SVM-based NLOS real-time identification method and system

Abstract

The invention provides an SVM (Support Vector Machine)-based NLOS (Non-Line Of Sight) real-time identification method and system, which are used for improving the positioning accuracy in an ultra wide band (UWB) system, and the method comprises the following steps: collecting signal data from UWB equipment, and extracting key features for identifying non-line-of-sight propagation, such as measured distance, received signal strength and first-path signal strength; preprocessing the extracted features, wherein the preprocessing comprises normalization, abnormal value elimination and smoothing; constructing a Lagrange function and carrying out optimization solution to obtain an optimal hyperplane parameter; mapping the data to a high-dimensional space by using a Gaussian kernel function to realize linear separability; training an SVM classifier and determining a decision function; and classifying the real-time data by using the trained model, and judging whether the real-time data is LOS or NLOS. The method and the system can autonomously learn signal characteristics, adapt to variable environments, meet the real-time requirement through lightweight design, and are effectively applied to a UWB system to solve the positioning problem in a barrier environment.

Description

Technical Field

[0001] The present invention relates to a simultaneous localization and mapping (SLAM) technology, and in particular to a NLOS real-time recognition method and system based on SVM. Background Art

[0002] Simultaneous localization and mapping (SLAM) is a complex technology that plays a vital role in the field of indoor autonomous mobile robots and unmanned navigation. It requires the system to be able to build a map in an unknown environment while accurately determining its real-time position. Currently, lidar systems and camera vision systems are common types of external sensors used for SLAM. However, on the one hand, the frequency range used by lidar and camera light sources is close to the visible light frequency range in the electromagnetic spectrum. Therefore, in challenging scenes such as those filled with smoke or containing a large number of mirrors, the signal-to-noise ratio of these optical measurement devices is relatively low, which makes SLAM algorithms that rely on such devices prone to failure and significantly reduces the functional performance of autonomous systems. On the other hand, when achieving relative positioning, both lidar and cameras are difficult to avoid the phenomenon of measurement errors gradually accumulating over time. Even slight initial measurement deviations will continue to amplify as the operation time increases, resulting in a decrease in positioning accuracy. Therefore, after a long period of continuous operation, the position data obtained by these sensors is likely to be inaccurate to a large extent.

[0003] In recent years, in view of the stability of radio frequency (RF) signals under harsh and extreme environmental conditions, there have been many studies exploring the use of RF bands instead of traditional dedicated sensors for SLAM. Due to the frequency characteristics of RF signals and the stability of anchor-tag systems based on absolute position positioning, robot positioning based on ultra-wideband (UWB) systems has been widely studied. However, in complex indoor environments, UWB signals will be interfered by obstacles, resulting in obstructed signal propagation paths, which in turn will cause ranging errors and seriously affect positioning accuracy. Therefore, how to accurately identify non-line-of-sight (NLOS) signals in real time is very important to ensure the accuracy of UWB absolute positioning in indoor environments with obstacles.

[0004] NLOS identification based on a neural network (NN) model, which only uses data from low-cost UWB devices, can be applied to different environments through the neural network model to mitigate the impact of NLOS ranging measurements. The model returns the probability that the ranging measurement is LOS based on the UWB data extracted from the UWB tag carried by the robot and input into the neural network model. The advantage of this method is that it has high environmental adaptability and can be adjusted and optimized according to the specific conditions of different scenarios, so that it can perform effectively in different environments and significantly reduce the ranging error caused by NLOS. Given that robot systems usually tend to be lightweight in design and have relatively low requirements for algorithm complexity, convolutional neural networks are not conducive to the construction of real-time systems.

[0005] The NLOS identification method based on the channel model uses the channel impulse response (CIR) metric to distinguish between non-line-of-sight and line-of-sight states. CIR is a key indicator that directly reflects the characteristics of the wireless channel. This method focuses on the detailed analysis of the channel impulse response (CIR). Through in-depth exploration of multiple CIR metrics, delay spread, multipath distribution characteristics, and energy delay distribution, it is possible to manually extract the signature features that distinguish between NLOS and LOS states. These features may include more dispersed multipath distribution, longer delay spread, and more complex energy distribution patterns in the NLOS environment, and the identification and classification of NLOS are completed according to the characteristics of NLOS. This method is limited by the prior knowledge of the signal processing channel model, has a time delay problem, is difficult to determine a suitable distribution function, and requires additional constraints.

[0006] With the increasing demand for indoor positioning services in production and life, the current accuracy improvement of non-line-of-sight (NLOS) positioning technology has reached a very high level. However, on the one hand, given that robot systems usually tend to be lightweight in design, the complexity of the algorithm is relatively low, so it is necessary to deploy a set of lightweight algorithm solutions that can not only meet actual needs but also efficiently adapt to such systems. This algorithm must ensure high recognition accuracy and high robustness; on the other hand, NLOS recognition is only the first step in UWB positioning in complex environments. In order to make full use of NLOS information, it is necessary to consider how to use the NLOS recognition results to correct the system and apply it in actual scenarios.

[0007] Therefore, in complex environments, the NLOS recognition system needs to meet the following requirements: 1. Widely applicable, it can complete the NLOS recognition task for different channel characteristics. 2. The NLOS recognition accuracy is high enough to effectively assist the system to make accurate corrections in multi-obstacle environments. 3. It has a lightweight design and can be deployed in robots or embedded devices to meet real-time requirements. 4. The system can be specifically applied to the actual scenario of UWB positioning in the NLOS environment.

[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 background technology and provide a NLOS real-time recognition method and system based on SVM.

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

[0011] In a first aspect of the present invention, a NLOS real-time recognition method based on SVM comprises the following steps:

[0012] S1. Extract key features for identifying non-line-of-sight propagation from signal data collected by UWB devices;

[0013] S2. Preprocessing the extracted features;

[0014] S3. Use the preprocessed data to construct and train the SVM classifier model, including: constructing a Lagrangian function, and obtaining hyperplane parameters for optimal data partitioning by optimizing and solving the Lagrangian function; using a Gaussian kernel function to map the preprocessed data to a high-dimensional feature space, and obtaining a linearly separable hyperplane in the high dimension; using the determined hyperplane parameters and support vectors to train the SVM classifier, and determining a decision function for classification based on the weights and support vectors obtained through training;

[0015] S4. Use the trained SVM classifier model to classify the real-time signal data into line-of-sight LOS or non-line-of-sight NLOS, thereby obtaining the NLOS recognition result.

[0016] In a second aspect of the present invention, a NLOS real-time recognition system based on SVM comprises:

[0017] The UWB positioning system includes a plurality of UWB base stations and at least one UWB tag mounted on a mobile carrier, which is used to measure the distance between the base station and the tag and provide raw positioning data;

[0018] An IMU sensor is mounted on the mobile carrier and is used to collect motion data of the mobile carrier to assist UWB positioning;

[0019] Mobile carrier, used to carry UWB tags and IMU sensors and move in the environment;

[0020] The processing module communicates with the UWB base station and the UWB tag, and is used to execute the SVM-based NLOS real-time recognition method.

[0021] In a third aspect of the present invention, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the SVM-based NLOS real-time identification method is implemented.

[0022] In a fourth aspect of the present invention, a computer program product includes a computer program, and when the computer program is executed by a processor, the SVM-based NLOS real-time identification method is implemented.

[0023] The present invention has the following beneficial effects:

[0024] The present invention provides a non-line-of-sight (NLOS) real-time identification method and system based on a support vector machine (SVM), which has the ability to autonomously learn signal characteristics without relying on additional channel constraints, so that it can flexibly adapt to variable obstacles and channel environments, thereby accurately completing the NLOS identification task. The system adopts a targeted data preprocessing strategy, so that it can operate stably in complex environments and can accurately distinguish the NLOS signal status. In addition, the scheme design achieves lightweight and low complexity, making it easy to integrate into a robot platform and ensure real-time response performance. By integrating NLOS identification information into the UWB positioning system, the present invention effectively solves the positioning problem caused by obstacles and improves the positioning accuracy and reliability in complex environments. In general, the present invention provides a high-precision and high-robustness solution for indoor positioning services through innovative SVM classifiers and real-time signal processing technology, which is particularly suitable for application scenarios that require stable and accurate positioning in environments with many obstacles.

[0025] Other beneficial effects of the embodiments of the present invention will be further described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a framework diagram of a NLOS real-time recognition algorithm based on SVM according to an embodiment of the present invention.

[0027] Figure 2 The figure is a schematic diagram of the recognition effect of UWB in a NLOS non-line-of-sight environment according to an embodiment of the present invention.

[0028] Figure 3 It is a schematic diagram of the recognition effect of UWB in the LOS line-of-sight environment according to an embodiment of the present invention.

[0029] Figure 4 This is a test result diagram of NLOS real-time recognition based on SVM according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] 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.

[0031] 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.

[0032] See also Figure 1 The embodiment of the present invention provides a NLOS real-time recognition method based on SVM, comprising the following steps:

[0033] Step S1. Feature extraction: extract key features for identifying non-line-of-sight propagation from signal data collected by the UWB device.

[0034] Specifically, the key features include the measured distance (distanced), the received signal strength (RxRssiRx) and the first signal strength (FpRssiFp).

[0035] In some embodiments, the observation data is derived from UWB signals collected from a UWB tag attached to a mobile robot. At each timestamp, data is extracted from the UWB signal in real time, including the measured distance, received signal strength, and first path signal strength, which is used as a criterion for identifying non-line-of-sight propagation.

[0036] Step S2. Data preprocessing: preprocess the extracted features.

[0037] In a preferred embodiment, the pre-processing comprises:

[0038] Normalization: Use the Z-score standardization method to normalize the feature values ​​and convert the numerical range of the data features into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0039] Outlier elimination: Use the outlier elimination method based on the 3σ rule to operate in each time window of the UWB data. For observations in the UWB time window where the deviation between the measured value and the average value of the data in the window exceeds three standard deviations, identify them as outliers and eliminate them;

[0040] Smoothing: Smoothing the data using a weighted moving average (WMA) gives relatively higher weights to recent data points and relatively lower weights to data points further away in time to highlight underlying trends or patterns in the signal.

[0041] Step S3. SVM classifier model construction and training: Use the preprocessed data to perform the following operations, including: constructing a Lagrangian function, and obtaining the hyperplane parameters for optimal data partitioning by optimizing and solving the Lagrangian function; using a Gaussian kernel function to map the preprocessed data into a high-dimensional feature space, and obtaining a linearly separable hyperplane in high dimensions; using the determined hyperplane parameters and support vectors to train the SVM classifier, and determining the decision function for classification based on the weights and support vectors obtained through training.

[0042] In a preferred embodiment, step S3 specifically includes: determining an objective function that aims to minimize the square of the modulus of the weight vector and the weighted sum of all misclassified sample penalty terms; introducing slack variables to allow certain data points to violate the classification rules, thereby improving the generalization ability of the model; using regularization parameters to balance the relationship between model complexity and classification error; constructing a Lagrangian function and solving it to obtain the parameters of the highest hyperplane that optimally divides the data; based on Lagrangian duality, converting the original optimization problem into a dual problem, and calculating the Lagrangian multiplier by solving the dual problem; the Lagrangian multiplier determines the importance and influence of each training sample in determining the final decision boundary; using the calculated Lagrangian multiplier and support vector to determine the final decision boundary, which can maximize the distinction between data points of different categories.

[0043] Furthermore, a Gaussian kernel function is used to map the preprocessed data and transform the data from the original feature space to a high-dimensional feature space in order to find a linear hyperplane in the new space that distinguishes different categories. The Gaussian kernel function is selected based on the characteristic Gaussian distribution characteristics of UWB data to optimize the classification performance of the data in the new feature space.

[0044] Step S4: Use the trained SVM classifier model to classify the real-time signal data and determine whether it is line-of-sight LOS or non-line-of-sight NLOS, thereby obtaining an NLOS recognition result.

[0045] In some embodiments, the classification operation of the SVM classifier specifically includes: using the hyperplane parameters determined during the training process to classify the real-time signal data through a decision function; using the support vector and the corresponding weights to calculate the output score of the test sample; based on the output score of the test sample, determine whether it is LOS or NLOS, if the output score is greater than or equal to zero, it is determined to be LOS; if the output score is less than zero, it is determined to be NLOS.

[0046] In a further embodiment, the SVM-based NLOS real-time identification method further includes: applying the NLOS identification result of the SVM classifier to the UWB positioning system, specifically including: using the weighted least squares method WLS for positioning, assigning different weights to each ranging data point according to the NLOS identification result, optimizing the positioning algorithm, and improving the positioning accuracy; wherein, a relatively low weight is given to the data point that is relatively greatly affected by NLOS, and a relatively high weight is given to the LOS data point. Optimization techniques such as the Newton-Gauss method or the Levenberg-Marquardt method can be used to minimize the WLS problem and solve the optimal positioning result.

[0047] The embodiment of the present invention also provides a NLOS real-time identification system based on SVM, including: a UWB positioning system, including multiple UWB base stations and at least one UWB tag mounted on a mobile carrier, used to measure the distance between the base station and the tag and provide original positioning data; an IMU sensor, mounted on the mobile carrier, used to collect motion data of the mobile carrier to assist UWB positioning; a mobile carrier, used to carry the UWB tag and the IMU sensor and move in an environment; a processing module, communicating with the UWB base station and the UWB tag, used to execute the NLOS real-time identification method based on SVM.

[0048] The present invention provides a real-time non-line-of-sight NLOS recognition method and system based on support vector machine (SVM), which can accurately extract key features from signal data collected by ultra-wideband (UWB) devices and use them to identify non-line-of-sight propagation conditions. The system can adapt to variable obstacles and channel environments by autonomously learning signal features without the need for additional channel constraints, thereby accurately completing the NLOS recognition task. In addition, the system adopts a targeted data preprocessing strategy, which not only improves the system stability in complex environments, but also ensures the accurate distinction of NLOS signal states.

[0049] In the SVM classifier model construction and training steps of the present invention, the hyperplane parameters for optimal data partitioning are obtained by constructing and optimizing the Lagrangian function, and the Gaussian kernel function is used to map the data into a high-dimensional feature space to achieve a linearly separable hyperplane, thereby improving the accuracy of classification. The system design has the advantages of lightweight and low complexity, making it easy to integrate into robots or other embedded devices and meet real-time requirements.

[0050] In addition, the present invention applies NLOS identification information to the UWB system, which can effectively handle positioning problems in obstacle environments and improve positioning accuracy and reliability in complex environments. This method is particularly suitable for indoor autonomous mobile robots and unmanned navigation fields, and solves the problems of low signal-to-noise ratio and accumulated measurement errors of traditional sensors such as lidar and cameras in harsh environments. By accurately identifying NLOS signals in real time, this system can provide strong support for SLAM technology that relies on precise positioning. In short, the present invention provides a high-precision and high-robustness solution for indoor positioning services through its innovative signal processing and machine learning technologies.

[0051] The following further describes specific examples and experimental verification of the present invention.

[0052] System composition

[0053] Software system composition

[0054] The software system runs on the ubuntu18.04 (Linux operating system mainly for desktop applications) operating system and ROS-melodic (open source robot operating system) software environment.

[0055] ROS Melodic greatly simplifies the development, testing and deployment of robot software with its powerful modular architecture, rich software package collection and flexible programming interface. It provides a series of plug-and-play tools and libraries including sensor data processing, motion control, path planning, visual recognition, etc., allowing developers to quickly build and iterate complex robot functions. Through seamless integration with Ubuntu 18.04, ROS Melodic can make full use of the resource management, network support and graphical interface provided by the operating system to create a highly integrated and efficient development environment for robot researchers and engineers.

[0056] Hardware system composition

[0057] The experiment uses a four-wheeled mobile robot RIKEBOT. RIKEBOT is a wheeled robot equipped with four omnidirectional Mecanum wheels, a LeTMC-520RGBD depth camera, a RP-LiDAR 2D laser radar, a NoopLoop:LinkTrack UWB device, an MPU6050 IMU sensor, and computing hardware. JetsonNano TM , with a quad-core ARM processor and 16GB of memory. On the robot, a bracket is installed to ensure that there are no obstacles in the robot's own structure that block the UWB signal. A tag is installed on the robot's bracket at the same height as the four base stations in the environment.

[0058] UWB

[0059] Nooploop's LinkTrack UWB is an advanced positioning system based on ultra-wideband (UWB) technology. It provides installation support on a variety of operating systems including Ubuntu, allowing users to quickly build and configure the system according to actual needs. In addition to hardware devices, Nooploop may also provide supporting software tools and algorithms to help users achieve higher-level positioning services, such as trajectory tracking, environmental mapping and data analysis.

[0060] Visualization Platform

[0061] The mapping algorithm uses RVIZ as the visualization platform. RVIZ is a powerful 3D visualization tool in the Robot Operating System (ROS), which is mainly used to display and debug various data and status information in the robot system. It is an open platform built on the Qt framework, which aims to provide users with an intuitive way to observe and understand the dynamic changes of robots and their environment. RVIZ is an indispensable part of the ROS ecosystem. It greatly facilitates the visualization of complex information in the process of robot development, testing and maintenance, and is a powerful auxiliary tool for robot developers and researchers.

[0062] Range-SLAM is based on the UWB mapping algorithm framework and runs on the Ubuntu 1 8.04 operating system and ROS-melodic software environment. It is compiled and run on the main control machine of the hardware control system. Figure 2 The function includes constructing the surrounding environment, refining the obstacle information, and completing the environmental positioning under obstacles.

[0063] NLOS Identification Algorithm

[0064] A real-time NLOS recognition method based on SVM mainly includes the following parts: feature extraction, extracting key features for subsequent processing; preprocessing of UWB data to improve the accuracy of NLOS recognition; building an SVM classifier to classify whether it is a NLOS signal; and specific applications after NLOS recognition.

[0065] 1. Feature Extraction

[0066] The observation data is derived from UWB signals collected from UWB tags attached to mobile robots. At each timestamp, data is extracted from the UWB signals in real time, and the data consists of the measured distance (distanced), received signal strength (RxRssiRx) and first path signal strength (FpRssiFp), which are used as criteria for identifying non-line-of-sight propagation.

[0067] In order to mitigate the impact of outliers, the UWB window is introduced for data preprocessing to enhance the robustness of the system. The UWB window maintains the most recent N W The window is updated when the system observes new UWB data, and the update takes into account the time series of the UWB data.

[0068] 2. Data Preprocessing

[0069] The data within the UWB window will be preprocessed for the subsequent LOS / NLOS identification, which can be divided into three parts: normalization, outlier removal and smoothing.

[0070] (1) Normalization:

[0071] Data normalization standardizes the numerical range of features, narrows the gap between different attributes, and prevents the imbalance of weights of certain features in SVM due to inconsistent scales. Use Z-score normalization to convert feature values ​​into a standard normal distribution with a mean of 0 and a standard deviation of 1. To simplify the expression, use [*] to represent d, Rx, and Fp respectively:

[0072]

[0073] in and Respectively represent the training UWB data set D train Mean and covariance of d, Rx and Fp.

[0074] (2) Outlier removal:

[0075] Outliers or abnormal data points can lead to overfitting or underfitting in the model. The presence of outliers may cause the non-line-of-sight recognition algorithm to fail, leading to potential failure of the mapping task, which in turn affects the overall functionality of the system. By properly excluding such abnormal data, noise interference can be reduced, allowing the model to better learn the general patterns inherent in the data.

[0076] UWB windows are utilized to perform anomaly handling tasks following the 3σ rule, which involves identifying observations that deviate significantly by three standard deviations from the mean, indicating potential anomalies in the data.

[0077]

[0078] in and They represent the mean and covariance of d, Rx and Fp in the UWB window respectively.

[0079] UWB data exhibits a certain degree of continuity in time series. Within the same UWB time window, the principle of data consistency must be observed to ensure that the data within the window conforms to a certain inherent pattern or expected trend, thereby effectively reducing the occurrence of outliers and improving the accuracy of data processing and analysis.

[0080] (3) Smoothing

[0081] Smoothing aims to remove noise, mutations or irregular fluctuations in data, thereby revealing potential underlying trends or patterns in the data. The smoothing process is crucial in data preprocessing and helps improve the prediction accuracy and stability of the model.

[0082] The Weighted Moving Average method (WMA) is used within the UWB window. In this method, higher weights are assigned to the most recent data points, while lower weights are assigned to data points that are farther away in time. The formula is as follows:

[0083]

[0084]

[0085] where k i represents the weight assigned to the ith lagged data point, represents the data point value at time t minus n within the UWB window, Represents the weighted moving average of the UWB window.

[0086] The result after applying the weighted moving average method WMA t As expected, it is then fused with the UWB data of the current frame to achieve the goal of smoothing. This approach effectively highlights the latest trends in the data series and provides more sensitive and accurate predicted NLOS identification results when processing UWB data characterized by significant temporal trends or transient effects.

[0087] 3. SVM Model

[0088] The essence of identifying LOS / NLOS conditions based on signal strength and distance information constitutes a classification problem. A support vector machine (SVM) classifier is used to solve this problem. The specific process is as follows:

[0089] First, during the training process, the main goal of SVM is to find an optimal decision boundary, which is usually a hyperplane. By maximizing the boundary, that is, increasing the distance between the boundary and each category of data points as much as possible, data points belonging to different categories are effectively separated. The objective function is as follows:

[0090]

[0091]

[0092] where ω=[ω1,ω2,ω3,b] T The vector represents the weights used for training, and N represents the number of elements in the training set, which is consistent with the number of support vectors. is a regularization process, which is the product of the penalty factor C multiplied by the total penalty for all misclassified samples. In this framework, e i Serves as a slack variable that quantifies how much sample i deviates from the ideal boundary separation, which needs to be optimized. In addition, C represents the regularization parameter, which is a hyperparameter that needs to be tuned to determine an appropriate setting. This tuning aims to achieve a balance between optimal model performance and ensuring strong generalization capabilities for non-line-of-sight propagation identification. (i) refers to the output labels for non-line-of-sight and line-of-sight conditions in the i samples manually annotated during the training process, where a value of 1 indicates line-of-sight conditions and a value of -1 indicates non-line-of-sight conditions, X (i) =[d (i) ,Fp (i) ,Rx (i) ,1] T Indicates the distance, signal strength and y (i) A vector of 1's associated with the labels.

[0093] The objective function formula deals with an optimization problem. In the solution process, the key step is to construct the Lagrangian function. By solving this Lagrangian function, the parameters of the highest hyperplane that optimally divides the data are obtained. The specific Lagrangian formula derivation result is shown in the following formula:

[0094]

[0095]

[0096] In mathematics, the symbol α i represents the Lagrange multiplier corresponding to each training sample, which reveals the relative importance of each data point in constructing the classification hyperplane. For those training samples that form support vectors, their respective Lagrange multipliers α i The value of actually determines the degree of their influence on the shape and location of the final decision boundary. Therefore, it can be said that αi The weights used as support vectors reflect the contribution of these samples to determining the optimal hyperplane in the process.

[0097] Afterwards, based on Lagrange duality, the original problem is transformed into its dual problem, and the Lagrange multiplier α is derived by solving the dual problem. By derivation, an equivalent equation can be established as shown below

[0098]

[0099]

[0100] Using kernel function techniques, the original data is mapped to a high-dimensional feature space, and a linearly separable hyperplane is sought in high dimensions. Given the characteristic Gaussian distribution of UWB data, the Gaussian kernel function is used. After obtaining the training weights and support vectors, in the prediction stage, given the preprocessed samples to be classified To classify, it is converted to Then x test is input into the trained SVM model. The sample is mapped into a high-dimensional feature space through the kernel function. Subsequently, the category to which the sample belongs is determined based on the position of the hyperplane in this space. The specific prediction formula is shown below.

[0101]

[0102] where x (i) and (i) represents training samples, also called support vectors; α i Refers to the weights just trained. The result y test represents the output score obtained from the test set by SVM. If y test ≥0, the classification output is LOS; otherwise, the classification result is NLOS.

[0103] IV. NLOS Information Application

[0104] In this section, a positioning method based on weighted least squares (WLS) will be introduced to mitigate the adverse effects of non-line-of-sight NLOS data. Its characteristic is that different weights are assigned to each ranging data point according to its NLOS recognition result, so that when solving the positioning problem, a lower weight is given to the data that is more affected by NLOS, while a higher weight is given to the line-of-sight LOS data. This processing method is expected to make the positioning algorithm more adaptable and robust in the face of complex and changeable non-line-of-sight environments, thereby significantly improving the positioning accuracy.

[0105] Considering the NLOS recognition results obtained by the support vector machine SVM model, the following WLS problem is constructed:

[0106]

[0107]

[0108] Where βi is the weight of the ranging measurement, which is related to the fraction of NLOS ranging measurements calculated by the SVM model, and λ is a constant coefficient. Minimization is achieved through optimization techniques such as the Newton-Gauss method or the Levenberg-Marquardt method.

[0109] Algorithm 1: NLOS identification system pseudo code

[0110]

[0111] Experimental Results

[0112] To ensure that UWB can accurately identify NLOS signals in a normal indoor environment, UWB signal strength data from 1.5m to 6m is collected as a training set. After training in MATLAB, the SVM support vector and weights are obtained and then imported into the ROS system. Then, the UWB NLOS environment test is carried out. The recognition effect of UWB in the NLOS non-line-of-sight environment is shown in the figure. Figure 2 As shown in Figure 2, the recognition effect of UWB in LOS line-of-sight environment is as follows: Figure 3 As shown, the NLOS accuracy in the test was 93.7%.

[0113] Compared with the traditional technology, the main advantages of the present invention are:

[0114] 1. Autonomous learning of signal characteristics: The system can autonomously learn signal characteristics without the need for additional channel constraints, allowing the system to flexibly adapt to changing obstacles and channel environments and accurately complete NLOS identification tasks.

[0115] 2. Targeted data preprocessing: By adopting targeted data preprocessing strategies, the system can operate stably under complex conditions and accurately identify whether it is a NLOS signal, thereby ensuring the accuracy and robustness of recognition.

[0116] 3. Lightweight design: The system design focuses on lightweight and low complexity, making it not only easy to integrate into robots or other embedded devices, but also able to meet real-time requirements and suitable for dynamic and real-time application scenarios.

[0117] 4. Improve positioning accuracy and reliability: Applying NLOS identification information to the UWB system effectively solves the positioning problem in obstacle environments and improves positioning accuracy and reliability in complex environments.

[0118] In summary, the present invention provides a NLOS recognition solution that can maintain high accuracy and high efficiency in complex environments, and is particularly suitable for indoor positioning services, robot navigation, and unmanned driving.

[0119] An embodiment of the present invention further provides a storage medium for storing a computer program, which at least performs the above method when executed.

[0120] An embodiment of the present invention further provides a control device, comprising a processor and a storage medium for storing a computer program; wherein the processor is configured to execute at least the method described above when executing the computer program.

[0121] An embodiment of the present invention further provides a processor, wherein the processor executes a computer program and at least executes the method described above.

[0122] The storage medium can be implemented by any type of non-volatile storage device, or a combination thereof. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The storage medium described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0123] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0124] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0125] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0126] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, etc. Various media that can store program codes.

[0127] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0128] The methods disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0129] The features disclosed in several product embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new product embodiments.

[0130] The features disclosed in several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0131] The above contents are further detailed descriptions 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 those skilled in the art of the present invention, several equivalent substitutions or obvious variations can be made without departing from the concept of the present invention, and the performance or use is the same, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. A NLOS real-time recognition method based on SVM, characterized in that: The following steps are involved: S1. Extract key features for identifying non-line-of-sight propagation from signal data collected by UWB devices; S2. Preprocessing the extracted features; S3. Use the preprocessed data to construct and train the SVM classifier model, including: constructing a Lagrangian function, and obtaining hyperplane parameters for optimal data partitioning by optimizing and solving the Lagrangian function; using a Gaussian kernel function to map the preprocessed data to a high-dimensional feature space, and obtaining a linearly separable hyperplane in the high dimension; using the determined hyperplane parameters and support vectors to train the SVM classifier, and determining a decision function for classification based on the weights and support vectors obtained through training; S4. Use the trained SVM classifier model to classify the real-time signal data into line-of-sight LOS or non-line-of-sight NLOS, thereby obtaining the NLOS recognition result.

2. The NLOS real-time identification method based on SVM according to claim 1, characterized in that: In step S2, the preprocessing includes: The Z-score standardization method is used to normalize the feature values, and the numerical range of the data features is uniformly converted into a standard normal distribution with a mean of 0 and a standard deviation of 1; The outlier elimination method based on the 3σ rule is used to operate in each time window of the UWB data. For the observations whose deviations between the measured values ​​and the mean value of the data in the window exceed three standard deviations in the UWB time window, they are identified as outliers and eliminated. The data is smoothed using a weighted moving average (WMA), which gives relatively higher weights to recent data points and relatively lower weights to data points further away in time to highlight underlying trends or patterns in the signal.

3. The NLOS real-time recognition method based on SVM according to claim 1 or 2, characterized in that: In step S3, the Lagrangian function is constructed, and the hyperplane parameters for optimally partitioning the data are obtained by optimizing and solving the Lagrangian function, which specifically includes: Determine the objective function, which aims to minimize the square of the modulus of the weight vector and the weighted sum of the penalty terms of all misclassified samples; Introducing slack variables to allow some data points to violate classification rules, thereby improving the generalization ability of the model; Use regularization parameters to balance the relationship between model complexity and classification error; Construct the Lagrangian function and solve it to obtain the parameters of the highest hyperplane that optimally divides the data; Based on Lagrangian duality, the original optimization problem is transformed into a dual problem, and the Lagrangian multiplier is calculated by solving the dual problem; the Lagrangian multiplier determines the importance and influence of each training sample in determining the final decision boundary; Using the calculated Lagrange multipliers and support vectors, the final decision boundary is determined, through which the data points of different categories are maximally distinguished.

4. The SVM-based NLOS real-time identification method according to any one of claims 1 to 3, characterized in that: In step S3, the preprocessed data is mapped to a high-dimensional feature space using a Gaussian kernel function to obtain a linearly separable hyperplane in the high dimension, specifically including: The preprocessed data is mapped using a Gaussian kernel function to transform the data from the original feature space to a high-dimensional feature space in order to find a linear hyperplane that distinguishes different categories in the new space; A Gaussian kernel function is selected based on the characteristic Gaussian distribution characteristics of UWB data to optimize the classification performance of the data in the new feature space.

5. The NLOS real-time identification method based on SVM according to any one of claims 1 to 4, characterized in that: In step S4, the classification operation of the SVM classifier specifically includes: Using the hyperplane parameters determined during the training process, the real-time signal data is classified through the decision function; Calculate the output score of the test sample using the support vector and the corresponding weights; According to the output score of the test sample, it is judged as LOS or NLOS. If the output score is greater than or equal to zero, it is judged as LOS; if the output score is less than zero, it is judged as NLOS.

6. The SVM-based NLOS real-time identification method according to any one of claims 1 to 5, characterized in that: Also includes: The NLOS recognition results of the SVM classifier are applied to the UWB positioning system, including: The weighted least squares method (WLS) is used for positioning. Different weights are assigned to each ranging data point according to the NLOS identification result to improve the positioning accuracy. Among them, relatively low weights are given to data points that are relatively greatly affected by NLOS, and relatively high weights are given to LOS data points.

7. The SVM-based NLOS real-time identification method according to claim 6, characterized in that: The Newton-Gauss method or the Levenberg-Marquardt method is used to minimize the WLS problem and obtain the optimal positioning result.

8. A NLOS real-time recognition system based on SVM, characterized in that: include: The UWB positioning system includes a plurality of UWB base stations and at least one UWB tag mounted on a mobile carrier, which is used to measure the distance between the base station and the tag and provide raw positioning data; An IMU sensor is mounted on the mobile carrier and is used to collect motion data of the mobile carrier to assist UWB positioning; Mobile carrier, used to carry UWB tags and IMU sensors and move in the environment; A processing module communicates with a UWB base station and a UWB tag, and is used to execute the SVM-based NLOS real-time identification method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the SVM-based NLOS real-time identification method according to any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the SVM-based NLOS real-time identification method according to any one of claims 1 to 7 is implemented.

Images