Indoor Integrated Positioning System Based on Distributed Extended Kalman Filter Deep Learning Optimization Algorithm

Abstract

This invention relates to an indoor integrated positioning system based on a distributed extended Kalman filter (DEKF) deep learning optimization algorithm, belonging to the field of indoor positioning. The system introduces a neural network, applying the CNN-LSTM deep learning method to UWB NLOS / LOS signal classification. Then, a distributed extended Kalman filter (DEKF) is introduced to fuse the CNN-LSTM-optimized UWB data with IMU data, suppressing the cumulative error caused by IMU drift over time, thus ensuring the accuracy of ranging data while preventing cumulative errors and time synchronization issues in the fusion of data from various sensors. The proposed CNN-LSTM-DEKF algorithm can automatically adapt to different sensor characteristics and environmental changes, providing more accurate and robust positioning results in complex indoor environments. Experimental results show that in indoor environments with significant multipath effects and signal obstruction gaps, the method of this invention significantly improves positioning accuracy.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of indoor positioning, and particularly relates to an indoor comprehensive positioning system based on a distributed extended Kalman filtering deep learning optimization algorithm. BACKGROUND

[0002] With the development of Internet of Things (IoT) and smart spaces, indoor positioning is in urgent demand in the fields of industrial production, medical care, Internet of Vehicles logistics and smart home, etc. The challenges faced by indoor positioning are much more complex than those in outdoor environments, including multipath effects of signals, non-line-of-sight (NLOS) propagation, environmental interference, etc. [1]. Traditional single positioning technologies, such as Ultra-Wideband (UWB) [2], Inertial Measurement Unit (IMU) [3], Wi-Fi [4], Bluetooth (BLE) [5], etc., cannot provide accurate and stable positioning effects in complex indoor environments. Therefore, multi-sensor fusion has gradually become an effective method to realize indoor positioning, which can make up for the shortcomings of single technology by finally fusing multiple sensor data. The UWB / IMU combined positioning method has attracted widespread attention due to its robustness and high accuracy in indoor environments, and the combination of UWB and IMU has obvious advantages in multi-sensor fusion. In the Ultra-Wideband (UWB) positioning system, commonly used ranging methods include Time of Arrival (TOA), Time Difference of Arrival (TDOA), Angle of Arrival (AOA), etc. [8], which can provide centimeter-level positioning accuracy. However, due to the non-line-of-sight (NLoS) (NLoS) in complex indoor environments, the accuracy of the positioning system is greatly affected, and the positioning system cannot provide accurate positioning results in the presence of NLoS. In order to solve the problem of NLoS in the UWB positioning system, the UWB / IMU combined positioning method is proposed, which can effectively solve the problem of NLoS in the UWB positioning system and provide accurate positioning results in the presence of NLoS. Figure 1) and multipath effect interference, there will be large distance positioning errors in UWB sensor data from time to time, and the ranging accuracy may be significantly reduced, affecting the system positioning accuracy. In recent years, with the rapid development of deep neural networks, multi-layer neural networks in deep learning have been widely used in the identification of LOS and NLOS propagation conditions. Liu et al. proposed a method that uses reversible transform denoising combined with convolutional neural network (CNN) to identify NLOS signals, with an identification accuracy of 81.68%[9]. Fang et al. used the original CIR as input and trained and tested it through a long short-term memory (LSTM) model, with an identification accuracy of 82.14%

[10] . Inertial measurement unit (IMU) has been widely studied and applied due to its low cost and environmental immunity. IMU uses three-axis gyroscopes and three-axis accelerometers as basic sensitive elements to provide real-time angular velocity and acceleration measurements, thereby obtaining the motion characteristics and attitude information of the carrier (fusion measurement data to calculate the attitude, velocity and position information of the target). However, due to the long-term drift of inertial sensors, the error of the calculated navigation state (position, velocity and attitude) will accumulate over time, so it cannot provide reliable long-term positioning information. UWB provides distance measurement of position when positioning, which is not easily affected by indoor environment, and is suitable for solving accurate positioning in most static scenes; while IMU can provide motion direction and acceleration information, which is suitable for real-time tracking of user motion trajectory. The complementarity of these two sensors makes the positioning fusion scheme based on UWB and IMU an ideal solution in complex environments. On this basis, [6] proposed a UWB / IMU combined positioning algorithm based on machine learning, which combines UWB and IMU separately to take advantage of each other and improve the system positioning accuracy. [7] proposed a fusion framework integrating IMU with multiple UWB tags based on a graph optimization model, which can provide high-precision position and attitude information for the system. Despite this, the fusion of UWB and IMU also faces some challenges. IMU is susceptible to cumulative accuracy (such as bias and bias), which eventually leads to cumulative positioning accuracy over time, while the data accuracy of UWB is affected by non-line-of-sight (NLoS) and multipath effects. To address these issues, Kalman Filter (KF) and its extended optimization version, Distributed Extended Kalman Filter (DEKF), are widely used in multi-sensor fusion, especially in dynamic positioning systems. Kalman filter is a recursive algorithm that updates the system's state estimate by combining measurement data and prediction models, thereby reducing the impact of measurement. Shi Jian-tao et al. proposed a new motion estimation algorithm that can reduce the drift of IMU sensors. This algorithm uses a Kalman filter to estimate and fuse UWB measurement data, and fuses the drift-free position obtained by the UWB positioning Kalman filter with the position calculated by the IMU

[11] .However, the performance of KF in nonlinear scenarios is limited, so the Extended Kalman Filter (EKF) is proposed to solve nonlinear problems. Yue Xiaodong et al. used the IMU / UWB loose combination precise positioning system to solve the precise position, velocity and attitude information of the robot, in which the positioning coordinates from the UWB base station (EKF) corrected the UWB error in real time

[12] . The distributed extended Kalman filter (DEKF) enhances the noise resistance and robustness of the system by dispersing the state estimation of different data sources into multiple nodes. Although DEKF can improve the positioning accuracy in theory, in practical applications, the accuracy of data is still easily affected by environmental fluctuations, sensor ranging accuracy and other factors. The performance of Kalman filter and its extended methods depends on the accuracy of the system model. In practical applications, the inconsistency of noise and dynamic environmental changes often lead to model deviation, thereby affecting the simulation performance. To solve this problem, in recent years, the adaptive Kalman filter (AKF) has been introduced, which adjusts the noise covariance matrix in real time to improve the auxiliary robustness. Liu et al. proposed an adaptive complementary Kalman filter to filter UWB and IMU data, track the error of variables such as position, velocity and direction, which can not only eliminate the multipath effect caused by shielding of UWB signals, but also correct the velocity and acceleration deviation caused by IMU drift

[13] . In the above discussion, people have done a lot of research on the fusion of different sensors for indoor positioning, and the commonly used method is to use the extended Kalman filter and the nonlinear filtering method evolved on its basis, which simplifies the nonlinear system into a local linear system, thereby realizing recursive estimation. However, these nonlinear filtering methods do not consider the influence of sensor fusion on the performance of the sensor itself, and the performance of sensor fusion is also crucial to improve positioning. Based on the analysis of the above related work, most UWB / IMU fusion methods usually ignore the data synchronization problem in the sensor fusion process. SUMMARY

[0003] The purpose of the present application is to overcome the defects existing in the prior art, further improve the robustness of the multi-sensor fusion positioning system in complex environments, and provide an indoor comprehensive positioning system based on a distributed extended Kalman filter deep learning optimization algorithm, and propose an optimization algorithm CNN-LSTM-DEKF that fuses convolutional long short-term memory network CNN-LSTM and distributed extended Kalman filter DEKF. The optimization algorithm performs non-line-of-sight identification on UWB ranging data and cumulative error suppression on IMU ranging data to ensure the positioning accuracy of the fusion positioning system. At the same time, by dispersing the state estimation of multiple nodes, the sensor data in the fusion process will not have data synchronization problems between various data sources, further improving the robustness and positioning accuracy of the sensor fusion positioning system.

[0004] To achieve the above object, the technical scheme of the present application is: an indoor comprehensive positioning system based on a distributed extended Kalman filtering deep learning optimization algorithm, comprising:

[0005] An ultra-wideband (UWB) module is configured to collect UWB ranging data, including channel impulse response (CIR) data of the UWB;

[0006] An inertial measurement unit (IMU) module is configured to collect IMU ranging data;

[0007] A distributed extended Kalman filtering deep learning optimization algorithm module is configured to propose an optimization algorithm CNN-LSTM-DEKF that fuses a convolutional long short-term memory network (CNN-LSTM) and a distributed extended Kalman filter (DEKF). The CNN-LSTM is configured to perform feature extraction and classification prediction on the CIR data of the UWB. The DEKF is configured to fuse the classified and predicted data with the IMU ranging data, suppress the cumulative error caused by the time drift of the IMU, and thus ensure the accuracy of the ranging data and prevent the cumulative error and time synchronization problem in the data fusion of the sensors.

[0008] In an embodiment of the present application, the network framework of the CNN-LSTM includes two basic components: a convolutional neural network (CNN) and a long short-term memory (LSTM). In the CNN-LSTM, the CIR data is used as the input vector of the CNN. The CNN learns the frequency characteristics of the CIR data to extract features using the non-time structure of the input signal. Then, the output of the CNN is input into the LSTM to learn the dependency of the time series data for non-line-of-sight (NLOS) signal and line-of-sight (LOS) signal classification and prediction.

[0009] In an embodiment of the present application, the CNN-LSTM is composed of a convolutional layer, a max-pooling layer, a fully connected layer, an LSTM layer, and a fully connected layer, and the generalization ability is improved through Dropout regularization.

[0010] In an embodiment of the present application, the CNN-LSTM performs feature extraction and classification prediction on the CIR data of the UWB, which includes the following stages:

[0011] The CNN component includes:

[0012] The CNN processes the original CIR data to extract spatial features that capture the changes in signal patterns caused by NLOS conditions. Multiple convolutional layers are used, and each convolutional layer uses the same activation function to detect the input data for features. To ensure the nonlinearity of the activation function, ReLU is used, and its equation is as follows:

[0013]

[0014] x represents the parameter of the ReLU function;

[0015] After the convolutional layer outputs to the pooling layer, the pooling layer reduces the dimension by down-sampling and prevents overfitting; then, the pooling layer outputs input to the fully connected layer, and the fully connected layer selects the classifier as the output through the SoftMax function;

[0016] LSTM component:

[0017] The output of the CNN is passed to the LSTM, which captures the time dependence in the CIR data sequence, and the gating architecture of the LSTM enables it to model long-term relationships in the data;

[0018] After the data extracted by the CNN is input to the LSTM, the LSTM is trained using a labeled data set with a cross-entropy loss function to ensure accurate binary classification of UWB signals as LOS or NLOS.

[0019] In an embodiment of the present application, the gating architecture of the LSTM includes an input, a forgetting, and an output gate.

[0020] In an embodiment of the present application, the input and output gate functions in the gating architecture of the LSTM are described as follows:

[0021] i t = sigma (W i · [h t-1 , x t ] + b i )

[0022]

[0023] o t = sigma (W o · [h t-1 , x t ] + b o )

[0024] h t = o t · tanh (C t )

[0025] In the formula, i t is the input gate, which determines whether the spatial features extracted by the CNN, i.e., the classification-related information including NLOS and LOS, are used to update the time series state; h t-1 represents the hidden state of the previous time step, which contains the dynamic information of the feature sequence at the past time step; x t represents the feature vector output by the CNN at the current time step, which contains the UWB CIR spatial features at the current time step; W i, b i represent the weight and bias parameters of the input gate, W o , b o represent the weight and bias parameters of the output gate, which are optimized by gradient descent during the training process to adapt the influence of different input features; is the candidate state, which is used to generate the new candidate state at the current time step, capturing the potential features of the current time step CNN features and historical hidden states, retaining important information related to LOS / NLOS; tanh represents the hyperbolic tangent activation function, which compresses the candidate state to the range of (-1, 1) for representing possible state updates; W C , b C represent the weight and bias parameters of the candidate state, which can effectively represent the nonlinear relationship of the time series after optimization; o t is the output gate, which extracts the most relevant information for LOS / NLOS classification from the candidate state and outputs to h t ; h t represents the hidden state at the current time step, which is the output of the LSTM and also one of the inputs for the next time step; tanh(C t ) represents a nonlinear transformation of the current time series state, keeping its range within (-1, 1).

[0026] In an embodiment of the present application, the DEKF calculation process is as follows:

[0027] Step one, local prediction stage

[0028] Each node i independently runs the prediction step, updating its state estimate based on the local model:

[0029]

[0030] is the predicted state estimate of node i at the current time step k compared to the previous time step k-1; represents the predicted value of the state of node i at time k-1; u k-1,i represents the control input of node i at time k-1; P k∣k-1,i is the predicted error covariance matrix of node i at the current time step k compared to the previous time step k-1, F k,i represents the state transition function of node i at time k; P k-1∣k-1,i represents the predicted error covariance matrix of node i at time k-1; represents the state transition matrix obtained by differentiating the state transition function; Q k,i represents the system noise covariance matrix, describing the uncertainty of state transition;

[0031] Step two, local update stage

[0032] Each node updates its local state based on the observed values:

[0033]

[0034] P k∣k,i =(IK k,i H k,i )P k∣k-1,i

[0035] K k,i H is the Kalman gain of node i, used to balance the impact of predictions and observations; k,i The Jacobian matrix representing the observation function; z k,i h represents the measurement value of node i; i R represents a nonlinear observation function that describes the mapping relationship from state to observation; k,i It represents the observation noise covariance, describing the uncertainty of sensor measurements.

[0036] Step 3, Information Sharing Phase: Each node shares local state estimates and covariance information through communication, forming a global information set. The shared content includes... P k∣k,i K k,i ;

[0037] Step 4, Global Fusion Stage: The optimal state estimate of the entire system is obtained through global information fusion. The weighted average method is then used to sum the estimates of each node based on the weights of the covariance matrix.

[0038]

[0039] Let W represent the optimal state estimate after global fusion. k,i This represents the weight of the i-th node. This represents the local state estimate of the i-th node;

[0040] Step 5, Global Update Feedback Phase: The fusion results are fed back to each node, and the local state is adjusted to maintain global consistency.

[0041] In one embodiment of the present invention, DEKF is verified by MAE and RMSE.

[0042] In one embodiment of the present invention, the ultra-wideband (UWB) module collects distance measurement data of anchor points through an ultra-wideband (UWB) tag.

[0043] In one embodiment of the present invention, the inertial measurement unit (IMU) module acquires ranging data through a triaxial accelerometer.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The application provides an optimization algorithm fusing CNN-LSTM and DEKF, which extracts features and makes classification prediction on UWB CIR data through CNN-LSTM, and fuses the data with IMU data through DEKF, thereby reducing the influence of non-line-of-sight and multipath effect on UWB and suppressing the time accumulation error of IMU, and significantly improving the accuracy of indoor positioning.

[0046] 2. Compared with the previous research, the CNN-LSTM-DEKF algorithm provided by the application not only considers the ranging error caused by the influence of complex environment on the sensor, but also considers the error accumulation and data synchronization between various data sources in the sensor fusion process, thereby effectively improving the noise resistance and robustness of the sensor fusion positioning system. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 Examples of NLoS and LoS propagation in UWB IPS.

[0048] Figure 2 System structure of the CNN-LSTM-DEKF optimization algorithm of the UWB and IMU fusion positioning system.

[0049] Figure 3 Three deep learning architectures for UWB channel identification, wherein (a) is CNN, (b) is LSTM, and (c) is CNN-LSTM.

[0050] Figure 4 Confusion matrix for classification performance on various line-of-sight obstacles (Sce.1 to Sce.5 correspond to table, human body, glass, door and wall, respectively, wherein (a) is DT, (b) is SVM, (c) is MLP, (d) is CNN, (e) is LSTM, and (f) is CNN-LSTM.

[0051] Figure 5 Real path and original data and motion trajectory estimated by four different methods.

[0052] Figure 6 Error box plot of five different optimization algorithms, wherein (a) is root mean square error, and (b) is absolute mean error. DETAILED DESCRIPTION

[0053] The technical solutions of the application will be specifically described below with reference to the drawings.

[0054] It should be pointed out that the following detailed description is exemplary and is intended to provide further description of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0055] It is to be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments consistent with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, devices, components and / or combinations thereof, but do not preclude the presence or addition of one or more other features, steps, operations, devices, components and / or combinations thereof.

[0056] The application provides an indoor comprehensive positioning system based on a distributed extended Kalman filtering deep learning optimization algorithm, comprising:

[0057] An ultra-wideband (UWB) module is configured to collect UWB ranging data, including channel impulse response (CIR) data of the UWB;

[0058] An inertial measurement unit (IMU) module is configured to collect IMU ranging data;

[0059] A distributed extended Kalman filtering deep learning optimization algorithm module is configured to propose an optimization algorithm CNN-LSTM-DEKF that fuses a convolutional long short-term memory network (CNN-LSTM) and a distributed extended Kalman filter (DEKF). The CNN-LSTM is configured to extract features and make a classification prediction on the CIR data of the UWB. The DEKF is configured to fuse the classified and predicted data with the IMU ranging data, suppresses the cumulative error caused by the time drift of the IMU, and thus ensures the ranging data accuracy and prevents the cumulative error and time synchronization problem in the data fusion of the sensors.

[0060] The following is a specific implementation process of the application.

[0061] As shown in Figure 2 , the embodiment provides an indoor comprehensive positioning system based on a distributed extended Kalman filtering deep learning optimization algorithm, comprising:

[0062] An ultra-wideband (UWB) module is configured to collect UWB ranging data, including channel impulse response (CIR) data of the UWB; as shown in Figure 2 , in this example, the UWB module collects ranging data of anchor points 1, 2, 3 and 4 based on an UWB tag;

[0063] An inertial measurement unit (IMU) module is configured to collect IMU ranging data; as shown in Figure 2 , in this example, the IMU module collects ranging data through a three-axis acceleration sensor;

[0064] Based on the distributed extended Kalman filter deep learning optimization algorithm module, an optimization algorithm CNN-LSTM-DEKF is proposed by fusing convolutional long short-term memory network CNN-LSTM and distributed extended Kalman filter DEKF, wherein the CNN-LSTM is used for feature extraction and classification prediction of channel impulse response CIR data of ultra-wideband UWB (such as Figure 2 As shown in the figure, the CNN-LSTM is used for UWB NLOS / LOS signal classification based on the channel impulse response CIR data of ultra-wideband UWB, and then the DEKF is used for fusing the classified and predicted data with inertial measurement unit IMU ranging data, so as to suppress the cumulative error caused by the time drift of the IMU, thereby ensuring the ranging data accuracy and preventing the cumulative error and time synchronization problem of the sensor data fusion, and finally realizing indoor positioning; the proposed CNN-LSTM-DEKF algorithm can automatically adapt to different sensor characteristics and environmental changes, and provide more accurate and robust positioning results in a complex indoor environment. Experimental results show that in an indoor environment with significant multipath effect and signal shielding gap, the method significantly improves the positioning accuracy.

[0065] 1. CNN-LSTM optimization model

[0066] Figure 3 is three kinds of deep learning architecture for UWB channel identification. The optimization algorithm framework used in the application is as shown in Figure 3 (c), which contains two basic components CNN and LSTM. In the CNN-LSTM, the original CIR data is used as the input vector of the CNN, the CNN learns the frequency characteristics of the CIR to utilize the non-time structure of the input signal for feature extraction, and then the output of the CNN is input into the LSTM for learning the dependency of the time series data for NLOS and LOS signal classification and prediction.

[0067] Figure 3 The running process of the CNN-LSTM model framework in (c) in processing UWB CIR data includes the following stages:

[0068] CNN component:

[0069] The convolutional neural network (CNN) processes the original UWB CIR data to extract spatial features that capture the signal pattern changes caused by NLOS conditions. Multiple convolutional layers are used, and each convolutional layer uses the same activation function to detect the input data for features. In order to ensure the nonlinearity of the activation function, ReLU (Rectified Linear Unit)

[16] is used, and its equation is as follows:

[0070]

[0071] Here, x represents the parameter of the ReLU function.

[0072] After the convolutional layer outputs to the pooling layer, the pooling layer reduces the dimension by down-sampling and prevents overfitting. Then, the pooling layer outputs are input to the fully connected layer, which selects a classifier as the output by the SoftMax function.

[0073] LSTM component:

[0074] The output of the CNN is passed to a long short-term memory (LSTM) network, which captures the temporal dependencies in the CIR sequence. The gating architecture of the LSTM (input, forget, and output gates) enables it to model long-term relationships in the data, and the “input gate” and “output gate” functions

[17]

[18] are described as follows:

[0075] i t = σ(W i · [h t-1 , x t ] + b i ) (3-2)

[0076]

[0077] o t = σ(W o · [h t-1 , x t ] + b o ) (3-4)

[0078] h t = o t · tanh(C t ) (3-5)

[0079] where i t is the input gate, which decides whether the spatial features extracted by the CNN, i.e., the classification-related information including NLOS and LOS, are used to update the time series state; h t-1 represents the hidden state of the previous time step, containing the dynamic information of the feature sequence at the past time steps; x t represents the feature vector output by the CNN at the current time step, containing the spatial features of the UWB CIR at the current time step; W i and b i represent the weight and bias parameters of the input gate, and W o and b o represent the weight and bias parameters of the output gate, which are optimized by gradient descent during the training process to adapt to the influence of different input features. For candidate state, the current time step new candidate state captures the potential features of the current time step CNN features and historical hidden state, retaining important information related to LOS / NLOS; tanh represents the hyperbolic tangent activation function, which compresses the candidate state to the range of (-1, 1) for representing possible state updates; W C , b C represent the weight and bias parameters of the candidate state, which can effectively represent the nonlinear relationship of the time series after optimization; o t is the output gate, which extracts the most relevant information for LOS / NLOS classification from the candidate state and outputs to h t ; h t represents the hidden state of the current time step, which is the output of the LSTM and also one of the inputs of the next time step; tanh(C t ) represents the nonlinear transformation of the current time series state, which keeps its range in (-1, 1).

[0080] After the data extracted by the CNN feature is input into the LSTM layer, the LSTM layer is trained using a labeled data set with a cross-entropy loss function to ensure that the UWB signal is accurately classified as LOS or NLOS. The optimized CNN-LSTM model significantly reduces the UWB positioning error by improving the signal classification accuracy, providing a robust preprocessing step for subsequent sensor fusion.

[0081] The basic technology used in this part of the training sample collection is to use sensor data for position solution.

[0082] 2. DEKF optimization based on CNN-LSTM

[0083] EKF is a classic filtering method for handling state estimation problems of linear and nonlinear dynamic systems

[19] . It linearizes the nonlinear system by first-order Taylor expansion, combines the prediction model of the system with the sensor measurement value, and recursively calculates the state estimation. DEKF is a distributed version of EKF, which distributes the state estimation task to multiple nodes to optimize the state estimation in a distributed manner. DEKF not only can handle the fusion problem of multiple sensors, but also can significantly improve the robustness and anti-noise performance of the system. Therefore, the UWB ranging data and IMU data fusion after training can be reasonably corrected and set through DEKF to avoid the occurrence of ranging outliers. And verified by MAE and RMSE. Distributed Extended Kalman Filter (DEKF) calculation process.

[0084] Step 1: Local Prediction Phase

[0085] Each node i runs the prediction step independently, updating its state estimate based on the local model:

[0086]

[0087] Here The predicted state estimate of node i relative to the previous time k-1 and the current time k; This represents the predicted state of node i at time k-1; u k-1,i P represents the control input of node i at time k-1; k∣k-1,i Let F be the prediction error covariance matrix of node i compared to the previous time k-1 and the current time k. k,i P represents the state transition function (Jacobi matrix) of node i at time k; k-1∣k-1,i Let represent the prediction error covariance matrix of node i at time k-1; Q represents the state transition matrix (Jacobi matrix) obtained by differentiating the state transition function; k,i This represents the system noise covariance matrix, describing the uncertainty of state transitions.

[0088] Step Two: Local Update Phase

[0089] Each node updates its local state based on the observed values:

[0090]

[0091] P k∣k,i =(IK k,i H k,i )P k∣k-1,i

[0092] Here K k,i H is the Kalman gain of node i, used to balance the impact of predictions and observations; k,i The Jacobian matrix representing the observation function; z k,i h represents the measurement value of node i; i R represents a nonlinear observation function that describes the mapping relationship from state to observation; k,i It represents the observation noise covariance, describing the uncertainty of sensor measurements.

[0093] Step 3: Information Sharing Phase. Each node shares local state estimates and covariance information through communication, forming a global information set. The shared content includes... P k∣k,i K k,i wait.

[0094] Step four: Global Fusion Phase, the optimal state estimation of the whole system is obtained by global information fusion, and the weighted average method is used to weight and sum the node estimates according to the weight of the covariance matrix:

[0095]

[0096] Here, represents the optimal state estimation after global fusion, W k,i represents the weight of the i-th node, represents the local state estimation of the i-th node.

[0097] Step five: Global Feedback Phase, the fusion results are fed back to each node to adjust the local state to maintain global consistency.

[0098] According to the above iterative process, DEKF reduces the computational burden of centralized methods through distributed computing, while enhancing the robustness of the system. Single node failure will not cause the entire system to fail. Local computing speeds up state updating and improves system response speed. It can adapt to heterogeneous sensor data, especially suitable for multi-sensor fusion systems such as UWB / IMU. DEKF can effectively utilize the data fusion of multiple UWB base stations, reducing the influence of NLOS and multipath effects. Through distributed nodes, the drift error of the IMU is corrected in real time. DEKF can integrate multi-source data in complex indoor environments and improve positioning accuracy and system robustness. The contribution of DEKF in indoor positioning lies in its strong fusion ability and distributed architecture, especially in complex environments and densely distributed nodes, which can significantly improve system performance.

[0099] In summary, the CNN-LSTM-DEKF algorithm provides a comprehensive solution for UWB / IMU-based indoor positioning, solving key challenges in sensor fusion while providing higher accuracy and robustness in complex indoor environments.

[0100] Experiment and data analysis

[0101] 1. Non-line-of-sight recognition performance evaluation of CNN-LSTM optimization algorithm

[0102] To evaluate the performance of the CNN-LSTM algorithm in UWB CIR data signal classification, the present invention conducts comparative experiments with public data sets and benchmark algorithms to evaluate the accuracy of various models in UWB LOS / NLOS signal classification, verifying the advantages of the CNN-LSTM algorithm.

[0103] The experimental uses a public UWB LOS / NLOS dataset

[20] . Specifically, the dataset covers 7 indoor scenarios (Office 1, Office 2, Small Apartment, Small Workshop, Kitchen with Living Room, Bedroom, and Boiler Room), which can avoid the bias between various models caused by the construction of specific locations. The dataset contains 42,000 samples (3,000 LOS and 3,000 NLOS), each containing 1,016 CIR raw data with a resolution of 1 nanosecond. The dataset is divided into a training dataset (33,600), a validation dataset (4,200), and a test dataset (4,200), and the 42,000 sample dataset is assigned in a random order, with an input vector dimension of (1x1016), directly input into the model for training. By selecting 5 different training models as benchmarks for comparison, the 5 training models are DT (decision tree), SVM (support vector machine), MLP (multilayer perception network), CNN (convolutional neural network), and LSTM (long short-term memory network). Among them, the CNN-LSTM (convolutional long short-term memory network) model used by the application is composed of a convolutional layer, a maximum pooling layer, a fully connected layer, an LSTM layer, and a fully connected layer, and the generalization ability is improved through Dropout regularization. All experiments are processed on a computer equipped with an Intel Core i5-10500H CPU, 16 GB RAM, and an NVIDIA GeForce RTX 3050Ti Laptop GPU.

[0104] To evaluate the classification performance of various methods in complex indoor environments for various line-of-sight obstructions, the confusion matrix as shown in Table 1 is used to reflect the classification performance of various methods in 5 kinds of (Sce.1 to Sce.5 correspond to table, human body, glass, door and wall respectively) shielding scene recognition. Figure 4 Both the CNN-LSTM and LSTM classification algorithms can achieve NLOS recognition of UWB to some extent, but in some scenes where the attenuation characteristics are relatively ambiguous, the two algorithms perform poorly and are difficult to distinguish in the recognition scene of the human body (Sce.2). Compared with other obstructions, human body obstructions interfere the most with NLOS recognition of UWB, and the misclassification rate of human body obstruction recognition is more than 10% for the five methods, with poor classification effect. For SVM and CNN algorithms, the error is more obvious, and the misclassification error rate of Sce.2 is more than 1 / 3. Overall, the CNN-LSTM algorithm has Figure 4 (f)It can be seen that it has good performance in the classification of various line-of-sight obstructions in the UWB non-line-of-sight scene. Compared with the other several algorithms with generally low recognition rate in the Sce.2 scene, the CNN-LSTM algorithm still achieves a recognition accuracy of no less than 89%, and the overall average misclassification rate is only 4.38%.

[0105] The performance of various models in NLoS identification is tested on the public dataset, and the performance comparison of NLOS obstruction identification is shown in Table 1. In terms of algorithm identification accuracy, precision and recall, the CNN-LSTM algorithm has higher advantages compared with other identification algorithms.

[0106] Table 1 Performance comparison of various algorithms for NLOS obstruction identification

[0107]

[0108] 2. Performance evaluation of fusion optimization algorithm CNN-LSTM-DEKF

[0109] In order to evaluate the performance of the CNN-LSTM-DEKF optimization algorithm proposed in the present application, the positioning effect of the original data and four different optimization algorithms (EKF, DEKF, CNN-LSTM, and CNN-STM-DEKF) is compared through simulation experiments. In the simulation experiment, it is assumed that the coordinates of the four base stations are (0, 0), (0, 50), (50, 0), and (50, 50) m, and the target to be measured moves in a uniform acceleration straight line in a two-dimensional plane, with an initial position of (20, 20) m, an acceleration of 0.01 m / s, an initial velocity of 0.5 m / s, a sampling period of 1 s, and a motion path of a broken line path and a square path. In order to simulate the interference of the external environment on the positioning accuracy of the UWB module, Gaussian white noise is added to the simulated trajectory data, with a mean of 0.06 and a variance of 0.1. At the same time, an IMU error model is set in the simulation framework, and the sampling data of the IMU module is fused into the motion trajectory at a sampling frequency of 100 Hz. The original data and the motion trajectory optimized by the four algorithms under two different real paths are shown in Figs. Figure 5 (a) and (b). Among them, the green solid line trajectory represents the real path, the red dot represents the unoptimized original data, the blue, pink, and cyan solid lines respectively represent the motion trajectories optimized by EKF, DEKF, and CNN-LSTM, and the black dashed line trajectory is the motion trajectory optimized by CNN-LSTM-DEKF.

[0110] From Figure 5As can be seen from (a) and (b), with the acceleration of the target to be measured, the oscillation amplitudes of the four algorithms will increase, and there will be greater deviation from the actual position. The oscillation of the DEKF and EKF algorithms is more obvious than that of the CNN-LSTM and CNN-LSTM-DEKF, indicating that the positioning accuracy is more affected by the errors of the IMU and UWB, and the trajectory accuracy of the CNN-LSTM and CNN-LSTM-DEKF is higher and closer to the true coordinates. At the same time, the DEKF and the CNN-LSTM fusion cost the proposed CNN-LSTM-DEKF optimization algorithm, and the range of abnormal values is much less than that of DEKF, because the combination of the CNN-LSTM optimization algorithm and the DEKF filtering algorithm can ensure to reduce the influence of the NLOS propagation, multipath interference and IMU drift cumulative error of UWB over time, and the distributed architecture of DEKF can prevent problems caused by asynchronous sensor data input, thereby maintaining consistent system state. Overall, the proposed system can dynamically adapt to environmental changes and ensure stable operation in different indoor scenes, achieving high robust positioning.

[0111] Table 2 Comparison of root mean square error and absolute average error of five different optimization algorithms

[0112]

[0113] Table 2 is a comparison of the root mean square error and absolute average error of five optimization algorithms (no algorithm, EKF, DEKF, CNN-LSTM, CNN-STM-DEKF). The root mean square error (RMSE) value of the CNN-STM-DEKFRMSE optimization algorithm proposed in the application is 0.242 and 0.205 (broken line path and square path) respectively, which is significantly lower than that of other algorithms; the absolute average error (MAE) value is 0.208 and 0.192 respectively, which further proves its optimization effect. In the broken line path, the error of each algorithm is large, but the error of CNN-LSTM-DEKF is significantly lower than that of other algorithms; in the square path, CNN-LSTM-DEKF also maintains the lowest error, showing high adaptability to regular trajectories. Comprehensive analysis shows that CNN-LSTM-DEKF significantly reduces the influence of NLOS interference and IMU drift through CNN-LSTM optimization of the deep features of UWB signals, and combines the distributed filtering method of DEKF to realize the complementarity of time and space information. In complex indoor scenes, the proposed algorithm performs outstanding positioning accuracy and robustness, and is suitable for dynamic and variable environments.

[0114] In order to more clearly show the performance evaluation results of the algorithm proposed in the application and other algorithms, Figure 6The ranging error box plot of five optimization algorithms (no algorithm, EKF, DEKF, CNN-LSTM, CNN-STM-DEKF) is shown, Figure 6 (a) is the box plot of the root mean square error (RMSE), and the data reflects the error distribution of each optimization algorithm; Figure 6 (b) is the box plot of the mean absolute error (MAE), which intuitively compares the positioning accuracy of the five algorithms. Through comparison, it can be clearly seen that the CNN-LSTM-DEKF algorithm proposed in the application performs better in RMSE and MAE, and the error distribution range is the smallest.

[0115] The application proposes an indoor comprehensive positioning system based on a distributed extended Kalman filter deep learning optimization algorithm, proposes a CNN-LSTM DEKF optimization algorithm based on UWB and IMU tight coupling, an indoor positioning method based on UWB and IMU tight coupling, acquires ranging information through IMU and UWB sensors, and then uses the CNN-LSTM-DEKF optimization algorithm to improve the positioning accuracy. Experimental results show that compared with the traditional UWB / IMU fusion method, the CNN-LSTM-DEKF method realizes obviously higher positioning accuracy. In order to evaluate the robustness of the proposed method, by comparing the RMSE and MAE of the optimization algorithm proposed in the application and other algorithms, it can be concluded that the CNN-LSTM-DEKF optimization algorithm proposed in the application performs better, and the RMSE values and MAE values under two different paths are 0.242, 0.208 and 0.205, 0.192. Therefore, the CNN-LSTM-DEKF optimization algorithm can be a better choice for improving the positioning accuracy of fused sensors in indoor environments. In addition, in future work, it is also necessary to verify the positioning accuracy of the optimization algorithm in complex environments and explore the performance of integrating camera input into the system.

[0116] The application also provides an electronic device, comprising a memory, a processor and computer program instructions stored on the memory and capable of being executed by the processor, when the processor executes the computer program instructions, the method steps involved in the system described above can be realized.

[0117] The application also provides a computer readable storage medium, which stores computer program instructions capable of being executed by a processor, when the processor executes the computer program instructions, the method steps involved in the system described above can be realized.

[0118] Those skilled in the art will appreciate that embodiments of the application can be readily used as a method, a system or a computer program product. Accordingly, the application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code thereon.

[0119] The application is described in reference to the flow diagrams and / or block diagrams of the methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing element or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flow diagram and / or block diagram block or blocks. Figure 1 one or more functions specified in the flow diagram and / or block diagram block or blocks.

[0120] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow diagram and / or block diagram block or blocks. Figure 1 one or more functions specified in the flow diagram and / or block diagram block or blocks. Figure 1 one or more functions specified in the flow diagram and / or block diagram block or blocks.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow diagram and / or block diagram block or blocks. Figure 1 one or more functions specified in the flow diagram and / or block diagram block or blocks. Figure 1 one or more functions specified in the flow diagram and / or block diagram block or blocks.

[0122] References:

[0123] [1]Picard Joseph S,Weiss Anthony J,et al.Direct Position Determination Sensitivity to NLOS Propagation Effects on Doppler-Shift[J].IEEE Transactions on signal processing,2019,67(14),3870-3881.

[0124] [2]Hyunseob Oh,Seungbeom Seo,Jinwook Seo,et al.An Efficient Ultra-Wideband Infrastructure Planning for Indoor Positioning Systems Using DL-TDOA[J].IEEE conference on global communications,GLOBECOM,2023,201-206.

[0125] [3]Kim,Moon-Sik,Yu,Si-Bok,Lee,Kwang-Soo,et al.Development of a High-Precision Calibration Method for Inertial Measurement Unit[J].

[0126] International journal of precision engineering and manufacturing,2014,15(3),567-575.

[0127] [4]Gentner Christian,Avram Diana,et al.WiFi-RTT-SLAM:Simultaneouslyestimating the positions ofmobiledevices and WiFi-RTT access points[J].ION GNSS+,2021,3142-3148.

[0128] [5] Zhang Xiaotian, Jia Guanghui, Huang Hai, et al. An approach for constituting double / multi wall BLE by single wall BLE of spacecraft shield[J]. International journal of impact engineering, 2014, 69, 114-121.

[0129] [6] Hao Pan, Yu Chen, Xiaogang Qi, Meili Liu, et al. A UWB / IMU-Assisted Fingerprinting Localization Framework with Low Human Efforts[J].

[0130] China communications, 2024, 21(6), 40-52.

[0131] [7] Zheng Shuaikang, Li Zhitian, Liu Yunfei, et al. An Optimization-Based UWB-IMU Fusion Framework for UGV[J]. IEEE sensors journal, 2022, 22(5), 4369-4377.

[0132] [8] Zhu Si, Sun Fei, Chen Xiaohui, et al. Direct Position Determination Sensitivity to NLOS Propagation Effects on Doppler-Shift[J]. IEEE International Conference on Communications in China, 2013, 321-326.

[0133] [9]Jiang Changhui,Chen Shuai,Chen Yuwei,et al.An UWB Channel ImpulseResponse De-Noising Method for NLOS / LOS Classification Boosting[J].IEEEcommunications letters,2020,24(11),2513-2517.

[0134]

[10] Kim Dae-Ho,FarhadArshad,Pyun Jae-Young,et al.UWB PositioningSystem Based on LSTM Classification With MitigatedNLOS Effects[J].IEEEinternet of things journal,2023,10(2),1822-1835.

[0135]

[11] Lee Gang Toe,Seo Seung Beom,Jeon Wha Sook,et al.IndoorLocalization by Kalman Filter based Combining of UWB-Positioning and PDR[J].IEEE consumer communications and networking conference,2021.

[0136]

[12] Wang Yan,Li Xin,et al.An improved robust EKF algorithm based onsigma points for UWB and foot-mounted IMU fusion positioning[J].IEEETransactions on signal processing,2021,66(2),329-350.

[0137]

[13] Liu Fei, Li Xin, Wang Jian, et al. An Adaptive UWB / MEMS-IMU Complementary Kalman Filter for Indoor Location in NLOS Environment [J]. Remote sensing, 2019, 11(22), 2628.

[0138]

[14] J Zeng, J Liu, T Zou, et al. Distributed extended Kalman filtering for wastewater treatment processes [J]. Industrial and engineering chemistry research, 2016, 55(28), 7720-7729.

[0139]

[15] Li Xiaojie, Yin Xunyuan, et al. Exploring the linkage between partition-based distributed extended Kalman filter and distributed full-information estimation [J]. IEEE 7th international conference on industrial cyber-physical systems, 2024.

[0140]

[16] de Hidenori, Kurita Takio, et al. Improvement of Learning for CNN with ReLU Activation by Sparse Regularization [J]. IEEE international joint conference on neural networks, 2017, 2684-2691.

[0141]

[17] Yadav Hemant, Thakkar Amit, et al. NOA-LSTM: An efficient LSTM cell architecture for time series forecasting[J]. Expert systems with applications, 2023, 122 333.

[0142]

[18] Luo Long, et al. Research and analysis of manipulator control method based on deep learning[J]. Journal of engineering research, 2024, 12(3), 412-420.

[0143]

[19] X Li, S Bo, Y Qin, et al. Iterative distributed moving horizon estimation of linear systems with penalties on both system disturbances and noise[J]. Chemical engineering research and design, 2023, 194, 878-893.

[0144]

[20] Changhui Jiang, Jichun Shen, Shuai Chen, et al. UWB NLOS / LOS Classification Using Deep Learning Method[J]. IEEE Transactions on signal processing, 2020, 24(10), 2226-2230.

[0145] The above description is only the preferred embodiment of the present application, not other forms of the present application, any skilled in the art may use the above disclosed technical content to make changes or modifications as equivalent embodiments. However, any simple modification, equivalent change and modification made on the basis of the technical essence of the present application without departing from the technical scheme of the present application shall still fall within the protection scope of the present application.

Claims

1. An indoor integrated positioning system based on a distributed extended Kalman filtering deep learning optimization algorithm, characterized in that, Comprise: An ultra-wideband (UWB) module configured to collect UWB ranging data, including channel impulse response (CIR) data of the UWB; An inertial measurement unit (IMU) module configured to collect IMU ranging data; An optimization algorithm module based on a distributed extended Kalman filter (DEKF) deep learning, which proposes an optimization algorithm CNN-LSTM-DEKF that fuses a convolutional long short-term memory network (CNN-LSTM) and a distributed extended Kalman filter (DEKF). The CNN-LSTM extracts features and makes classification predictions from the CIR data of the UWB, and then the DEKF fuses the classified and predicted data with the IMU ranging data to suppress the cumulative error caused by the time drift of the IMU, thereby ensuring the accuracy of the ranging data and preventing cumulative error and time synchronization problems in the data fusion of various sensors; The DEKF calculation process is as follows: Step one, local prediction stage Each node i independently runs the prediction step and updates its state estimation based on the local model: is the predicted state estimate of node i at time k compared to the previous time k-1; denotes the state prediction of node i at time k-1; denotes the control input of node i at time k-1; is the predicted error covariance matrix of node i at time k compared to the previous time k-1, denotes the state transition function of node i at time k; denotes the predicted error covariance matrix of node i at time k-1; denotes the state transition matrix derived by differentiating the state transition function; denotes the system noise covariance matrix, which describes the uncertainty of the state transition; Step two, local update stage Each node updates the local state according to the observation value: Kalman gain for node i, used to weigh the influence of the prediction and the observation; Jacobian matrix of the observation function; measurement of node i; nonlinear observation function describing the mapping from the state to the measurement; measurement noise covariance describing the uncertainty of the sensor measurement; Step three, information sharing stage, each node shares local state estimation and covariance information through communication to form a global information set, the shared content includes , , ; Step four, global fusion stage, the optimal state estimation of the whole system is obtained through global information fusion, and the weighted sum of the estimation of each node is obtained through weighted average method according to the weight of the covariance matrix: represents a global fused optimal state estimation, represents a weight of the i-th node, represents a local state estimation of the i-th node; Step five, global update feedback stage, the fusion result is fed back to each node to adjust the local state to maintain global consistency. 2.The indoor integrated positioning system based on the distributed extended Kalman filter deep learning optimization algorithm of claim 1, wherein, The network framework of CNN-LSTM includes two basic components: convolutional neural network (CNN) and long short-term memory (LSTM). In CNN-LSTM, CIR data is used as the input vector of CNN, and CNN learns the frequency characteristics of CIR data to extract features using the non-time structure of input signals. Then the output of CNN is input into LSTM to learn the dependency of time series data for non-line-of-sight (NLOS) signal and line-of-sight (LOS) signal classification and prediction. 3.The indoor integrated positioning system based on the distributed extended Kalman filter deep learning optimization algorithm according to claim 1 or 2, characterized in that, CNN-LSTM is composed of a convolutional layer, a max pooling layer, a fully connected layer, an LSTM layer, and a fully connected layer, and the generalization ability is improved through Dropout regularization. 4.The indoor integrated positioning system based on the distributed extended Kalman filtering deep learning optimization algorithm of claim 2, wherein, CNN-LSTM extracts features and makes classification predictions from the CIR data of the UWB, which includes the following stages: CNN component: CNN processes the original CIR data to extract spatial features that capture changes in signal patterns caused by NLOS conditions; multiple convolutional layers are used, each using the same activation function to detect input data for features. To ensure the nonlinearity of the activation function, ReLU is used, whose equation is as follows: x represents the parameter of the ReLU function; After the convolutional layer outputs to the pooling layer, the pooling layer reduces the dimension through downsampling and prevents overfitting. Then, the pooling layer outputs to the fully connected layer, which selects the classifier as the output through the SoftMax function; LSTM component: The output of CNN is passed to LSTM to capture the time dependency in the CIR data sequence. The gating architecture of LSTM enables it to model long-term relationships in the data. After the data extracted by CNN is input to LSTM, LSTM is trained using a labeled dataset with a cross-entropy loss function to ensure accurate binary classification of UWB signals as LOS or NLOS. 5.The indoor integrated positioning system based on the distributed extended Kalman filter deep learning optimization algorithm of claim 4, wherein, The gating architecture of LSTM includes input, forget, and output gates. 6.The indoor integrated positioning system based on the distributed extended Kalman filter deep learning optimization algorithm according to claim 4 or 5, characterized in that, The input and output gate functions in the gating architecture of LSTM are described as follows: In the formula is the input gate, which determines whether the spatial features extracted by CNN, including the classification-related information of NLOS and LOS, are used to update the time series state; represents the hidden state of the previous time step, containing the dynamic information of the feature sequence of the past time step; represents the feature vector output by the current time step CNN, containing the UWB CIR spatial features of the current time step; , represents the weight and bias parameters of the input gate, , represents the weight and bias parameters of the output gate, which are optimized by gradient descent during the training process to adapt to the influence of different input features; is the candidate state, which is used to generate the new candidate state of the current time step to capture the potential features of the current time step CNN features and historical hidden states, and retain important information related to LOS / NLOS; represents the hyperbolic tangent activation function, which compresses the candidate state to the range of (-1, 1) and is used to represent the possible state update; represents the weight and bias parameters of the candidate state, which can effectively represent the nonlinear relationship of the time series after optimization; is the output gate, which extracts the most relevant information of LOS / NLOS classification from the candidate state and outputs to ; represents the hidden state of the current time step, which is the output of the LSTM and also one of the inputs of the next time step; represents the nonlinear transformation of the current time series state, which keeps its range in (-1, 1).

7. The indoor integrated positioning system based on distributed extended Kalman filter deep learning optimization algorithm according to claim 1, characterized in that, The DEKF is verified by MAE and RMSE. 8.The indoor integrated positioning system based on the distributed extended Kalman filter deep learning optimization algorithm of claim 1, wherein, The ultra-wideband UWB module collects ranging data of anchor points through an ultra-wideband UWB tag. 9.The indoor integrated positioning system based on the distributed extended Kalman filter deep learning optimization algorithm of claim 1, wherein, The inertial measurement unit IMU module collects ranging data through a three-axis acceleration sensor.

Images