Inertial-based autonomous navigation method based on multi-source motion constraints and satellite images

Abstract

The application discloses a kind of inertial base autonomous navigation method based on multi-source motion constraint and satellite image, comprising: constructing elevation calculation model based on satellite image;Local filter is constructed based on zero velocity correction, vehicle kinematic model, vehicle dynamics model and elevation calculation model;Global filter is constructed based on local filter, determine available local filter in current global filter;Each available local filter is time updated and measurement updated, and state estimation result is fused, so as to complete the inertial base autonomous navigation of fusion multi-source motion constraint.The application is applied to navigation positioning field, ensures that the estimation and correction result of navigation error is not affected when the precision of vehicle motion constraint is reduced, improves navigation precision and robustness, and comprehensively utilizes multiple vehicle motion constraints, and realizes the ability of real-time switching available constraint according to driving state through availability evaluation, ensures that reliable constraint correction can be obtained during vehicle driving.

Description

Technical Field

[0001] This invention relates to the field of navigation and positioning technology, specifically an inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery. Background Technology

[0002] Navigation information is fundamental to the autonomous driving of unmanned ground vehicles (UGVs) and is a necessary prerequisite for capabilities such as motion control, environmental perception, and path planning. However, in complex conditions such as wilderness scenarios and adversarial environments, the performance of the most widely used and classic navigation systems, such as Global Navigation Satellite Systems (GNSS), can be severely affected, or even fail for extended periods, leading to a rapid divergence in positioning errors and ultimately rendering the platform unable to operate normally. Therefore, ensuring short-term accuracy and long-term reliability of navigation results under conditions of prolonged satellite and communication signal rejection is of significant research importance and engineering application value.

[0003] Vehicle model-aided navigation (VM-AINS) is a commonly used autonomous navigation solution for UGVs (Urban Vehicles). Vehicle model-aided inertial navigation (VM-AINS), represented by the combination of wheel speed measurement (OD) and inertial navigation system (INS), is the most widely used in-vehicle autonomous navigation method. Currently, VM-AINS commonly uses three types of models: non-holonomic constraints (NHC), vehicle kinematics models (VKM), and vehicle dynamics models (VDM). Based on the different models used, VM-AINS can be further divided into NHC-AINS, VKM-AINS, and VDM-AINS.

[0004] NHC assumes that the vehicle's lateral and vertical velocities are zero, or small-amplitude random noise with a mean of zero. Due to its simplicity and low computational cost, NHC-AINS has become the most widely used vehicle model. Building upon NHC-AINS, researchers use the measured values ​​of the rear wheel OD (Outward Departure) as the forward velocity, combining them with NHC to construct a three-dimensional velocity vector, which is called Holistic Constraint Assisted Navigation. This method is based on the same principle as OD / INS and is also referred to in some studies as Dynamic Zero Velocity Update (ZUPT). This type of method can be considered a simple VKM, but because it does not involve other kinematic knowledge, it can still be classified within the NHC-AINS category. Furthermore, when the vehicle is stationary, the three-dimensional velocity information output by NHC is all zero; in this case, NHC-AINS is equivalent to the Zero Velocity Update (ZUPT) method. VKM describes the vehicle's motion from a geometric perspective, analyzing the changes in state variables such as heading, velocity, and position over time, as well as the relationships between these state variables, through a geometric model. This model is based on the fundamental assumption that tire slippage and sideslip do not occur. Therefore, compared to NHC, VKM-AINS has a wider range of assumptions, higher accuracy, and is more in line with reality. VDM starts with the forces acting on various parts of the vehicle, including longitudinal and lateral forces on the tires, and describes the vehicle's acceleration and angular velocity through a mechanical model, thus constructing differential equations for heading, velocity, and position. Compared to VKM and NHC, VDM-AINS no longer relies on assumptions such as no tire sideslip or slippage, further improving its applicability and output accuracy. However, this method exhibits singularities at low speeds, where navigation errors diverge rapidly and cannot be recovered even after increasing speed. Furthermore, in complex road conditions and maneuvering states, various vehicle motion constraints are severely affected, significantly reducing the accuracy of information provided by these constraints. Individual vehicle motion constraints are difficult to apply to all situations, making it difficult for the vehicle to obtain effective constraint information throughout the entire process. Moreover, the error characteristics of each information source are unknown and change in real time, limiting the fusion effect. Furthermore, VM-AINS primarily provides constraints on the vehicle's planar motion, with weaker constraints on vertical errors. Its elevation errors continue to diverge rapidly, making it difficult to guarantee navigation accuracy over long periods. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention provides an inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery. This method ensures that the estimation and correction results of navigation errors remain unaffected when the accuracy of vehicle motion constraints decreases, thereby improving navigation accuracy and robustness. Furthermore, it comprehensively utilizes various vehicle motion constraints and elevation constraints provided by satellite imagery products, and achieves the ability to switch available constraints in real time according to the driving status through availability assessment, ensuring that reliable constraint corrections can be obtained throughout the entire vehicle driving process.

[0006] To achieve the above objectives, this invention provides an inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery, characterized by the following steps: Step 1: Construct an elevation calculation model based on satellite imagery to calculate the mean and variance of elevation in the area traversed by the vehicle; Step 2: Construct local filters based on the zero-speed correction model ZUPT, the vehicle kinematics model VKM, the vehicle dynamics model VDM, and the elevation calculation model, respectively. Step 3: Construct a global filter based on all the local filters, and determine the available local filters in the current global filter based on the current state information of the carrier; Step 4: Based on the filtering result of the global filter at the previous time step, obtain the initial parameters of each available local filter, and update each available local filter in time. Step 5: Based on the time update results, each available local filter is measured and updated to obtain the state estimate of each available local filter. The local filter corresponding to the zero-speed correction model ZUPT is measured and updated using the standard Kalman filter algorithm. The local filters corresponding to the vehicle kinematics model VKM, the vehicle dynamics model VDM, and the elevation calculation model are adaptively measured and updated using the variational Bayesian filter algorithm based on the central error entropy. Step 6: Fuse the state estimates of each available local filter to obtain the state estimate of the current global filter, thus completing the inertial basis autonomous navigation with fused multi-source motion constraints.

[0007] Compared with the prior art, the present invention has the following beneficial technical effects: 1. This invention proposes a vehicle motion constraint-assisted inertial navigation method based on central error entropy variational Bayesian method, which ensures that the estimation and correction results of navigation error are not affected when the accuracy of vehicle motion constraint decreases, thereby improving navigation accuracy and robustness; 2. This invention comprehensively utilizes multiple vehicle motion constraints and achieves the ability to switch available constraints in real time according to the driving status through availability assessment, ensuring reliable constraint correction throughout the vehicle's driving process; 3. This invention incorporates satellite imagery information, provides elevation constraints, and solves the problem of rapid elevation error divergence in recursive autonomous navigation; 4. This invention proposes a federated adaptive filtering method based on central error entropy variational Bayes and standard Kalman filtering, which improves the fusion accuracy of multi-source heterogeneous information and further ensures the accuracy and reliability of autonomous navigation. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0009] Figure 1 This is a flowchart of the inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the general scope of an embodiment of the present invention; Figure 3 This is a schematic diagram of road similarity analysis of satellite image tiles in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the approximate division of passable and impassable areas within a given scope in an embodiment of the present invention.

[0010] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0012] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0013] like Figure 1 The above describes an inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery disclosed in this embodiment, which mainly includes the following steps: Step 1: Construct an elevation calculation model based on satellite imagery to calculate the mean and variance of elevation in the area traversed by the vehicle; Step 2: Construct local filters based on the zero-speed correction model ZUPT, the vehicle kinematics model VKM, the vehicle dynamics model VDM, and the elevation calculation model, respectively. Step 3: Construct a global filter based on all local filters, and determine the available local filters in the current global filter based on the current state information of the carrier; Step 4: Based on the filtering result of the global filter at the previous time step, obtain the initial parameters of each available local filter, and update each available local filter in time. Step 5: Based on the time update results, each available local filter is measured and updated to obtain the state estimate of each available local filter. Among them, the local filter corresponding to the zero-speed correction model ZUPT is measured and updated using the standard Kalman filter algorithm, while the local filters corresponding to the vehicle kinematics model VKM, the vehicle dynamics model VDM, and the elevation calculation model are adaptively measured and updated using the variational Bayesian filter algorithm based on the central error entropy. Step 6: Fuse the state estimates of each available local filter to obtain the state estimate of the current global filter, thus completing the inertial basis autonomous navigation with fused multi-source motion constraints.

[0014] The elevation channel error divergence rate is higher than that of the horizontal channel in inertial-based autonomous navigation, and reliable elevation constraints are difficult to obtain in autonomous navigation systems. Therefore, this embodiment constructs an elevation calculation model and uses satellite imagery and DSM information to calculate and correct the elevation. In this embodiment, the specific implementation process of calculating the mean and variance of the elevation of the traversable area of ​​the carrier using the elevation calculation model can be divided into three parts: approximate range determination, traversable area segmentation, and elevation information calculation.

[0015] The specific implementation process for determining the general scope is as follows: Based on the real-time latitude and longitude obtained through recursion, the approximate location of the carrier can be determined. approximate location It is a two-dimensional vector whose elements are the latitude and longitude output by the autonomous navigation system at the current moment; Meanwhile, the global filter of the federated filtering can output the posterior state covariance matrix in real time. posterior state covariance matrix The eastward error corresponding to the diagonal in and northward error The elements represent the uncertainty of the current horizontal positioning in the navigation results, which provides the basis for subsequent elevation calculations. Individual eastward and northward errors cannot characterize the actual circular probability error; therefore, this embodiment uses the approximate range radius based on the posterior state covariance matrix, denoted as . : ; This allows us to determine the approximate range of the current actual location of the carrier, that is, by With the center, and A circle with radius , for example Figure 2 As shown.

[0016] The specific implementation process of segmenting the passable area is as follows: Before the mission begins, satellite imagery maps of level 19 or higher are required. Simultaneously, the global path for this mission is determined through a higher-level global planning module, or by the operator directly annotating the satellite imagery. Based on this, an improved UNet network architecture is used to process satellite image patches of size 500×500. After the reconstruction training task is completed, information is extracted from the previous layer of the reconstructed image to generate a feature matrix of size 500×500×64. The penultimate layer of the decoder is chosen as the feature representation layer because this layer not only contains the spatial details recovered through the upsampling step but also incorporates high-level abstract features from the encoding stage. Therefore, each pixel in the image can be represented by a 64-dimensional vector. After obtaining the 64-dimensional features of each pixel in the original image patch, as shown... Figure 3 As shown, firstly, the features of the task path points are clustered to obtain road feature basis vectors. Then, the cosine similarity between the feature basis vectors of the path points in a 500×500 image patch and the surrounding environment is explicitly calculated. After normalizing the similarity, a local drivability map with the same size as the original image can be generated. Then, every 20 meters along the road, a 500×500 image patch centered on the path point is taken to calculate the similarity. The results of each similarity patch are accumulated to finally obtain a global satellite road similarity map, that is, to obtain the drivability score of all points within a general range. For example, Figure 4 This is a schematic diagram showing the approximate division of passable and impassable areas within a given range.

[0017] The specific implementation process for calculating elevation information is as follows: Based on the established approximate location and the segmentation of passable areas, the possible range of the current actual location can be further narrowed by eliminating impassable areas. Using a satellite DSM pre-aligned with satellite imagery, the elevation information of all points within this area can be obtained, denoted as _____. i The elevation of each point is Next, we can use a weighted method to obtain the approximate elevation mean and elevation variance, as follows: ; ; ; in, For the general scope of the first The weight of each point For the general scope of the first The accessibility score for each point The average elevation for a general area. For the general scope of the first Elevation information of each point, The number of points within a general range. The elevation variance represents the approximate range.

[0018] Using elevation calculation models can yield more robust elevation estimation results. Specifically, in flat areas, the elevation changes near the vehicle's location are relatively small, resulting in relatively accurate elevations with smaller variances. In steep mountainous areas, the surrounding elevation changes drastically. Elevation estimation based on accessibility can effectively reduce or eliminate the impact of abnormal areas such as slopes and gullies on the final elevation estimation results, thus obtaining more robust results.

[0019] Elevation estimation based on satellite imagery can provide vertical constraints for autonomous navigation systems, correcting vertical position and velocity. Vehicle motion constraints primarily address planar errors, further suppressing the divergence of horizontal velocity and position errors in the autonomous navigation system. Therefore, in this embodiment, based on the elevation calculation model, local filters are constructed using the Zero-Voltage Correction Model (ZUPT), the Vehicle Kinematics Model (VKM), and the Vehicle Dynamics Model (VDM). The ZUPT, VKM, and VDM models are as follows: ; ; ; in, The velocity vector output by the zero-velocity correction model ZUPT, with superscript... This is the transpose of the matrix. The velocity vector output by the vehicle kinematics model (VKM). , The lateral and longitudinal velocities output by the vehicle kinematics model (VKM) are the vehicle's lateral and longitudinal velocities. The velocity vector output by the vehicle kinematics model (VDM). , The vehicle's lateral velocity and longitudinal velocity are output by the vehicle kinematics model (VDM).

[0020] In the specific implementation process, the vehicle kinematic model VKM adopts a single-track VKM, and the vehicle dynamics model VDM adopts a linear tire model + single-track VDM, without considering the angular motion and load transfer of the sprung parts.

[0021] After selecting the ZUPT, VKM, and VDM models, it is necessary to further analyze their applicability. ZUPT has the most clearly defined applicability; it can only be used when the vehicle is stationary, and this can be quantitatively determined using measurements from the odometer and gyroscope. The applicability of VKM and VDM is mainly determined by the vehicle's speed. Therefore, in this embodiment, determining the available local filters in the current global filter based on the vehicle's current state information specifically includes: First, obtain the current status information of the carrier, including the standard deviation of the odometer's measurement noise. Standard deviation of measurement noise of gyroscope in stationary state The relative speed of the vehicle measured by the odometer Angular velocity about the z-axis of the vehicle body coordinate system measured by a gyroscope And the vehicle's travel speed, including the standard deviation of the measured noise. Typically, the standard deviation of measurement noise of a gyroscope in a stationary state is within the range of 0.05 m / s to 0.15 m / s. Typically in the range of 1e-04 rad / s to 1.5e-03 rad / s; judge and If both conditions are met, the vehicle can be considered stationary, and the local filter corresponding to the zero-speed correction model ZUPT can be used; otherwise, it cannot be used. Determine whether the current vehicle speed is less than the upper speed limit threshold applicable to the vehicle kinematics model (VKM). If yes, the local filter corresponding to the vehicle kinematics model VKM is available; otherwise, it is not available. Determine whether the current vehicle speed exceeds the lower speed threshold applicable to the Vehicle Dynamics Model (VDM). If yes, the local filter corresponding to the vehicle dynamics model (VDM) is available; otherwise, it is not available. The local filter corresponding to the elevation calculation model remains available at all times.

[0022] In practical implementation, for passenger vehicles, the applicable speed upper limit threshold for the vehicle kinematics model (VKM) is... Typically, 15 m / s is taken as the lower speed threshold applicable to the Vehicle Dynamics Model (VDM). Then it can be calculated by the following formula: ; in, For vehicle control cycle, For carrier quality, For the longitudinal stiffness of the front tires, This refers to the longitudinal stiffness of the rear tire.

[0023] In this embodiment, the inertial-based autonomous navigation is based on federated filtering. Federated filtering can be divided into various forms depending on the selection strategy of the information allocation factor. The specific implementation process of the initial parameter settings and time updates for each available local filter in this embodiment is as follows: First, the information allocation factor is calculated based on the number of available local filters: ; in, For the first An information allocation factor for a usable local filter. This represents the number of currently available local filters; Then, based on the information allocation factor and the filtering result of the global filter at the previous time step, the initial parameters of each available local filter are obtained as follows: ; ; ; in, , , The first A number of available local filters are available in The system noise covariance matrix, state covariance matrix, and state estimate at each time step are as follows: , , The global filter is respectively in System noise covariance matrix, state covariance matrix, and state estimate at time step (superscript) Indicates the first i Local filtering, These correspond to the elevation information obtained using the elevation calculation model, and the velocity vectors output by ZUPT, VKM, and VDM, respectively. The time updates for each available local filter are as follows: ; ; ; in, For the first A number of available local filters are available in Time based The state prediction value of the time information, For the first Available local filters from Time's up The state transition matrix at time t, For the first A number of available local filters are available in Time based The predicted state covariance of time-space information. The system noise driving matrix, For the first A number of available local filters are available in The measurement vector at time t. For the first A number of available local filters are available in Measurement matrix at time, For the first A number of available local filters are available in The true state vector at time t, For the first A number of available local filters are available in The measurement noise vector at time step. In practical applications, the time update method of the global filter is the same as that of the local filter, so it will not be described in detail in this embodiment.

[0024] During the measurement update phase, among the four local filters, the ZUPT-based local filter uses the standard Kalman filter algorithm for state estimation, while the other three local filters employ an adaptive state estimation algorithm based on the central error entropy variational Bayesian filter algorithm. The use of the standard Kalman filter algorithm for state estimation is a conventional technique in this field and will not be elaborated upon in this embodiment. The specific real-time process of adaptive measurement update using the central error entropy variational Bayesian filter algorithm includes: First, based on the inverse Wishart distribution degrees of freedom and inverse scaling matrix of the previous time step, calculate the predicted values ​​of the inverse Wishart distribution degrees of freedom and inverse scaling matrix of the current time step, as follows: ; in, for Time based Predicted degrees of freedom of the inverse Wishart distribution for time-series information. for The degrees of freedom of the inverse Wishart distribution output at time step 1. for Time based The predicted value of the inverse scaling matrix of time information. for The inverse scaling matrix output at time step 1. Forgetting factor; Then, the local filter is initialized as follows: ; ; ; ; Among them, superscript ( i ) indicates the first i iteration i =0 indicates the initial time. The initial state estimate, for Time based The state prediction value of the time information, The initial inverse Wishart distribution degrees of freedom, Let be the initial inverse scaling matrix. For the first The estimated value of the observation noise covariance matrix in the next iteration. For the first Degrees of freedom of the inverse Wishart distribution in the next iteration. For the first The inverse scaling matrix of the next iteration The dimension of the observation vector; Then, the Cholesky decomposition is performed, resulting in: ; ; ; ; ; ; in, The joint error covariance matrix, For the joint error vector, for Time based Predicted value of state error covariance for time-space information. To observe the noise covariance matrix, for Cholesky decomposition factor, for Cholesky decomposition factor, To observe the Cholesky decomposition factor of the noise covariance matrix, For Cholesky decomposition operations, for Time based The predicted state error value of the time information. To measure the residual, For state error, for The real-time state of the system; Then, the standardized measurement vector and the standardized measurement matrix are calculated as follows: ; ; in, The standardized measurement vector, This is the preprocessed measurement vector. The standardized measurement matrix, This is the preprocessed measurement matrix; Then, the residual vector and weighting matrix are calculated as follows: ; ; in, For the first The measurement channel is at the first The residual of the next iteration For the first Standardized measurement values ​​for each measurement channel For the first Standardized measurement matrix row vectors for each measurement channel For the first The state estimate of the next iteration. For weighted matrices, , , , This refers to the cross-information between state and measurement; Then, the update iteration gain matrix, prediction covariance matrix, and observation noise covariance matrix are calculated as follows: ; ; ; in, For the first The gain matrix of the next iteration For the first The predicted covariance matrix of the next iteration. For the first The observation noise covariance matrix of the next iteration; Finally, determine if the iteration termination condition has been met. If so, update the degrees of freedom and inverse scaling matrix of the inverse Wishart distribution and output the final result. , , , Otherwise, continue iterating, where , Each serves as a corresponding local filter in The state estimate and state covariance matrix at each time step are then output to the global filter. , These are used as inputs for the next measurement update of the local filter.

[0025] In practice, the iteration termination condition is determined by whether one of the following conditions is met; if so, the iteration ends; otherwise, the iteration continues: ; in, The preset convergence threshold represents the value of the first convergence threshold. The iteration and the When the difference in the next iteration is very small, it is considered that convergence has reached the optimal solution. In this embodiment, we take... ; The preset iteration threshold indicates that reaching the predetermined number of iterations is considered sufficiently accurate.

[0026] In practice, the process of updating the degrees of freedom and inverse scaling matrix of the inverse Wishart distribution is as follows: ; ; ; in, For the first The observation noise covariance matrix of the next iteration. This is the preprocessed measurement vector.

[0027] After completing the measurement updates for all local filters, the state estimates and state covariance matrices of the local filters are transmitted to the global filter. The state estimates of each local filter are then fused to obtain the optimal global estimate, thus completing the inertial-based autonomous navigation with fused multi-source motion constraints. In this embodiment, the state estimate fusion process of the local filters is as follows: ; ; in, For global filters in The state covariance matrix at time t, For each available local filter in The state covariance matrix at time t, For global filters in State estimation at time 10:00 For each available local filter in State estimation at time 10:00 For global filters ink The state prediction covariance matrix at time t. For global filters in k The state prediction vector at time t.

[0028] It is worth noting that, although this embodiment Figure 1 The steps are shown sequentially as indicated by the arrows, but they are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are performed; they can be executed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0029] The above description is only a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the scope of protection of the present invention.

Claims

1. An inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery, characterized in that, Includes the following steps: Step 1: Construct an elevation calculation model based on satellite imagery to calculate the mean and variance of elevation in the area traversed by the vehicle; Step 2: Construct local filters based on the zero-speed correction model ZUPT, the vehicle kinematics model VKM, the vehicle dynamics model VDM, and the elevation calculation model, respectively. Step 3: Construct a global filter based on all the local filters, and determine the available local filters in the current global filter based on the current state information of the carrier; Step 4: Based on the filtering result of the global filter at the previous time step, obtain the initial parameters of each available local filter, and update each available local filter in time. Step 5: Based on the time update results, perform measurement updates on each available local filter to obtain the state estimate of each available local filter. Among them, the local filter corresponding to the zero-speed correction model ZUPT is updated using the standard Kalman filter algorithm, and the local filters corresponding to the vehicle kinematics model VKM, the vehicle dynamics model VDM, and the elevation calculation model are updated using an adaptive measurement update algorithm based on the central error entropy variational Bayesian filter algorithm. Step 6: Fuse the state estimates of each available local filter to obtain the state estimate of the current global filter, thus completing the inertial basis autonomous navigation with fused multi-source motion constraints.

2. The inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery according to claim 1, characterized in that, The process by which the elevation calculation model calculates the mean and variance of the elevation of the area accessible to the vehicle is as follows: Based on the real-time latitude and longitude obtained through recursion, the approximate location of the carrier can be determined. And obtain the posterior state covariance matrix of the real-time output of the global filter. Extract the posterior state covariance matrix The eastward error corresponding to the diagonal in and northward error ; Based on the eastward error and the aforementioned northward error Calculate the approximate range radius and will be Center of the circle A circle with radius is used as a rough approximation of the actual range of the current carrier; Based on satellite imagery, traversable areas are segmented to obtain a global satellite road similarity map, which yields traversability scores for all points within the approximate range. Using a satellite DSM pre-aligned with satellite imagery, elevation information for all points within the passable area is obtained. The mean and variance of the elevation for the approximate area are then calculated using weighted averages. ； ； ； in, For the general scope of the first The weight of each point For the general scope of the first The accessibility score for each point The average elevation for a general area. For the general scope of the first Elevation information of each point, The number of points within a general range. The elevation variance represents the approximate range.

3. The inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery according to claim 1, characterized in that, In step 2, the zero-velocity correction model ZUPT is: ； in, This is the velocity vector output by the zero-velocity correction model ZUPT. This is the transpose of the matrix; The vehicle kinematic model VKM is: ； in, The velocity vector output by the vehicle kinematics model (VKM). , The lateral and longitudinal velocities of the vehicle are output by the vehicle kinematics model (VKM). The vehicle dynamics model VDM is: ； in, The velocity vector output by the vehicle dynamics model (VDM) , The vehicle's lateral velocity and longitudinal velocity are output by the vehicle dynamics model (VDM).

4. The inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery according to claim 1, characterized in that, In step 3, determining the available local filters in the current global filter based on the current state information of the carrier specifically involves: Obtain the current status information of the carrier, including the standard deviation of the odometer's measurement noise. Standard deviation of measurement noise of gyroscope in stationary state The relative speed of the vehicle measured by the odometer Angular velocity about the z-axis of the vehicle body coordinate system measured by a gyroscope and the vehicle's travel speed; judge and Whether both conditions are met simultaneously, if so, the local filter corresponding to the zero-velocity correction model ZUPT is available; otherwise, it is not available. Determine whether the current vehicle speed is less than the upper speed limit threshold applicable to the vehicle kinematics model (VKM). If so, the local filter corresponding to the vehicle kinematics model VKM is available; otherwise, it is not available. Determine whether the current vehicle speed is greater than the applicable lower speed threshold of the Vehicle Dynamics Model (VDM). If so, the local filter corresponding to the vehicle dynamics model VDM is available; otherwise, it is not available. The local filter corresponding to the elevation calculation model remains available at all times.

5. The inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery according to claim 4, characterized in that, The applicable speed lower limit threshold for the Vehicle Dynamics Model (VDM) Specifically: ； in, For vehicle control cycle, For carrier quality, For the longitudinal stiffness of the front tires, This refers to the longitudinal stiffness of the rear tire.

6. The inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery according to any one of claims 1 to 5, characterized in that, Step 4 specifically includes: The information allocation factor is calculated based on the number of available local filters and is as follows: ； in, For the first An information allocation factor for a usable local filter. This represents the number of currently available local filters; The initial parameters of each available local filter are obtained based on the filtering result of the global filter at the previous time step: ； ； ； in, , , The first A number of available local filters are available in The system noise covariance matrix, state covariance matrix, and state estimate at each time step are as follows: , , The global filter is respectively in System noise covariance matrix, state covariance matrix, and state estimate at time 1; The time updates for each available local filter are as follows: ； ； ； in, For the first A number of available local filters are available in Time based The state prediction value of the time information, For the first Available local filters from Time's up The state transition matrix at time t, For the first A number of available local filters are available in Time based The predicted state covariance of time-space information. The system noise driving matrix, For the first A number of available local filters are available in The measurement vector at time t. For the first A number of available local filters are available in Measurement matrix at time, For the first A number of available local filters are available in The true state vector at time t, For the first A number of available local filters are available in The measurement noise vector at time step.

7. The inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery according to claim 6, characterized in that, In step 5, the adaptive measurement update using the variational Bayesian filtering algorithm based on center error entropy specifically involves: Based on the inverse Wishart distribution degrees of freedom and inverse scaling matrix of the previous time step, the predicted values ​​of the inverse Wishart distribution degrees of freedom and inverse scaling matrix of the current time step are calculated as follows: ； in, for Time based Predicted degrees of freedom of the inverse Wishart distribution for time-series information. for The degrees of freedom of the inverse Wishart distribution output at time step 1. for Time based The predicted value of the inverse scaling matrix of time information. for The inverse scaling matrix output at time step 1. Forgetting factor; The local filter is initialized as follows: ； ； ； ； in, The initial state estimate, for Time based The state prediction value of the time information, The initial inverse Wishart distribution degrees of freedom, Let be the initial inverse scaling matrix. For the first The estimated value of the observation noise covariance matrix in the next iteration. For the first Degrees of freedom of the inverse Wishart distribution in the next iteration. For the first The inverse scaling matrix of the next iteration The dimension of the observation vector; Performing Cholesky decomposition, we get: ； ； ； ； ； ； in, The joint error covariance matrix, For the joint error vector, for Time based Predicted value of state error covariance for time-space information. To observe the noise covariance matrix, for Cholesky decomposition factor, for Cholesky decomposition factor, To observe the Cholesky decomposition factor of the noise covariance matrix, For Cholesky decomposition operations, for Time based The predicted state error value of the time information. To measure the residual, For state error, for The real-time state of the system; The standardized measurement vectors and standardized measurement matrices are calculated as follows: ； ； in, The standardized measurement vector, The preprocessed measurement vector, The standardized measurement matrix, This is the preprocessed measurement matrix; The residual vector and weighting matrix are calculated as follows: ； ； in, For the first The measurement channel is at the first The residual of the next iteration For the first Standardized measurement values ​​for each measurement channel For the first Standardized measurement matrix row vectors for each measurement channel For the first The state estimate of the next iteration. For weighted matrices, , , , This refers to the cross-information between state and measurement; The update iteration gain matrix, prediction covariance matrix, and observation noise covariance matrix are calculated as follows: ； ； ； in, For the first The gain matrix of the next iteration For the first The predicted covariance matrix of the next iteration. For the first The observation noise covariance matrix of the next iteration; Determine if the iteration termination condition has been met. If so, update the inverse Wishart distribution degrees of freedom and inverse scaling matrix, and output the final value. , , , Otherwise, continue iterating, where , Each serves as a corresponding local filter in The state estimate and state covariance matrix at each time step are then output to the global filter. , These are used as inputs for the next measurement update of the local filter.

8. The inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery according to claim 7, characterized in that, The iteration termination condition specifically involves determining whether one of the following conditions is met: ； in, The preset convergence threshold, This is a preset threshold for the number of iterations; If yes, end the iteration; otherwise, continue the iteration.

9. The inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery according to claim 6, characterized in that, The process of updating the degrees of freedom and inverse scaling matrix of the inverse Wishart distribution is as follows: ； ； ； in, For the first The observation noise covariance matrix of the next iteration. This is the preprocessed measurement vector.

10. The inertial-based autonomous navigation method based on multi-source motion constraints and satellite imagery according to claim 6, characterized in that, In step 6, the state estimate of the global filter is: ； ； in, For global filters in The state covariance matrix at time t, For each available local filter in The state covariance matrix at time t, For global filters in State estimation at time 10:00 For each available local filter in State estimation at time 10:00 For global filters in k The state prediction covariance matrix at time t. For global filters in k The state prediction vector at time t.

Images