector to suppress the measurement errors of the UWB radio frequency module and the IMU sensor itself, reducing the high requirements for the accuracy of the UWB radio frequency module, thereby reducing the system cost. The entire system exhibits low computational complexity and high tolerance to sensor errors, enabling high-precision positioning even under resource-constrained conditions. Finally, by integrating the Kalman filter processing results and the position parameters optimized by the LM algorithm, the present invention can output the precise real-time position of the UWB tag to meet the needs of real-time positioning.

[0031] The UWB real-time positioning method of the present invention can realize accurate positioning of a mobile platform in an obstacle-blocking scene, effectively handle the influence of the blockage on the UWB signal, and improve the accuracy and reliability of positioning.

[0032] Other beneficial effects of the embodiments of the present invention will be further described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a physical picture of the unmanned vehicle according to an embodiment of the present invention.

[0034] Figure 2 This is a LOS scene diagram of an embodiment of the present invention.

[0035] Figure 3 Schematic diagram of a LOS scenario according to an embodiment of the present invention.

[0036] Figure 4 This is a NLOS scene diagram of an embodiment of the present invention.

[0037] Figure 5 Schematic diagram of an NLOS scenario according to an embodiment of the present invention.

[0038] Figure 6 This is a structural diagram of an improved LM nonlinear optimization algorithm according to an embodiment of the present invention.

[0039] Figure 7 4 is a structural diagram of a sight distance identifier according to an embodiment of the present invention.

[0040] Figure 8 It is a structural diagram of a GPR-based online sensor data corrector according to an embodiment of the present invention.

[0041] Fig. 9 This is a diagram of the overall module structure of the NLOS-resistant UWB positioning system according to an embodiment of the present invention.

[0042] Fig.10 It is a dynamic experimental positioning position curve and a real position curve of the UWB fusion positioning system in the LOS scenario of an embodiment of the present invention.

[0043] Fig.11 This is a UWB dynamic positioning position curve and a real position curve in the LOS scenario of an embodiment of the present invention.

[0044] Fig.12 It is a dynamic experimental positioning position curve and a real position curve of the UWB fusion positioning system in the NLOS scenario of an embodiment of the present invention.

[0045] Fig.13 This is a UWB dynamic positioning position curve and a real position curve in an NLOS scenario of an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following is a detailed description of the embodiments of the present invention. It should be emphasized that the following description is only exemplary and is not intended to limit the scope and application of the present invention.

[0047] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0048] See also Figures 3 to 9 The embodiment of the present invention provides an ultra-wideband (UWB) real-time positioning method, comprising the following steps:

[0049] S1: Initialize the UWB positioning system, configure the parameters of the error state Kalman filter (ESKF) and the Levenberg-Marquardt (LM) nonlinear optimization algorithm, and prepare to fuse IMU data and UWB ranging data;

[0050] S2: Start the UWB radio frequency module to collect ranging signals between multiple UWB base stations and UWB tags on mobile platforms such as unmanned vehicles, providing raw data for multilateral positioning; at the same time, use the IMU sensor to collect the acceleration and angular velocity data of the unmanned vehicle to provide motion model input for the state Kalman filter ESKF;

[0051] S3: Combining the UWB ranging data and IMU data obtained in step S2, executing the UWB multi-lateration positioning process, starting the LM algorithm for preliminary real-time positioning solution, and thus estimating the relative position error between the UWB base station and the UWB tag;

[0052] S4: Construct a Jacobian matrix based on the LM algorithm, which contains the relationship between the measured distance from the UWB base station to the tag and the predicted position; solve the descent vector of the position parameter of the UWB tag relative to the observation error, and update the position parameter of the UWB tag by the least squares method, and prepare the observation data of the UWB tag for the prediction step of the state Kalman filter ESKF;

[0053] S5: Iterative optimization of the LM algorithm is performed to solve the UWB multi-lateration positioning problem until the termination condition is met: the sum of squares of the error vector is less than the preset tolerance or the maximum number of iterations is reached, and the fine adjustment of the position parameters is completed;

[0054] S6: Using the IMU data collected in step S2, the unmanned vehicle posture information is updated through dead reckoning to provide the latest motion state for the state update step of the ESKF;

[0055] S7: Input the position parameters of the UWB tag optimized in step S5 as observation data into the state Kalman filter ESKF, and perform state prediction and error correction in combination with the motion state updated in step S6;

[0056] S8: In the state Kalman filter ESKF, the position and velocity estimates of the unmanned vehicle are updated based on the predicted and observed data, and the fusion is used to achieve robust processing of UWB ranging errors and IMU measurement noise;

[0057] S9: Combining the results of the state Kalman filter ESKF fusion processing in step S8 and the position parameters optimized by the LM algorithm in step S5, the precise real-time position of the UWB tag is output to complete high-precision positioning.

[0058] See also Figure 7 In a preferred embodiment, the method further comprises the following steps:

[0059] Before step S1, offline processing is performed, training parameters are set, UWB data is collected, and a data set is established according to the line-of-sight situation; the data set is used to train the SVM model for subsequent line-of-sight recognition; while collecting the UWB ranging signal in step S2, the trained SVM model is used to perform line-of-sight recognition on the signal, and the UWB signal is identified and classified as a "line-of-sight" LOS or a "non-line-of-sight" NLOS situation; if the SVM model determines that it is a "non-line-of-sight" NLOS situation, the corresponding UWB ranging signal is discarded and is not used for subsequent positioning calculations; if the SVM model determines that it is a "line-of-sight" LOS situation, the UWB ranging signal is retained and is input as valid data into the positioning solution process of steps S3 to S9.

[0060] See also Figure 8 In a preferred embodiment, the method further comprises the following steps:

[0061] Before step S1, an offline stage processing is performed, and training parameters are set and a feature data set is established based on the collected UWB measurement data; the real distance of the UWB measurement data in the data set is annotated for training the GPR model; the Gaussian process regression GPR model is trained using the annotated data set until the model parameters meet the preset requirements, so as to prepare for the measurement error correction in the online stage; in step S3, after the collected UWB ranging signal is initially positioned and solved, the characteristics of the ranging signal are input into the sensor data corrector; the measurement error of the UWB radio frequency module is corrected in real time using the trained GPR model, and the corrected accurate distance information is output; the corrected accurate distance information is used as observation data, and together with the UWB tag position parameters optimized in step S5, is input into the state Kalman filter ESKF for further position estimation and error correction.

[0062] In a preferred embodiment, in step S3, the system modeling of the multilateral positioning is implemented by the following method, including: determining a distance measurement method based on the arrival time TOA principle, which is used to calculate the straight-line distance between UWB radio frequency modules; wherein the UWB base station is a UWB radio frequency module with a known position, and the UWB tag is a UWB radio frequency module with a position to be estimated; using the timestamps of the signal sent by the UWB base station and the signal received by the UWB tag, and calculating the distance between the two by the known UWB signal wave speed; constructing a multilateral positioning equation group for determining the position of the UWB tag based on the distance measurement value from the UWB base station to the UWB tag; converting the multilateral positioning equation group into a linear equation group, constructing a coefficient matrix and a constant vector, which are used to numerically solve the position of the UWB tag; solving the linear equation group by a numerical method to obtain the three-dimensional position coordinates of the UWB tag, thereby achieving high-precision positioning.

[0063] In a preferred embodiment, in step S3: in the process of executing UWB multilateral positioning, the error between each UWB base station and the UWB tag is calculated by defining a positioning error function, which is based on the predicted position and the actual measured distance; an error vector is further constructed to summarize the positioning errors between all base stations and tags; in step S4: on the basis of constructing a Jacobian matrix, the partial derivative information of the positioning error function contained in the matrix with respect to the predicted position parameter is used to construct and solve the least squares problem, and the error function is approximated by a first-order Taylor expansion to solve the descent vector of the position parameter; in step S5: in the iterative optimization process, the optimization step size is adjusted using the trust region method, and the position parameter is improved according to the sum of squares of the error vector; and after each iteration, the radius of the trust optimization area is adjusted according to the new sum of squares of the error vector until the termination condition is met; preferably, the tolerance and the maximum number of iterations of the algorithm are set as the termination condition to ensure the convergence and stability of the algorithm; in steps S3 to S5: a position feedback mechanism is introduced, and the solution of the previous positioning cycle is used as the initial optimization point of the next cycle during the iteration process to improve the insensitivity of the algorithm to the selection of initial parameters and enhance the robustness of the algorithm.

[0064] See also Fig. 9In a preferred embodiment, in step S2, while starting the UWB radio frequency module to collect the ranging signal, the IMU sensor is synchronously started to collect the acceleration and angular velocity data of the unmanned vehicle, which will be used for subsequent track calculation and state prediction; using the IMU data, track calculation is performed based on the Euler numerical integration method to determine the initial displacement and velocity state variables of the unmanned vehicle at the next moment; in step S6, the IMU data collected in step S2 is used to perform track calculation based on the Euler numerical integration method to determine the posture change of the unmanned vehicle at the next moment; the IMU sensor is used to calculate the acceleration and angular velocity data of the unmanned vehicle. The high-frequency data of the IMU is fused with the low-frequency data of the UWB radio frequency module, and the posture information of the unmanned vehicle is updated through short-time track calculation, and an accurate initial state estimation is provided for the error state Kalman filter; considering that the sending frequency of IMU data is higher than the sending frequency of UWB data, the posture change of the unmanned vehicle during the UWB data update interval is estimated through track calculation, and the state variables are updated; in step S7, in the error state Kalman filter, the IMU data and the UWB ranging data are combined to perform a fusion process of the state variables to improve the positioning accuracy.

[0065] See also Fig. 9 In a preferred embodiment, before using the collected UWB ranging signal for preliminary positioning solution in step S3, the UWB ranging signal is firstly filtered by a mean filter, and the output filtering result is used as the input data of the LM algorithm and the error state Kalman filter ESKF. In a more preferred embodiment, the mean filter has a kernel of a set size for smoothing a continuous sequence of measured values; each element in the filter kernel has an equal weight to ensure the uniformity of the filtering process; and a series of filtered continuous measured values ​​are obtained by convolving the measured value sequence with the filter kernel and calculating the filtering result of each measured value point in the sequence in turn.

[0066] See also Fig. 9In a preferred embodiment, in step S3: in the process of executing UWB multi-lateration positioning, the distance measurement information, first path signal strength indication (FPSSI) and total received signal strength indication (TRSSI) collected by the Nooploop LinkTrack PA UWB radio frequency module are used to analyze the changes in these signal strengths to identify whether there is multipath phenomenon in UWB signal transmission; based on the support vector machine SVM model, a training sample set is constructed using the collected UWB ranging information and signal strength indication, each sample in the sample set corresponds to a label, indicating line-of-sight (LOS) or non-line-of-sight (NLOS) conditions, and the UWB signal is classified by training the SVM model to identify whether the current UWB signal is LOS or NLOS. In step S4: if the SVM model determines that the current UWB signal is NLOS, the signal is discarded and not used in subsequent positioning calculations to reduce the impact of multipath effects on positioning accuracy; if it is determined to be LOS, the signal is retained and input as valid data into the subsequent positioning solution process. In the offline stage, the training set is established using the UWB measurement data and the corresponding real distance, and the training is performed through the Gaussian process regression (GPR) model, which uses the square exponential kernel function as the covariance function and uses the automatic correlation determination kernel function (ARDK) to better retain the data characteristics. In step S6: In the online stage, the measurement error of the UWB radio frequency module is corrected in real time using the trained GPR model, and the corrected accurate distance information is output, and combined with the optimized UWB tag position parameters, it is input into the state Kalman filter ESKF for further position estimation and error correction.

[0067] In the above preferred embodiments of the present invention, a line-of-sight classifier based on support vector machine is designed, a sensor data corrector based on Gaussian process regression is designed, and a real-time positioning algorithm for a UWB positioning system that resists NLOS is proposed.

[0068] An embodiment of the present invention also provides an ultra-wideband UWB positioning system for obstacle-occlusion scenarios, the system comprising: multiple UWB radio frequency modules, including multiple UWB base stations distributed at predetermined positions and UWB tags mounted on a mobile platform, for measuring the distance between the multiple UWB base stations and the UWB tags; an inertial measurement unit IMU, mounted on the mobile platform, for providing moving speed and direction information to assist UWB positioning; a mobile platform, as a mobile carrier, carrying the above-mentioned UWB tags for actual movement; a main control machine, communicating with the multiple UWB radio frequency modules and the inertial measurement unit IMU, for executing the ultra-wideband UWB real-time positioning method.

[0069] The specific embodiments of the present invention are further described below.

[0070] System composition

[0071] The NLOS-resistant UWB positioning system includes a real-time positioning algorithm and a hardware experimental platform.

[0072] The real-time positioning algorithm runs on the ubuntu20.04 (Linux operating system mainly for desktop applications) operating system and ROS-noetic (open source robot operating system) software environment.

[0073] The hardware platform consists of a main control machine, a sensing device (UWB radio frequency module, IMU), and a communication device between the upper and lower computers. The hardware system is deployed on a four-wheeled unmanned vehicle, and a rechargeable battery is deployed on the unmanned vehicle to power it, so that the sensing device can obtain sensing information in real time and transmit it to the upper computer for real-time positioning and solving.

[0074] System Modeling

[0075] The straight-line distance between the transmitting end and the receiving end of the UWB radio frequency module can be measured based on TOA to calculate the straight-line distance according to the signal transmission time, or based on the signal strength model to measure the straight-line distance according to the attenuation of the signal strength. The UWB radio frequency module used in the present invention measures the distance based on TOA.

[0076] In the UWB positioning system, the UWB RF module with a known real position is called a UWB base station, and the UWB RF module with an estimated position is called a UWB tag. Assuming the UWB wave speed is v, the timestamp when the UWB signal is sent from the UWB base station is stamp1, and the timestamp when the UWB tag receives the UWB signal is stamp2, the measured distance is:

[0077]

[0078] The multi-lateration positioning method measures the distance from each UWB base station to the UWB tag, and obtains the position of the measured UWB tag through the relationship equation between coordinates and distance.

[0079] Suppose there are N UWB base stations, and the coordinates of the i-th base station are p i (x i ,y i ,z i ), d i is the distance between the UWB base station and the UWB tag, p(x, y, z) is the coordinate of the UWB tag to be determined, and we can get:

[0080]

[0081] Hardware Experiment Platform

[0082] In the hardware platform design, different modules communicate with each other using Topic in the ROS system (a communication mechanism in the ROS operating system).

[0083] Use a Scout unmanned vehicle equipped with a laser radar, an IMU module, and a UWB radio module, such as Figure 1 The upper computer communicates with the lower computer carried by the vehicle body based on WiFi network remote communication. Taking the initial point of the UWB tag of the unmanned vehicle as the origin of the positioning position, the positions of each UWB base station relative to the unmanned vehicle are shown in Table 1.

[0084] Table 1 UWB base station coordinates

[0085]

[0086]

[0087] The models of each module are shown in Table 2.

[0088] Table 2 Introduction of experimental equipment

[0089]

[0090] LOS (no obstacles) experimental scenario Figure 2 The scene diagram is as follows Figure 3 .

[0091] By placing a metal plate in front of the UWB base station 2 as an obstacle, a NLOS positioning scenario is formed, such as Figure 4 The schematic diagram of NLOS scenario is as follows: Figure 5 .

[0092] Real-time positioning algorithm

[0093] The UWB positioning algorithm designed by the present invention consists of a multilateral positioning module and an error state Kalman filter module. The multilateral positioning module uses an improved LM nonlinear optimization algorithm as the core to perform real-time positioning calculation on the UWB ranging signal, integrates the positioning calculation result with the accelerometer and gyroscope data of the IMU on the unmanned vehicle to estimate the position of the UWB tag, and outputs the real-time positioning result.

[0094] The LM nonlinear least squares optimization algorithm has the characteristics of fast convergence speed and convenient parameter adjustment, and is widely used in fields such as nonlinear optimization problems. Taking into account that the traditional LM nonlinear optimization algorithm relies on the selection of initial optimization parameters, the present invention improves the LM nonlinear optimization algorithm by introducing position feedback design.

[0095] For the i-th UWB base station and UWB tag, the positioning error is:

[0096]

[0097] Define the error vector:

[0098]

[0099] The Jacobian matrix is:

[0100]

[0101] Construct a least squares problem:

[0102]

[0103] Expand f(p) into the first order Taylor:

[0104] f(p+Δp)≈f(p)+J(p)Δp (7)

[0105] The descending vector is taken as:

[0106]

[0107] Among them, u k The radius of the trust optimization region, also known as the optimization step size.

[0108] Formula (8) can be simplified by taking the derivative and setting it to zero:

[0109] (J(p) T J(p)+u k I) Δp = -J(p) T f(p) (9)

[0110] Where I is the unit matrix, the new parameters are:

[0111] p new =p+Δp=p-[J(p) T J(p)+u k I] -1 J(p) T f(p) (10)

[0112] During the iteration process, when the new error vector sum of squares is smaller than the original one, the radius of the trust optimization area u is reduced. k , otherwise increase the radius u of the trust optimization area k . Set the tolerance ∈ and the maximum number of iterations max_iter. When the sum of squares of the error vector is less than the tolerance ∈ or the number of iterations is greater than the maximum number of iterations max_iter, the iteration stops.

[0113] The above is the calculation process of the LM nonlinear optimization algorithm for a positioning cycle. By feeding back the positioning point of the positioning cycle as the initial optimization point of the next positioning cycle, an improved LM nonlinear optimization algorithm is designed. The structure diagram of the algorithm is shown in Figure 6 .

[0114] This method does not rely on the selection of initial optimization point parameters. In the positioning scenario of the dynamic operation of the unmanned vehicle, the multilateral positioning algorithm will not fall into the wrong local optimal solution, causing the algorithm to be distorted.

[0115] The position solution problem in the present invention sets the initial step length u of the LM nonlinear optimization algorithm k =0.5, tolerance∈=1×10 -6 , the maximum number of iterations max_iter = 100, the multilateral positioning problem can be solved through the above process.

[0116] The accelerometer and gyroscope carried by the inertial measurement unit are used to measure the acceleration and angular velocity of the robot. As a lightweight sensor device, IMU is often carried on drones and unmanned vehicles and combined with other sensors for fusion positioning. Based on the multilateral positioning system based on the improved LM method, the UWB fusion positioning system is designed based on the error state Kalman filter by fusing the acceleration data and angular velocity data measured by IMU to further improve the positioning accuracy of the UWB positioning system.

[0117] In ESKF, the original state variable is usually called the nominal state variable, and the state variable in ESKF is called the error state variable. Let the state variable be:

[0118] x=[p,v,R,b g ,b a ,g] T (11)

[0119] Among them, p is displacement, v is velocity, R is rotation, b g is the gyroscope bias, b a is the accelerometer zero bias, and g is gravity.

[0120] The discrete-time equation of motion for the nominal state variables can be written as:

[0121]

[0122]

[0123]

[0124] b g (t+Δt)=b g (t) (15)

[0125] b a (t+Δt)=b a (t) (16)

[0126] g(t+Δt)=g(t) (17)

[0127] The discrete form of the error state variable is:

[0128] δp(t+Δt)=δp+δvΔt(18)

[0129]

[0130]

[0131] δb g (t+Δt)=δb g +η g (twenty one)

[0132] δb a (t+Δt)=δb a +η a (twenty two)

[0133] δg(t+Δt)=δg (23)

[0134] Write the above discrete form into matrix form:

[0135] δx(t+Δt)=Fδx+w,w~N(0,Q) (24)

[0136] Among them, w is Gaussian noise, and the covariance matrix Q is:

[0137] Q=diag(03,Cov(η v ),Cov(η θ ),Cov(η g ),Cov(η a ),03) (25)

[0138] The matrix F is:

[0139]

[0140] The prediction process of the ESKF includes the prediction of the nominal state based on the IMU integral and the prediction of the error state variable:

[0141] P pred =FPF T +Q (27)

[0142] Formula (27) is the prediction of covariance, where P pred is the predicted covariance matrix, and P is the corrected covariance matrix. Since the error state variable is reset to zero in each update iteration, the prediction of the error state variable is not considered in the design of the error state Kalman filter.

[0143] The observation equation is abstracted as h, v is the observation noise, z is the observation data, and V is the covariance matrix of the noise. Here the observation data is the three-dimensional positioning position estimated by the multilateral positioning method:

[0144] z=h(x)+v,v~N(0,V)(28)

[0145] Error state update process:

[0146] K=P pred H T (HP pred H T +V) -1 (29)

[0147] δx=K(zh(x pred )) (30)

[0148] x=x pred +δx (31)

[0149] P=(I-KH)P pred (32)

[0150] H is obtained by the chain rule:

[0151]

[0152]

[0153]

[0154] The data transmission frequency (100Hz) of the IMU sensor used in the experiment is greater than the data transmission frequency (50Hz) of the UWB radio module. In the engineering implementation, the IMU data is subjected to short-term trajectory calculation to update the robot's posture. The state variables at the next moment are obtained from the acceleration data and angular velocity data measured by the IMU. The trajectory calculation based on the Euler numerical integration method is as follows:

[0155]

[0156]

[0157]

[0158] Based on the above UWB positioning system design, UWB tags can be positioned with high precision in LOS scenarios.

[0159] The present invention designs a mean filter to suppress the high-frequency noise in the ranging signal of the UWB radio frequency module. The mean filter with a size of k is designed as follows:

[0160] Assume that the measurement value sequence Tn =(x n-k+1 ,...,x n ) 1×k , x n is the measured value at the current moment, and the mean filter kernel is:

[0161]

[0162] The filtered result of the nth distance measurement is:

[0163] x n =T n ·kenel (40)

[0164] The present invention adopts a kernel design mean filter with a size of 5 to filter UWB measurement data to suppress high-frequency noise in the UWB radio frequency module.

[0165] There is a multipath phenomenon in the data transmission between UWBs. The first path signal strength indication (FPSSI) refers to the signal strength of the directly received signal of the UWB signal transmission. In the LOS scenario, it is usually a straight-line propagation channel, and in the NLOS scenario, it is usually the channel with the shortest transmission time in the multi-channel. The total received signal strength indication (TRSSI) refers to the total signal strength of the UWB multi-channel channel. The Nooploop LinkTrack P-AUWB RF module can be used to obtain distance measurement information, first path signal strength indication and total received signal strength indication. According to experimental results, when a metal plate is used as an obstacle, the UWB RF module distance measurement signal produces a positive deviation relative to when there is no obstacle, and the first path signal strength indication and the total received signal strength indication are reduced. Based on this, the present invention trains a support vector machine to identify the NLOS channel.

[0166] Assume that the UWB ranging information obtained by the i-th group of measurements is The total received signal strength indicator is The first path signal strength indicator is Construct a training sample set:

[0167]

[0168] Among them, n is the number of samples, and the labels of the sample set are as follows:

[0169] y i ∈{-1,+1}(42)

[0170] In the case of LOS, y is +1, and in the case of NLOS, y is -1.

[0171] For each sample, let it be a vector in the sample space:

[0172]

[0173] The present invention uses the Gaussian kernel function:

[0174]

[0175] Where σ is the bandwidth of the Gaussian kernel, satisfying σ>0.

[0176] set up Represents the mapping of x to the feature vector in the high-dimensional feature space. The mapping relationship satisfies:

[0177]

[0178] Then the model corresponding to the hyperplane that divides the feature space is:

[0179]

[0180] In order to solve the parameters ω and b that can divide the hyperplane with the maximum margin, the following planning problem needs to be solved:

[0181]

[0182] The dual problem of this problem is:

[0183]

[0184] Substituting into the kernel function, the above problem is transformed into:

[0185]

[0186] The hyperplane obtained by solving is:

[0187]

[0188] The kernel function of the SVM model designed by the present invention is a Gaussian kernel function, and the Gaussian kernel bandwidth is σ=1. The structure of the sight distance identifier designed based on SVM is as follows: Figure 7 shown.

[0189] The present invention also takes into account the sensor measurement error of the UWB radio frequency module itself, and designs a sensor data corrector based on Gaussian process regression, which can effectively alleviate the large measurement error of the UWB radio frequency module, making the algorithm of the present invention more adaptable.

[0190] A Gaussian process is a set of any finite random variables that have a joint Gaussian distribution, and its properties are completely determined by the mean function and the covariance function. Gaussian process regression based on Gaussian process is a non-parametric learning method for supervised learning. For the training set D = {X, y}, where X is the n-dimensional input data and y is the corresponding output data, for the input x, its output satisfies the Gaussian process:

[0191] y~GP(m(x),k(x,x * ))(51)

[0192] The design of the Gaussian process regressor is to obtain the mean function m(x) and the covariance function k(x,x) based on the training sample set. * ).

[0193] It is difficult to use the common linear regression model to fit the regression relationship between the actual distance of the UWB RF module and the distance measurement signal and signal strength signal. Therefore, a Gaussian process regression model is constructed using the data set to correct the instrument measurement error.

[0194] Assume that the i-th group of sample data is The output is y i , assuming that the output data forms a high-dimensional vector y={y1,...,y n The prior distribution of} satisfies the multivariate Gaussian distribution:

[0195] y~(0,K(X,X)+σ n 2 I)(52)

[0196] Where I is the identity matrix, σ n 2 is the noise variance, and in the model of the present invention, σ n 2 =0.3, K is the covariance matrix determined by the Gaussian function, satisfying:

[0197]

[0198] Where n and m are X and X respectively. * The dimension of .

[0199] The Gaussian process regression model often uses the square exponential kernel function as the covariance function:

[0200]

[0201] Among them, l and is the hyperparameter of the kernel function.

[0202] In order to better preserve the data feature information and make the training Gaussian process regression model have a better mapping relationship with the data features, the present invention constructs an automatic relevance determination kernel (ARDK) based on the square exponential kernel function. Considering the three-dimensional input space, we have:

[0203]

[0204]

[0205] in, l1, l2, and l3 are the hyperparameters of the kernel function. By setting a hyperparameter for each dimensional input space, the Gaussian process regression model can retain more data features.

[0206] The output y of the training sample set and the output y of the test set * Satisfies the following joint prior Gaussian distribution:

[0207]

[0208] According to Bayes' theorem, the test set outputs y * The conditional distribution is:

[0209]

[0210]

[0211] cov(y * )=K(X * ,X * )-K(X,X * )[K(X,X)+σ n 2 I] -1 K(X,X * ) T (60)

[0212] The hyperparameters of the kernel function are solved using maximum likelihood estimation, that is, to obtain:

[0213]

[0214] Where K = K(X,X) + σ n 2 I.

[0215] The measured UWB distance measurement information, total received signal strength indication, and first path signal strength indication are used as training samples, and the measured real distance is used as the corresponding sample output label to construct a training set, solve the maximum relief estimation problem, and obtain the kernel function hyperparameters l1, l2, l3, and σ 2 f .

[0216] Since the present invention only considers the distance error problem and does not involve variance information, in order to reduce computing resources, for a set of input data to be predicted, only the mean prediction is considered. * , the mean of the Gaussian process regression output is taken as the output of the data, that is,

[0217] y * =K(X,x * ) T [K(X,X)+σ n 2 I] -1 y (62)

[0218] Let α=[K(X,X)+σ n 2 I] -1 y, α are n-dimensional column vectors and are independent of the input data. α can be calculated in advance to increase the calculation rate.

[0219] y * =K(X,x * ) T α (63)

[0220] The above simplification improves the calculation rate. The structure of the sensor data corrector based on Gaussian process regression is as follows: Figure 8 .

[0221] The UWB positioning system against NLOS scenarios is designed by combining the line-of-sight classification model, sensor data corrector and improved LM method combined with error Kalman filter obtained by the above training. Its structure is as follows: Fig. 9 .

[0222] The UWB positioning system uses the SVM line-of-sight classifier to identify the line-of-sight situation between UWB radio frequency modules in real time. If it is NLOS, the UWB distance measurement signal of the NLOS channel is removed from the UWB positioning algorithm; if it is LOS, the UWB distance measurement signal is retained and input into the UWB positioning module for position solution and estimation based on subsequent signal processing. The above process can eliminate the error introduced by the NLOS channel in the UWB positioning system.

[0223] The following comparative experiments are conducted to verify the performance of the algorithm.

[0224] First, only the UWB positioning system is used for positioning in the LOS scenario, and the handle is used to control the unmanned vehicle to move. The motion curve and position estimation are as follows: Fig.10 The calculated root mean square error is 0.2117m.

[0225] By using the NLOS-resistant UWB positioning system designed by the present invention, the same experiment is carried out to obtain the positioning result as shown in the figure, and the root mean square error is 0.13313m. Fig.11 The UWB dynamic positioning position curve and the real position curve in the LOS scenario are shown.

[0226] In the LOS scenario, since there are no obstacles, the SVM module does not work. The mean filter and sensor data corrector play a role in suppressing the measurement error of the UWB RF module, and the positioning accuracy is significantly improved.

[0227] By introducing a metal plate as an obstacle, the unmanned vehicle was controlled to conduct an experiment. Only the UWB positioning system was used for positioning in this NLOS scenario, and the handle was used to control the unmanned vehicle to move. The motion curve and position estimation are shown in the figure below. The calculated root mean square error is 0.5290m. Fig.12 The dynamic experimental positioning curve and the actual positioning curve of the UWB fusion positioning system in the NLOS scenario are shown.

[0228] By using the NLOS-resistant UWB positioning system designed by the present invention, the same experiment is carried out to obtain the positioning result as shown in the figure, and the root mean square error is 0.143822m. Fig.13 Shows the UWB dynamic positioning position curve and the real position curve in the NLOS scenario

[0229] The introduction of the line-of-sight classifier based on support vector machine eliminates the impact of the NLOS channel on the positioning system. At the same time, the UWB measurement data of the LOS channel is corrected in real time through the sensor data corrector based on Gaussian process regression. In actual experiments, the UWB positioning system still has a high positioning accuracy in the NLOS scenario, and the root mean square error of positioning is less than 15cm in dynamic experiments.

[0230] The present invention can well meet the positioning accuracy requirements of the UWB positioning system under the influence of obstacles:

[0231] 1. Lightweight UWB positioning solution, suitable for a variety of scenarios.

[0232] 2. Highly robust to the environment, no need for extensive offline training when changing obstacle scenes.

[0233] 3. Effectively suppress the impact of obstacles on the UWB RF module.

[0234] 4. Effectively suppress the measurement error of the sensor itself.

[0235] The present invention designs a NLOS-resistant UWB positioning system with the following functions:

[0236] 1. Low computational complexity, achieving high-precision positioning with low computing resources.

[0237] 2. Detect obstacles in real time and eliminate the impact of non-line-of-sight channels on the positioning system.

[0238] 3. Effectively suppress the measurement error of the sensor and have low requirements for the UWB RF module.

[0239] The technical advantages of the present invention are:

[0240] 1. A high-precision UWB positioning system is realized. By designing a sensor data corrector, the influence of the measurement error of the UWB radio frequency module itself on the UWB positioning system is suppressed. Therefore, in a line-of-sight scenario, the UWB positioning system of the present invention has higher positioning accuracy than the traditional UWB positioning system, and has lower requirements on the accuracy of the UWB radio frequency module, thereby reducing the equipment cost.

[0241] 2. Achieved UWB high-precision positioning in obstacle-blocked scenarios, suppressing the impact of obstacles on the UWB positioning system.

[0242] 3. The UWB positioning system designed by the present invention only needs the data of the impact of obstacles on the UWB radio frequency module. When the environment changes to a certain extent, the positioning system is still applicable.

[0243] 4. The UWB positioning system of the present invention does not need to model the scene, thus reducing computing resources.

[0244] The above content is a further detailed description of the present invention in combination with specific / preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, without departing from the concept of the present invention, it can also make several substitutions or modifications to these described embodiments, and these substitutions or modifications should be regarded as belonging to the protection scope of the present invention. In the description of this specification, the description of reference terms "an embodiment", "some embodiments", "preferred embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in combination with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily target the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In the absence of mutual contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples. Although the embodiments of the present invention and their advantages have been described in detail, it should be understood that various changes, substitutions and modifications can be made in the present invention without departing from the scope of protection of the patent application.

Claims

1. An ultra-wideband UWB real-time positioning method for obstacle-occluded scenes. It is characterized in that The following steps are involved: S1: Initialize the UWB positioning system, configure the parameters of the error state Kalman filter ESKF and the LM (Levenberg-Marquardt) nonlinear optimization algorithm, and prepare to fuse IMU data and UWB ranging data; S2: Start the UWB radio frequency module to collect the ranging signals between multiple UWB base stations and the UWB tags on the mobile platform to provide raw data for multilateral positioning; at the same time, use the IMU sensor to collect the acceleration and angular velocity data of the mobile platform to provide the motion model input for the state Kalman filter ESKF; S3: Combine the UWB ranging data and IMU data obtained in step S2, perform the UWB multi-lateration positioning process, start the LM algorithm to perform preliminary real-time positioning solution, and estimate the relative position error between the UWB base station and the UWB tag; S4: Construct a Jacobian matrix based on the LM algorithm, which contains the relationship between the measured distance from the UWB base station to the tag and the predicted position; solve the descent vector of the position parameter of the UWB tag relative to the observation error, and update the position parameter of the UWB tag by the least squares method, and prepare the observation data of the UWB tag for the prediction step of the state Kalman filter ESKF; S5: performing iterative optimization of the LM algorithm to solve the UWB multi-lateration positioning problem until a termination condition is met: preferably, the sum of squares of the error vector is less than a preset tolerance or a maximum number of iterations is reached; S6: Using the IMU data collected in step S2, the mobile platform posture information is updated by dead reckoning to provide the latest motion state for the state update step of the state Kalman filter ESKF; S7: Input the position parameters of the UWB tag optimized in step S5 as observation data into the state Kalman filter ESKF, and perform state prediction and error correction in combination with the motion state updated in step S6; S8: In the state Kalman filter ESKF, the position and velocity estimates of the mobile platform are updated based on the predicted and observed data, and the fusion is used to achieve robust processing of UWB ranging errors and IMU measurement noise; S9: Combining the result of the state Kalman filter ESKF fusion processing in step S8 and the position parameters optimized by the LM algorithm in step S5, the precise real-time position of the UWB tag is output.

2. The ultra-wideband (UWB) real-time positioning method according to claim 1, It is characterized in that The following steps are also included: Before step S1, an offline phase is performed to set training parameters, collect UWB data, and establish a data set according to the line-of-sight conditions; Use the dataset to train the SVM model for subsequent sight distance recognition; While collecting the UWB ranging signal in step S2, the trained SVM model is used to perform line-of-sight recognition on the signal, and the UWB signal is identified and classified as line-of-sight LOS or non-line-of-sight NLOS; If the SVM model determines that it is a non-line-of-sight NLOS situation, the corresponding UWB ranging signal is discarded and not used for subsequent positioning calculations; If the SVM model determines that it is a line-of-sight LOS situation, the UWB ranging signal is retained and input into the positioning solution process of steps S3 to S9 as valid data.

3. The ultra-wideband (UWB) real-time positioning method according to claim 1 or 2, It is characterized in that The following steps are also included: Before step S1, an offline phase processing is performed to set training parameters and establish a feature data set based on the collected UWB measurement data; The real distance annotation is performed on the UWB measurement data in the dataset for training the GPR model; Use the labeled data set to train the Gaussian process regression (GPR) model until the model parameters meet the preset requirements, preparing for the measurement error correction in the online stage; In step S3, after the collected UWB ranging signal is initially positioned and solved, the characteristics of the ranging signal are input into the sensor data corrector; The trained GPR model is used to correct the measurement error of the UWB RF module in real time, and the corrected accurate distance information is output; The corrected accurate distance information is used as observation data and input into the state Kalman filter ESKF together with the UWB tag position parameters optimized in step S5 for further position estimation and error correction.

4. The ultra-wideband (UWB) real-time positioning method according to any one of claims 1 to 3, It is characterized in that In step S3, the system modeling of the multilateration positioning is implemented by the following method, including: Determine a distance measurement method based on the TOA principle for calculating the straight-line distance between UWB radio frequency modules; wherein the UWB base station is a UWB radio frequency module with a known position, and the UWB tag is a UWB radio frequency module with a position to be estimated; Using the timestamps of the signal sent by the UWB base station and the signal received by the UWB tag, the distance between the two is calculated based on the known UWB signal wave speed; Based on the distance measurement from the UWB base station to the UWB tag, a multilateral positioning equation group is constructed for determining the position of the UWB tag; The multilateral positioning equations are transformed into a linear equation system, and the coefficient matrix and constant vector are constructed to numerically solve the position of the UWB tag; The linear equations are solved by numerical methods to obtain the three-dimensional position coordinates of the UWB tag.

5. The ultra-wideband (UWB) real-time positioning method according to any one of claims 1 to 4, It is characterized in that In step S3: in the process of performing UWB multilateration positioning, the error between each UWB base station and the UWB tag is calculated by defining a positioning error function, which is based on the predicted position and the actual measured distance; and an error vector is further constructed to summarize the positioning errors between all base stations and tags; In step S4: on the basis of constructing the Jacobian matrix, using the partial derivative information of the positioning error function contained in the matrix with respect to the predicted position parameter, a least squares problem is constructed and solved, and the error function is approximated by a first-order Taylor expansion to solve the descent vector of the position parameter; In step S5: during the iterative optimization process, the optimization step size is adjusted using the trust region method, and the position parameter is improved according to the sum of squares of the error vector; and after each iteration, the radius of the trust optimization region is adjusted according to the new sum of squares of the error vector until the termination condition is met; In steps S3 to S5: a position feedback mechanism is introduced, and the solution of the previous positioning cycle is used as the initial optimization point of the next cycle during the iteration process to improve the insensitivity of the algorithm to the initial parameter selection.

6. The ultra-wideband (UWB) real-time positioning method according to any one of claims 1 to 5, It is characterized in that In step S2: while starting the UWB radio frequency module to collect the ranging signal, the IMU sensor is synchronously started to collect the acceleration and angular velocity data of the mobile platform, which will be used for subsequent track calculation and state prediction; using the IMU data, track calculation is performed based on the Euler numerical integration method to determine the initial displacement and velocity state variables of the mobile platform at the next moment; In step S6: using the IMU data collected in step S2, track calculation is performed based on the Euler numerical integration method to determine the posture change of the mobile platform at the next moment; the high-frequency data of the IMU sensor is fused with the low-frequency data of the UWB radio frequency module, and the posture information of the mobile platform is updated through short-time track calculation, and an accurate initial state estimation is provided for the error state Kalman filter; considering that the transmission frequency of the IMU data is higher than the transmission frequency of the UWB data, the posture change of the mobile platform during the UWB data update interval is estimated through track calculation, and the state variables are updated; In step S7: in the error state Kalman filter, the IMU data and the UWB ranging data are combined to perform fusion processing of the state variables.

7. The ultra-wideband (UWB) real-time positioning method according to any one of claims 1 to 6, It is characterized in that Before using the collected UWB ranging signal for preliminary positioning in step S3, the UWB ranging signal is firstly filtered by a mean filter, and the output filtering result is used as the input data of the LM algorithm and the error state Kalman filter ESKF; Preferably, the mean filter has a kernel of a set size for smoothing a continuous sequence of measured values; each element in the filter kernel has an equal weight to ensure uniformity of the filtering process; and a series of filtered continuous measured values ​​are obtained by convolving the measured value sequence with the filter kernel and calculating the filtering result of each measured value point in the sequence in turn.

8. The ultra-wideband (UWB) real-time positioning method according to any one of claims 1 to 7, It is characterized in that In step S3: in the process of performing UWB multilateration positioning, the distance measurement information, the first path signal strength indication FPSSI and the total received signal strength indication TRSSI collected by the Nooploop LinkTrack P-AUWB radio frequency module are analyzed to identify whether there is a multipath phenomenon in the UWB signal transmission by analyzing the changes in these signal strengths; Based on the support vector machine (SVM) model, a training sample set is constructed using the collected UWB ranging information and signal strength indication. Each sample in the sample set corresponds to a label, indicating line-of-sight (LOS) or non-line-of-sight (NLOS) conditions. The UWB signal is classified by training the SVM model to identify whether the current UWB signal is in LOS or NLOS condition. In step S4: if the SVM model determines that the current UWB signal is in NLOS condition, the signal is discarded and not used in subsequent positioning calculations to reduce the impact of multipath effects on positioning accuracy; if it is determined to be in LOS condition, the signal is retained and input as valid data into the subsequent positioning solution process; In the offline stage, the UWB measurement data and the corresponding real distance are used to establish a training set, and the Gaussian process regression (GPR) model is used for training. The model uses the square exponential kernel function as the covariance function and uses the automatic correlation kernel function ARDK to better retain the data characteristics. In step S6: In the online stage, the trained GPR model is used to correct the measurement error of the UWB RF module in real time, and the corrected accurate distance information is output. Combined with the optimized UWB tag position parameters, it is input into the state Kalman filter ESKF for further position estimation and error correction.

9. An ultra-wideband UWB positioning system for obstacle-occluded scenes. It is characterized in that The system includes: A plurality of UWB radio frequency modules, including a plurality of UWB base stations distributed at predetermined locations and a UWB tag carried on a mobile platform, for measuring the distance between the plurality of UWB base stations and the UWB tag; Inertial measurement unit (IMU), mounted on the mobile platform, is used to provide information on movement speed and direction to assist UWB positioning; A mobile platform, as a mobile carrier, carries the UWB tag and performs actual movement; A main control machine communicates with the plurality of UWB radio frequency modules and the inertial measurement unit IMU, and is used to execute the ultra-wideband UWB real-time positioning method according to any one of claims 1 to 8.

10. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, the ultra-wideband (UWB) real-time positioning method according to any one of claims 1 to 8 is implemented.

Images