Pedestrian inertial navigation positioning method and device based on lstm zero speed detection and EKF

Abstract

The application discloses a pedestrian inertial navigation positioning method based on LSTM zero speed detection and EKF, wherein the LSTM is used to identify the pedestrian state as zero speed or walking state according to IMU data; an error state vector is constructed; an error differential propagation equation is constructed according to the error state vector; a state transition matrix is obtained by discretizing the continuous time state equation; the prior state estimation value and the prior state covariance matrix at the current time are calculated; zero is taken as a virtual absolute observation value; the difference between the actual observation speed and the current predicted speed is taken as the measurement innovation; the Kalman gain is calculated according to the measurement innovation; the error state vector is corrected by using the Kalman gain, and the posterior state covariance matrix is updated; the false positive label noise is eliminated from the source, the pure and high-fidelity observation update signal can be provided for the EKF filter, and thus the inertial navigation drift problem in a complex dynamic scene is greatly inhibited.

Description

Technical Field

[0001] This invention relates to the field of measurement, and more particularly to a pedestrian inertial navigation and positioning method and apparatus based on LSTM zero-velocity detection and EKF. Background Technology

[0002] In indoor or complex urban environments with limited satellite signals, PINS (Pedestrian Inertial Navigation Systems) based on foot-worn inertial measurement units (IMUs) have become an important means of autonomous positioning. However, low-cost microelectromechanical inertial sensors inevitably suffer from severe zero-bias errors and measurement noise, and simple inertial recursion leads to a rapid accumulation and divergence of positioning errors over time. To effectively suppress this drift, existing technologies widely employ the Zero Velocity Update (ZUPT) algorithm. The core principle of this algorithm is to utilize the biomechanical characteristics of pedestrian gait: in a complete gait cycle, when the foot is in the intermediate support phase with full contact with the ground, the theoretical instantaneous velocity of the human body is zero. By capturing these brief zero-velocity intervals, the system introduces "zero velocity" as an absolute pseudo-observation into the extended Kalman filter (EKF), thereby achieving periodic correction of accumulated state errors.

[0003] In this technical architecture, the positioning accuracy of the entire navigation system fundamentally depends on the accuracy of the front-end zero-velocity detector. Currently, mainstream technologies primarily rely on statistical signal processing methods, such as angular rate energy (ARE) detectors, acceleration variance (MV) detectors, and generalized likelihood ratio test (GLRT) detectors. The shortcomings of these traditional methods lie in their heavy reliance on simplified gait physics model assumptions and the need for extensive experimental calibration of fixed or adaptive decision thresholds. Because pedestrian gait patterns are complex and varied in real-world scenarios, this threshold-based heuristic decision-making mechanism has extremely poor generalization ability. For example, in high-dynamic scenarios such as rapid walking or irregular foot swaying, traditional detectors experience severe performance degradation due to drastic changes in impact force and signal. ARE detectors are prone to misclassification at the start of walking, MV detectors are prone to false detections during the swaying phase or missed detections due to minor movements during the support phase, and the GLRT algorithm also exhibits a high misclassification rate under high dynamic conditions.

[0004] The aforementioned technical flaws can trigger a fatal chain reaction: if the detector misjudges a situation, it will incorrectly mark a non-zero velocity state as zero velocity, thus injecting extremely poor nonlinear observation noise into the downstream EKF filter. This erroneous spurious observation data will directly undermine the convergence of the traditional Kalman filter, causing the trajectory extrapolation to diverge rapidly. Summary of the Invention

[0005] The following is an overview of the topics described in detail in this article.

[0006] The purpose of this application is to at least partially solve one of the technical problems existing in the related technologies. The embodiments of this application provide a pedestrian inertial navigation and positioning method and device based on LSTM zero velocity detection and EKF.

[0007] An embodiment of the first aspect of this application provides a pedestrian inertial navigation and localization method based on LSTM zero-velocity detection and EKF, comprising:

[0008] Acquire inertial measurement unit (IMU) data;

[0009] The LSTM neural network identifies the pedestrian's state as either zero speed or walking based on IMU data.

[0010] An error state vector is constructed based on the position error, velocity error, attitude error, gyroscope zero bias error, and accelerometer zero bias error in the global coordinate system. The error state vector is used to characterize the deviation between the actual state of the system and the purely calculated state of the inertial navigation system.

[0011] The error differential propagation equation of the inertial navigation system is constructed based on the error state vector. The continuous-time state equation of the inertial navigation system is determined based on the error differential propagation equation. The continuous-time state equation is discretized within the sampling time to obtain the state transition matrix. The prior state estimate and prior state covariance matrix at the previous moment are calculated using the nonlinear state transition function and the state transition matrix.

[0012] When the pedestrian is in a zero-speed state, zero is taken as a virtual absolute observation value. The difference between the actual observed speed and the current predicted speed is calculated as the measurement information. The Kalman gain is calculated based on the measurement information. The error state vector is corrected using the Kalman gain, and the posterior state covariance matrix is ​​updated.

[0013] According to certain embodiments of the first aspect of this application, during the training phase of the LSTM neural network, the LSTM neural network is optimized using a loss function;

[0014] The loss function is expressed by the formula:

[0015] ;

[0016] In the formula, For loss function, The length of the sliding window. For true gait labeling, Predict probabilities for the network.

[0017] According to certain embodiments of the first aspect of this application, the error state vector is expressed by the formula:

[0018] ;

[0019] In the formula, Let be the error state vector. This represents the position error in the global coordinate system. The velocity error is in the global coordinate system. This represents the attitude error in the global coordinate system. This refers to the zero bias error of the gyroscope. This is the zero bias error of the accelerometer.

[0020] According to certain embodiments of the first aspect of this application, the error differential propagation equation is expressed by the following formula:

[0021] ; ; ;

[0022] In the formula, Let the direction cosine matrix be the distance from the vehicle coordinate system to the navigation coordinate system. This is the specific force measurement value from the accelerometer.

[0023] According to certain embodiments of the first aspect of this application, the prior state estimate at the current moment is expressed by the formula: ;

[0024] The prior state covariance matrix at the current moment can be expressed by the formula: ;

[0025] In the formula, This is the estimated prior state value at the current moment. It is a nonlinear state transition function. This is the state estimate from the previous moment. For IMU sensor data, Let be the prior state covariance matrix at the current moment. This is the linearized state transition matrix. Let be the state covariance matrix of the previous time step. Let be the noise covariance matrix of the inertial navigation system.

[0026] According to certain embodiments of the first aspect of this application, calculating the Kalman gain based on the measured innovation includes:

[0027] The innovation covariance matrix is ​​calculated based on the linear observation matrix, the prior state covariance matrix at the current time, and the constant noise covariance matrix of the zero-velocity measurement.

[0028] The Kalman gain is calculated based on the linear observation matrix, the prior state covariance matrix at the current time, and the innovation covariance matrix.

[0029] According to certain embodiments of the first aspect of this application, the novel covariance matrix is ​​expressed by the formula: ;

[0030] The Kalman gain is expressed by the formula: ;

[0031] In the formula, The new information covariance matrix, For linear observation matrices, Let be the prior state covariance matrix at the current moment. The constant noise covariance matrix for zero-velocity measurements. For Kalman gain.

[0032] According to certain embodiments of the first aspect of this application, the corrected error state vector is expressed by the formula: ;

[0033] The updated empirical state covariance matrix is ​​expressed by the formula: ;

[0034] In the formula, This is the corrected error state vector. For Kalman gain, To measure the new information, The updated empirical state covariance matrix, For linear observation matrices, Let be the prior state covariance matrix at the current moment.

[0035] According to a second aspect of this application, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the pedestrian inertial navigation and positioning method based on LSTM zero-velocity detection and EKF as described in the first aspect of this application.

[0036] According to a third aspect of this application, a computer storage medium stores computer-executable instructions for performing the pedestrian inertial navigation and positioning method based on LSTM zero-velocity detection and EKF as described in an embodiment of the first aspect of this application.

[0037] The above scheme has at least the following beneficial effects: it eliminates the cumbersome manual threshold setting, constructs a sliding window from continuous six-axis IMU normalized data, and directly inputs it into a lightweight LSTM network. Utilizing the powerful gating mechanism of LSTM to extract deep gait temporal features, it can robustly and with extremely high accuracy output binary classification decisions for gait phases under various complex stride lengths and movement speeds. By eliminating false positive label noise at the source, it can provide clean, high-fidelity observation update signals for the backend EKF filter, thereby significantly suppressing inertial navigation drift problems in complex dynamic scenes. Attached Figure Description

[0038] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0039] Figure 1 This is a flowchart illustrating the steps of a pedestrian inertial navigation and localization method based on LSTM zero-velocity detection and EKF.

[0040] Figure 2 This is a schematic diagram of the principle of an inertial navigation system;

[0041] Figure 3 This is a structural diagram of the LSTM zero-rate detector network;

[0042] Figure 4 This is a schematic diagram of the internal structure of an LSTM cell;

[0043] Figure 5 This is a comparison chart of the output waveforms of different zero-velocity detection algorithms;

[0044] Figure 6 This is a comparison chart of the straight-line walking trajectories corresponding to different zero-velocity detectors;

[0045] Figure 7 This is a graph of the cumulative distribution function of horizontal positioning error. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0048] The embodiments of this application provide a pedestrian inertial navigation and positioning method and apparatus based on LSTM zero-velocity detection and EKF.

[0049] The embodiments of this application will be further described below with reference to the accompanying drawings.

[0050] Reference Figure 1 The pedestrian inertial navigation and positioning method includes the following steps:

[0051] Step S100: Acquire inertial measurement unit (IMU) data;

[0052] Step S200: The pedestrian's state is identified as either zero speed or walking state based on the inertial measurement unit (IMU) data using an LSTM neural network.

[0053] Step S300: Construct an error state vector based on the position error, velocity error, attitude error, gyroscope bias error, and accelerometer bias error in the global coordinate system.

[0054] Step S400: Construct the error differential propagation equation of the inertial navigation system based on the error state vector; determine the continuous-time state equation of the inertial navigation system based on the error differential propagation equation; discretize the continuous-time state equation within the sampling time to obtain the state transition matrix; and calculate the prior state estimate and prior state covariance matrix at the current time using the state estimate and state covariance matrix of the previous time step through the nonlinear state transition function and the state transition matrix.

[0055] In step S500, when the pedestrian is in a zero-speed state, zero is taken as a virtual absolute observation value. The difference between the actual observed speed and the current predicted speed is calculated as the measurement information. The Kalman gain is calculated based on the measurement information. The error state vector is corrected using the Kalman gain, and the posterior state covariance matrix is ​​updated.

[0056] In step S100, a six-axis inertial measurement unit (IMU) is rigidly fixed to the forefoot or heel of the pedestrian's foot. This location is the core force point of the gait support phase, where the zero-velocity characteristic is most pronounced. The IMU sampling frequency is configured to be between 200Hz and 500Hz to adapt to the characteristic that the gait cycle of a pedestrian's single step is less than 1 second, ensuring the integrity of the gait timing data.

[0057] Collect triaxial acceleration and triaxial angular velocity data from the inertial measurement unit, such as Acc X, Acc Y, Acc Z, Gyro X, Gyro Y, and Gyro Z. Construct a data structure with a length of... (like A continuous-time sliding window is used as the input sequence and fed into the inertial navigation system. (Refer to...) Figure 2 , Figure 2 This is a schematic diagram of the principle of an inertial navigation system.

[0058] To eliminate the influence of different sensor range dimensions, the six-axis IMU data within the sliding window were uniformly normalized and scaled, mapping the data to the [0,1] interval. During normalization, the temporal characteristics and relative change trends of the original data for each axis were preserved, eliminating only the feature distortion caused by differences in sensor range and dimensions. The normalized IMU data was then fed into the neural network.

[0059] Reference Figure 3 For step S200, the neural network includes a core feature extraction module, which is an LSTM zero-rate detector network comprising multiple LSTM units, for example, six LSTM units. (Refer to...) Figure 4 An LSTM unit consists of a forget gate, an input gate, and an output gate. Normalized time-series data is input into a hidden layer containing multiple LSTM units.

[0060] LSTM networks utilize their gating mechanism to extract deep gait temporal features and generate a binary classification probability distribution through the Softmax function of the output layer. Specifically, tanh is used as the activation function and sigmoid as the gating activation function, enabling forgetting, input, and output gates for unit states to achieve long-term memory and effective extraction of gait temporal features. Fully connected layers map the output features of the hidden layers into two-dimensional vectors, corresponding to the support and swing phases respectively. The Softmax function converts the two-dimensional vectors into a binary probability distribution, outputting two probability values ​​in the range [0,1], representing the confidence that the current window belongs to the support / swing phase state.

[0061] During the network training phase, binary cross-entropy is used as the loss function to optimize the LSTM model. ;in, Represents the loss function. Indicates the true gait label, Indicates the probability of network prediction. The length of the sliding window.

[0062] During the actual inference phase, when the probability value output by Softmax is greater than the set classification threshold (e.g., 0.5), the pedestrian is determined to be in the support phase, i.e., the zero-speed state, and the output is marked as 1; if it is less than the threshold, it is determined to be in the swing phase, and the output is marked as 0. This detection result serves as a hard trigger switch for subsequent error correction.

[0063] For step S300, in the back-end calculation of the strapdown inertial navigation system, due to the strong nonlinearity problem in directly estimating the absolute navigation state, a 15-dimensional error state vector is constructed. It is used to characterize the deviation between the actual state of the system and the purely calculated state of the inertial navigation system.

[0064] This vector is defined as: ;

[0065] In the formula, Let be the error state vector. This represents the position error in the global coordinate system. The velocity error is in the global coordinate system. This represents the attitude error in the global coordinate system. This refers to the zero bias error of the gyroscope. This is the zero bias error of the accelerometer.

[0066] By constructing an error state vector instead of directly estimating the absolute navigation state, the strong nonlinearity problem of strapdown inertial navigation is transformed into a weak nonlinearity problem of the error state, which significantly reduces the error caused by EKF linearization and improves the stability and accuracy of filtering.

[0067] For step S400, in order to predict the state at discrete time points, the error differential propagation equation of the system in continuous time is first derived based on the basic principle of strapdown inertial navigation.

[0068] After neglecting terms such as Earth's rotation, which have negligible impact on slow-moving pedestrians, the error differential propagation equation can be expressed as:

[0069] Differential of position error: ;

[0070] Differential of velocity error: ;

[0071] Attitude error derivative: ;

[0072] In the formula, Let the direction cosine matrix be the distance from the vehicle coordinate system to the navigation coordinate system. This is the specific force measurement value from the accelerometer.

[0073] The velocity error is caused by the acceleration projection error resulting from the attitude error and the acceleration measurement error; the attitude error is caused by the gyroscope measurement error.

[0074] The above continuous-time model is applied at the sampling time. Discretize the matrix using a first-order Taylor expansion to obtain a 15×15 state transition matrix. ,as follows:

[0075] ;

[0076] In the formula, for identity matrix and These are the correlation times of the first-order Markov process with zero bias of the sensor.

[0077] In the prediction phase of the Extended Kalman Filter (EKF), the prior state estimate and prior state covariance matrix at the current time are calculated using the previous time step's state estimate and state covariance matrix through a nonlinear state transition function and state transition matrix.

[0078] The prior state estimate at the current moment is expressed by the formula: ;

[0079] The prior state covariance matrix at the current moment can be expressed by the formula: ;

[0080] In the formula, This is the estimated prior state value at the current moment. It is a nonlinear state transition function. This is the state estimate from the previous moment. Let be the prior state covariance matrix at the current moment. For IMU sensor data, This is the linearized state transition matrix. Let be the state covariance matrix of the previous time step. Let be the noise covariance matrix of the inertial navigation system.

[0081] For step S500, when the LSTM detector output flag is 1, that is, the pedestrian state is zero velocity state and the pedestrian's feet are in the support phase, the zero velocity observation update mechanism is triggered.

[0082] During the support phase, the actual physical velocity of the feet relative to the ground is zero. .

[0083] Using zero as a virtual absolute observation value Calculate and measure new information That is, the measurement information is the difference between the actual observed velocity and the current predicted velocity of the inertial navigation system. The difference. The measurement of information is expressed by the following formula: .

[0084] Since this observation focuses only on velocity physical quantities, a linear observation matrix is ​​constructed to extract velocity error components from the 15-dimensional global error state. The linear observation matrix has a dimension of 3×15. (Linear observation matrix) This can be expressed as follows: .

[0085] Calculate the new information covariance matrix and the Kalman gain that satisfies the minimum mean square error (MMSE) criterion.

[0086] The new information covariance matrix is ​​expressed by the following formula: ;

[0087] The Kalman gain is expressed by the formula: ;

[0088] In the formula, The new information covariance matrix, For linear observation matrices, Let be the prior state covariance matrix at the current moment. The constant noise covariance matrix for zero-velocity measurements is used to weigh the confidence level of LSTM decisions. For Kalman gain.

[0089] Finally, the EKF update phase is performed, using Kalman gain to correct the error state and update the posterior error covariance matrix. .

[0090] The corrected error state vector is expressed by the following formula: ;

[0091] The updated empirical state covariance matrix is ​​expressed by the formula: ;

[0092] In the formula, This is the corrected error state vector. For Kalman gain, To measure the new information, The updated empirical state covariance matrix, For linear observation matrices, Let be the prior state covariance matrix at the current moment.

[0093] Using the Kalman gain matrix mentioned above Through the cross-coupling effect, the system not only directly eliminates the accumulated velocity error, but more importantly, it utilizes the relationship between the velocity error, attitude error, and sensor zero-bias error in the covariance matrix. The strong mathematical correlation demonstrated indirectly calculates and compensates for horizontal attitude angle (roll and pitch) errors that are difficult to observe directly, as well as the device bias of the IMU. This closed-loop correction mechanism fundamentally severs the vicious nonlinear error transmission chain caused by acceleration projection errors due to attitude errors, resulting in velocity divergence and position drift, ultimately achieving high-precision pedestrian trajectory reconstruction.

[0094] Reference Figure 2 In general, inputting raw gyroscope data and raw accelerometer data The original data is corrected for sensor errors, and the corrected data in the carrier coordinate system is output. and Based on the inertial mechanics equations and data in the corrected carrier coordinate system. and Perform strapdown inertial navigation calculations to obtain the attitude direction cosine matrix in navigation coordinates in real time. ,speed and location Based on the data in the corrected carrier coordinate system and Gait temporal features are extracted using an LSTM network to identify whether the current moment is a support phase. If it is a support phase, zero-velocity correction is performed. , Based on the error state differential equation, the prior error state and prior covariance matrix at the current time are discretized and calculated using the previous time step's error state and covariance. Using a support phase velocity of 0 as a virtual observation, a zero-velocity observation equation is constructed, and the Kalman gain is calculated to obtain the error estimates for attitude, velocity, and position. The attitude direction cosine matrix ,speed and location and the error estimate of the EKF output. Make corrections, attitude corrections: Speed ​​correction: Position correction: The corrected attitude, velocity, and position are returned to the strapdown inertial navigation system for calculation, achieving closed-loop iteration.

[0095] To comprehensively validate the pedestrian inertial navigation and localization method based on an LSTM zero-velocity detector, this scheme underwent rigorous comparative experiments in a real outdoor scenario. The experiments required pedestrians to walk on an outdoor straight track, with microelectromechanical inertial sensors mounted on their feet to collect motion data. To ensure absolute objectivity in the evaluation, a high-precision RTK GNSS module with centimeter-level positioning accuracy was used as the trajectory reference ground truth. This pedestrian inertial navigation and localization method based on an LSTM zero-velocity detector was comprehensively quantitatively compared with a traditional threshold detector (angular rate energy ARE, acceleration variance MV, and generalized likelihood ratio test GLRT) after optimal parameter search.

[0096] In practical pedestrian navigation, dynamic changes in walking speed are the main cause of failure for traditional detectors. Experiments comprehensively tested three dynamic scenarios: slow speed (3-4 km / h), normal speed (4-6 km / h), and fast walking / running (8-10 km / h). Refer to Table 1, which compares the accuracy of gait phase detection at different walking speeds. Test results show that the performance of traditional detectors based on fixed thresholds drops drastically with increasing walking speed. In fast walking conditions, the strong impact force of the feet and rapid signal changes cause the MV and ARE detectors to miss a large number of support phases, resulting in a sharp drop in detection accuracy to 56.9% and 63.8%, respectively. Even the GLRT detector, which has relatively good overall performance, drops to 77.5%. In contrast, the LSTM detector proposed in this method, by learning the deep temporal dependencies of gait, completely eliminates the dependence on a single signal amplitude threshold, maintaining extremely high stability across all speed ranges, with its zero-speed interval detection accuracy consistently above 97%.

[0097] Table 1. Comparison of Gait Phase Detection Accuracy at Different Walking Speeds

[0098]

[0099] Traditional detectors, relying on a single physical feature, are prone to false positives. For example, ARE detectors are prone to false positives in the early stages of walking; MV detectors may false positives during the swing phase due to irregular foot movements, and may miss detections due to minute movements in the zero-velocity phase; GLRT detectors also suffer from detection errors under highly dynamic conditions. The LSTM zero-velocity detector proposed in this invention does not rely on a single threshold but learns features from data, significantly outperforming traditional algorithms. It can accurately estimate the zero-velocity interval even in highly dynamic motion, outputting an extremely clean and accurate zero-velocity decision state.

[0100] Reference Figure 5In the comparison of the output waveforms of the four zero-velocity detection algorithms, the blue line represents the acceleration amplitude, and the red line represents the detector output, where 1 represents the support phase and 0 represents the oscillation phase. It can be seen that the red line output of LSTM is the most stable, with no glitches or false positives.

[0101] The significant improvement in front-end zero-velocity detection accuracy enables this invention to provide extremely high-fidelity error update signals for the back-end filter, effectively curbing trajectory divergence. The final physically calculated trajectory shows that the trajectories predicted by GLRT and MV detectors have considerable errors, while the error based on the LSTM detector is almost negligible, closely resembling the ideal true trajectory.

[0102] Reference Figure 6 , Figure 6 The purple line (Proposed LSTM) almost completely overlaps with the black line (Ground Truth).

[0103] Table 2. Comparison of Positioning Error Quantification for Different Zero-Voltage Detectors

[0104]

[0105] Referring to Table 2, which compares the positioning errors of different zero-velocity detectors, the average positioning errors based on traditional ARE, MV, and GLRT detectors are as high as 2.259m, 7.545m, and 3.382m, respectively. After adopting the LSTM detector of this method, the average positioning error of the system is significantly reduced to only 0.226m. Compared with the three traditional detection methods mentioned above, the positioning accuracy of this method achieves significant improvements of 90.0%, 97.0%, and 93.3%, respectively. Furthermore, in terms of the root mean square error (RMSE), a measure of system stability, this method achieves only 0.265m, also representing a significant improvement of 88.6% to 96.9%.

[0106] Reference Figure 7 According to the cumulative distribution function (CDF) of horizontal positioning error, the purple line (Proposed LSTM) is closest to the top left corner, proving that its error distribution is the most concentrated and its error value is the smallest. Further statistical evidence demonstrates the superiority of this method.

[0107] Embodiments of this application provide an electronic device. The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the pedestrian inertial navigation and positioning method based on LSTM zero-velocity detection and EKF as described above.

[0108] This electronic device can be any smart terminal, including computers.

[0109] In general, for the hardware structure of electronic devices, the processor can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, to execute relevant programs and implement the technical solutions provided in the embodiments of this application.

[0110] The memory can be implemented in the form of read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory and is called and executed by the processor.

[0111] Input / output interfaces are used to implement information input and output.

[0112] The communication interface is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0113] The bus transmits information between various components of a device, such as the processor, memory, input / output interfaces, and communication interfaces. The processor, memory, input / output interfaces, and communication interfaces communicate with each other within the device via the bus.

[0114] Embodiments of this application provide a computer storage medium. The computer storage medium stores computer-executable instructions for executing the pedestrian inertial navigation and positioning method based on LSTM zero-velocity detection and EKF as described above.

[0115] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium. In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0116] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0117] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0118] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0119] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0120] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed between each other may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms. Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0121] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A pedestrian inertial navigation and positioning method based on LSTM zero-velocity detection and EKF, characterized in that, include: Acquire inertial measurement unit (IMU) data; The LSTM neural network identifies the pedestrian's state as either zero speed or walking based on IMU data. An error state vector is constructed based on the position error, velocity error, attitude error, gyroscope zero bias error, and accelerometer zero bias error in the global coordinate system. The error state vector is used to characterize the deviation between the actual state of the system and the purely calculated state of the inertial navigation system. The error differential propagation equation of the inertial navigation system is constructed based on the error state vector. The continuous-time state equation of the inertial navigation system is determined based on the error differential propagation equation. The continuous-time state equation is discretized within the sampling time to obtain the state transition matrix. The prior state estimate and prior state covariance matrix at the previous moment are calculated using the nonlinear state transition function and the state transition matrix. When the pedestrian is in a zero-speed state, zero is taken as a virtual absolute observation value. The difference between the actual observed speed and the current predicted speed is calculated as the measurement information. The Kalman gain is calculated based on the measurement information. The error state vector is corrected using the Kalman gain, and the posterior state covariance matrix is ​​updated. 2.The pedestrian inertial navigation positioning method based on LSTM zero velocity detection and EKF of claim 1, wherein, During the training phase of the LSTM neural network, the LSTM neural network is optimized using a loss function; The loss function is expressed by the formula: ； wherein, is a loss function, is the length of the sliding window, is the true gait label, is the network prediction probability. 3.The pedestrian inertial navigation positioning method based on LSTM zero velocity detection and EKF of claim 1, wherein, The error state vector is expressed by the following formula: ； In the formula, Let be the error state vector. This represents the position error in the global coordinate system. The velocity error is in the global coordinate system. This represents the attitude error in the global coordinate system. This refers to the zero bias error of the gyroscope. This is the zero bias error of the accelerometer. 4.The pedestrian inertial navigation positioning method based on LSTM zero velocity detection and EKF of claim 3, wherein, The error differential propagation equation is expressed as follows: ； ； ； In the formula, Let the direction cosine matrix be the distance from the vehicle coordinate system to the navigation coordinate system. This is the specific force measurement value from the accelerometer.

5. The pedestrian inertial navigation positioning method based on LSTM zero velocity detection and EKF according to claim 1, characterized in that, The a priori state estimate at the current time is represented by the equation ; The a priori state covariance matrix at the current time is expressed by the equation ; In the formula, This is the estimated prior state value at the current moment. It is a nonlinear state transition function. This is the state estimate from the previous moment. For IMU sensor data, Let be the prior state covariance matrix at the current moment. This is the linearized state transition matrix. Let be the state covariance matrix of the previous time step. Let be the noise covariance matrix of the inertial navigation system. 6.The pedestrian inertial navigation positioning method based on LSTM zero velocity detection and EKF of claim 1, wherein, The calculation of the Kalman gain based on the measured information includes: The innovation covariance matrix is ​​calculated based on the linear observation matrix, the prior state covariance matrix at the current time, and the constant noise covariance matrix of the zero-velocity measurement. The Kalman gain is calculated based on the linear observation matrix, the prior state covariance matrix at the current time, and the innovation covariance matrix.

7. The pedestrian inertial navigation positioning method based on LSTM zero velocity detection and EKF according to claim 6, characterized in that, The innovation covariance matrix is expressed by the equation: ; The Kalman gain is expressed by the equation: ; In the formula, The new information covariance matrix, For linear observation matrices, Let be the prior state covariance matrix at the current moment. The constant noise covariance matrix for zero-velocity measurements. For Kalman gain. 8.The pedestrian inertial navigation positioning method based on LSTM zero velocity detection and EKF of claim 1, wherein, The corrected error state vector is expressed by the following formula: ; The updated state covariance matrix is represented by the equation: ; In the formula, This is the corrected error state vector. For Kalman gain, To measure the new information, The updated empirical state covariance matrix, For linear observation matrices, Let be the prior state covariance matrix at the current moment.

9. An electronic device, comprising: include: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the pedestrian inertial navigation and positioning method based on LSTM zero-velocity detection and EKF as described in any one of claims 1 to 8.

10. A computer storage medium, characterized in that, The system stores computer-executable instructions for performing the pedestrian inertial navigation and positioning method based on LSTM zero-velocity detection and EKF as described in any one of claims 1 to 8.

Images