Vehicle navigation positioning method and system when navigation signal is lost

Abstract

This invention discloses a vehicle navigation and positioning method and system when navigation signals are lost, relating to the field of vehicle autonomous positioning technology. The method includes performing tight-coupling matching based on geomagnetic feature maps, obtaining geomagnetic sequence similarity through a sliding window dynamic time warping algorithm, and generating a fused positioning trajectory. The method extracts road network topological connectivity from high-precision maps, constructs a weighted road connection graph Laplacian matrix, and combines this with the fused positioning trajectory to obtain topological constraint correction coefficients through a graph convolutional network. Furthermore, the method constructs a joint optimization function based on the topological constraint correction coefficients and employs an asymmetric Gauss-Newton iterative algorithm to solve for the optimal positioning coordinates. This method overcomes the shortcomings of traditional methods in terms of rigid topological modeling and unstable nonlinear optimization, thereby achieving high-precision and highly robust continuous vehicle positioning in scenarios where GNSS signals are lost.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle autonomous positioning, and in particular to a vehicle navigation positioning method and system when a navigation signal is lost. Background Art

[0002] Traditional global navigation satellite systems (GNSS) are widely adopted because they provide continuous, real-time positioning services worldwide. However, in environments with severe signal obstruction, such as urban canyons, tunnels, and underground garages, GNSS signals are prone to interruption or degradation, resulting in significantly increased positioning errors or even complete failure. To address the shortcomings of GNSS in signal-absent scenarios, researchers have proposed a variety of auxiliary positioning technologies in recent years, such as inertial navigation systems (INS), visual odometry (VO), geomagnetic matching positioning, and constraint methods based on road network topology. Among them, geomagnetic matching positioning has gradually become a research hotspot due to its characteristics such as no need for external infrastructure support and strong anti-interference capabilities.

[0003] While existing technologies have improved vehicle positioning performance in GNSS-restricted environments to a certain extent, several key issues remain to be addressed. First, most geomagnetic matching methods rely on static geomagnetic maps, making them difficult to adapt to dynamically changing road environments. They also lack the ability to model the temporal and spatial evolution of geomagnetic features, which affects matching accuracy and stability. Second, existing fusion positioning algorithms often rely on Kalman filtering when dealing with nonlinear and non-Gaussian noise. These algorithms are prone to divergence when faced with long periods of signal loss and large-scale trajectory drift, resulting in unreliable positioning results. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a vehicle navigation and positioning method when the navigation signal is lost, which solves the problems of static geomagnetic feature modeling and poor stability of nonlinear optimization in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a vehicle navigation and positioning method when the navigation signal is lost, which includes: constructing a space-time tensor field when the navigation signal is lost, predicting the time window when the navigation signal is lost based on the space-time tensor field, and collecting road environment data, and generating a geomagnetic feature map of the current road environment through an adversarial generative network; performing tight coupling matching based on the geomagnetic feature map, and obtaining the similarity of the geomagnetic sequence through a sliding window dynamic time warping algorithm to generate a fused positioning trajectory; extracting the road network topological connection relationship from a high-precision map, constructing a weighted road connection graph Laplacian matrix, and combining the fused positioning trajectory to obtain a topological constraint correction coefficient through a graph convolutional network; constructing a joint optimization function based on the topological constraint correction coefficient, and using an asymmetric Gauss-Newton iterative algorithm to solve the optimal positioning coordinates.

[0008] As a preferred solution of the vehicle navigation and positioning method when the navigation signal is lost according to the present invention, the carrier-to-noise ratio of the satellite navigation signal is continuously monitored, and when the carrier-to-noise ratio is lower than the carrier-to-noise ratio threshold, it is determined that the navigation signal is lost.

[0009] As a preferred solution of the vehicle navigation and positioning method when the navigation signal is lost according to the present invention, the steps of constructing the space-time tensor field when the navigation signal is lost and predicting the time window when the navigation signal is lost according to the space-time tensor field are as follows:

[0010] Collect satellite geometry, ionospheric delay gradient, and multipath interference distribution, align and normalize them, and then fuse them into a space-time tensor field through tensor product.

[0011] Tucker decomposition is used to reduce the dimension of the space-time tensor field, and a state-space model is established through dynamic Bayesian prediction to predict the signal loss window.

[0012] As a preferred solution of the vehicle navigation and positioning method when the navigation signal is lost according to the present invention, wherein: the road environment data includes six-degree-of-freedom data, traffic density and high-precision map road type;

[0013] The six-degree-of-freedom data, traffic density, and high-precision map road types are spliced ​​into feature vectors, and the geomagnetic feature map is obtained through the adversarial generative network generator.

[0014] As a preferred solution of the vehicle navigation and positioning method when the navigation signal is lost according to the present invention, wherein: the method performs tight coupling matching based on the geomagnetic feature map, obtains the geomagnetic sequence similarity through the sliding window dynamic time warping algorithm, and generates a fused positioning trajectory, the specific steps are as follows:

[0015] Tightly couple the geomagnetic feature map with the six-degree-of-freedom motion data to generate the initial positioning trajectory;

[0016] Extract continuous geomagnetic intensity sequences from geomagnetic feature maps and combine them with high-precision maps to form a reference geomagnetic sequence library;

[0017] Retrieve local candidate sequences from the reference geomagnetic sequence library based on the initial positioning trajectory, and generate sequence similarity scores using a dynamic programming algorithm;

[0018] Candidate trajectories are obtained based on the sequence similarity score, and abnormal candidate trajectories are eliminated to generate a fused positioning trajectory.

[0019] As a preferred solution of the vehicle navigation and positioning method when the navigation signal is lost according to the present invention, the road network topology connection relationship is extracted from the high-precision map, a weighted road connection graph Laplacian matrix is ​​constructed, and the topology constraint correction coefficient is obtained through the graph convolution network in combination with the fused positioning trajectory. The specific steps are as follows:

[0020] Analyze lane-level geometry data from high-precision maps and extract lane centerline nodes;

[0021] The topological connection relationship is defined by the lane attributes of the lane centerline nodes, and the lane length and traffic efficiency are extracted. The steering constraint weight is generated through linear superposition;

[0022] Based on the steering constraint weights, a sparse matrix is ​​constructed. A diagonal matrix is ​​generated by accumulating the steering constraint weights of each row of the sparse matrix. The road connection graph Laplace matrix is ​​generated by combining the Laplace matrix formula.

[0023] The symmetric normalized adjacency matrix is ​​obtained according to the Laplacian matrix of the road connection graph, and combined with the fused positioning trajectory, the topological constraint correction coefficient is generated through the graph convolutional network.

[0024] As a preferred solution of the vehicle navigation positioning method when the navigation signal is lost according to the present invention, wherein: the joint optimization function is constructed based on the topological constraint correction coefficient, and the asymmetric Gauss-Newton iterative algorithm is used to solve the optimal positioning coordinates. The specific steps are as follows:

[0025] Based on the fused positioning trajectory, topology constraint correction coefficient and lane centerline node, a joint optimization function is constructed through the objective function;

[0026] The trajectory smoothing term, topology fitting term and observation consistency term in the joint optimization function are constructed as residual vector components respectively;

[0027] Based on the residual components, the Jacobian matrix is ​​constructed through partial derivatives, and the trajectory increment is solved through sparse LU decomposition to generate the optimal positioning coordinates.

[0028] In the second aspect, the present invention provides a vehicle navigation and positioning system when the navigation signal is lost, including a data acquisition module for constructing a space-time tensor field when the navigation signal is lost, predicting the time window of the navigation signal loss based on the space-time tensor field, and collecting road environment data, and generating a geomagnetic feature map of the current road environment through an adversarial generative network; a matching positioning module for performing tight coupling matching based on the geomagnetic feature map, and obtaining the similarity of the geomagnetic sequence through a sliding window dynamic time warping algorithm to generate a fused positioning trajectory; a positioning correction module for extracting the road network topological connection relationship from a high-precision map, constructing a weighted road connection graph Laplacian matrix, and combining the fused positioning trajectory to obtain the topological constraint correction coefficient through a graph convolutional network; a joint optimization module for constructing a joint optimization function based on the topological constraint correction coefficient, and using an asymmetric Gauss-Newton iterative algorithm to solve the optimal positioning coordinates.

[0029] The beneficial effects of the present invention are as follows: by constructing a space-time tensor field to predict the navigation signal loss time window, and combining it with a generative adversarial network to dynamically generate a geomagnetic feature map, the problems of static and poor adaptability of existing geomagnetic modeling are solved; at the same time, a graph convolutional network is introduced to generate dynamic topological constraint correction coefficients, and an asymmetric Gauss-Newton algorithm is used for joint optimization and solution, which overcomes the shortcomings of traditional methods in terms of rigid topological modeling and instability of nonlinear optimization, thereby achieving high-precision and high-robustness continuous vehicle positioning in the GNSS signal loss scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0031] Figure 1 The present invention is a flow chart of a vehicle navigation and positioning method when the navigation signal is lost.

[0032] Figure 2 Schematic diagram of the vehicle navigation and positioning system when the navigation signal is lost.

[0033] Figure 3 Flowchart for constructing the Laplacian matrix of a weighted road connection graph.

[0034] Figure 4 Flowchart for space-time tensor field construction and signal loss window prediction. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0036] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0037] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0038] Reference Figure 1 , is an embodiment of the present invention, which provides a vehicle navigation and positioning method when the navigation signal is lost, comprising the following steps:

[0039] S1: Please refer to Figure 4 , construct the space-time tensor field when the navigation signal is lost, predict the time window when the navigation signal is lost based on the space-time tensor field, collect road environment data, and generate the geomagnetic characteristic map of the current road environment through the adversarial generative network;

[0040] S1.1: Continuously monitor the carrier-to-noise ratio of the satellite navigation signal. When the carrier-to-noise ratio falls below the carrier-to-noise ratio threshold, the navigation signal is considered lost.

[0041] Furthermore, the original carrier-to-noise ratio is collected in real time by the vehicle-mounted GNSS (Global Navigation Satellite System) receiver, and the high-frequency noise interference is suppressed by Hamming window moving average filtering to generate the carrier-to-noise ratio. The carrier-to-noise ratio threshold is calculated by the carrier-to-noise ratio threshold reference value and the dynamic compensation of the satellite elevation angle. The expression is:

[0042] ;

[0043] in, is the carrier-to-noise ratio threshold, is the carrier-to-noise ratio threshold value ( ), is the satellite elevation angle.

[0044] S1.2: Collect satellite geometry, ionospheric delay gradient, and multipath interference distribution, construct a space-time tensor field, and predict the time window of navigation signal loss based on the space-time tensor field;

[0045] Furthermore, the satellite geometric configuration includes the satellite elevation angle and azimuth angle, the ionospheric vertical delay is obtained through the dual-frequency signal transmitted by the satellite, and the multipath intensity distribution is generated through the signal reflection model, and the spatial gradient tensor of the ionospheric vertical delay is used as the ionospheric delay gradient;

[0046] The L1 and L2 frequency pseudorange observations transmitted by the satellite are calculated using pseudorange measurement technology, and the carrier phase smoothing pseudorange method is used to improve the observation accuracy. Based on the inverse square relationship between dual-frequency ionospheric delay and frequency, a dual-frequency ionospheric delay correction model is applied, and the L1 and L2 frequency pseudorange observations are substituted into the inverse square formula to calculate the ionospheric vertical delay.

[0047] A signal reflection model is established based on the principle of geometric optics reflection to extract the multipath interference intensity distribution. The first-order partial derivatives of the ionospheric vertical delay along the longitude, latitude, and elevation directions are calculated using the central difference method. The longitude partial derivative, latitude partial derivative, and elevation partial derivative are arranged in three-dimensional spatial coordinate order to form the ionospheric delay gradient.

[0048] It should be noted that high-precision maps are collected, and the vertical height, horizontal width, surface normal vector of the building facade geometric parameters and the real and imaginary parts of the dielectric constant of the surface dielectric characteristics parameters are extracted through the three-dimensional model vertex coordinate data in the high-precision map; based on the principle of geometric optics reflection, the incident angle and reflection angle are calculated according to the incident direction vector of the satellite signal and the normal vector of the reflecting surface; the Fresnel reflection coefficient formula is used in combination with the dielectric constant parameters and the incident angle to calculate the amplitude attenuation and phase shift of the reflected signal; the geometric length difference between the direct signal propagation path and the reflected signal propagation path is calculated respectively to generate the phase difference; the composite field strength is calculated by combining the amplitude and phase difference of the direct signal and the reflected signal through the complex superposition formula; finally, the logarithm of the power ratio of the composite field strength to the direct signal field strength is taken to generate a spatial distribution result describing the multipath interference intensity.

[0049] Satellite geometry, ionospheric delay gradient, and multipath interference distribution are aligned and normalized and then fused into a space-time tensor field through tensor product.

[0050] Tucker decomposition is used to reduce the dimension of the space-time tensor field, and dynamic Bayesian prediction is used to establish a state space model to predict the signal loss window;

[0051] Furthermore, Tucker decomposition is performed on the space-time tensor field to decompose the three-dimensional space-time tensor field into a core tensor and three factor matrices. The core tensor retains the main characteristic patterns of the space-time tensor field, and the three factor matrices correspond to the time dimension, space dimension, and signal parameter dimension respectively.

[0052] Construct state space models based on dynamic Bayesian prediction framework;

[0053] In the state-space model, three latent state variables are defined: signal strength attenuation state, multipath interference accumulation state, and ionospheric disturbance state. Historical carrier-to-noise ratio data are unsupervisedly clustered using a Gaussian mixture model to generate three discrete state categories: normal, attenuated, and severely attenuated. The K-means clustering algorithm is used to classify the multipath interference intensity into low, medium, and high levels. The total electron content time series of the ionospheric vertical delay is smoothed using a Kalman filter. Periods of ionospheric disturbance are identified using a mutation point detection algorithm, defining two states: stable and disturbed. The transition frequencies between latent state variables at adjacent moments in the historical data are statistically analyzed. Zero-probability events are processed using Laplace smoothing to generate an initial state transition probability matrix. The time dimension factor matrix after Tucker decomposition is processed using a sliding window mean smoothing method to generate a time trend feature sequence. This time trend feature sequence is input into a one-dimensional convolutional neural network and encoded as a state transition propensity vector. This is element-wise multiplied with the initial state transition probability matrix and normalized to generate a dynamically adjusted state transition probability matrix. The satellite geometry and ionospheric delay gradient are then converted into observation likelihood probabilities using a probability density function. Based on the state transition probability matrix and observation likelihood, a forward algorithm is used to obtain the joint probability distribution of the hidden states for future time steps. When severe attenuation, high multipath interference intensity, and ionospheric disturbances occur simultaneously in the joint probability distribution of the hidden states, the signal loss event is determined to have occurred. The optimal path of the hidden state sequence is backtracked using the Viterbi algorithm. Combined with the duration statistics of similar events in historical data, the start time and duration range of the signal loss window are predicted. Using the newly acquired carrier-to-noise ratio, satellite geometry, ionospheric delay gradient, and multipath interference distribution, the state estimation results are recursively corrected using the Bayesian update rule to dynamically adjust the predicted boundaries of the signal loss window.

[0054] S1.3: Generate a geomagnetic feature map of the current road environment through a generative adversarial network;

[0055] Real-time collection of vehicle motion status, environmental perception data, geomagnetic observations and satellite signal prediction information; specifically, the vehicle's six-degree-of-freedom data is collected in real time as the vehicle's motion status through the on-board inertial measurement unit. The vehicle's six-degree-of-freedom data includes three-axis acceleration and angular velocity (yaw, pitch and roll); environmental perception data includes traffic density and high-precision map road type. The number of vehicles within a radius of 100 meters is obtained through the Internet of Vehicles, and the traffic density is extracted by normalizing the real-time number of vehicles with the lane capacity. By reading the high-precision map data, the current road type (such as highway, urban road or tunnel) is extracted and encoded as the high-precision map road type; geomagnetic observations are collected through the three-axis magnetometer carried by the vehicle; satellite signal prediction information is based on the signal loss time window predicted by the space-time tensor field to obtain the start time and frequency of the adversarial generative network generator; the six-degree-of-freedom data, traffic density and high-precision map road type are spliced ​​into an environmental feature vector;

[0056] The adversarial generative network generator specifically includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the environmental feature vector. The hidden layer is composed of three fully connected layers and one transposed convolution layer. The first three fully connected layers are combined with the activation function LeakyReLU (with leaky rectified linear unit) to process the environmental feature vector to generate a one-dimensional environmental feature vector. In the last transposed convolution layer, the one-dimensional environmental feature vector is processed by the reshape (data reshaping) operation to generate the environmental spatial feature. The output layer compresses the environmental spatial feature through convolution to generate a geomagnetic feature map, and uses tanh (hyperbolic positive Cutting function) The activation function will output the corresponding geomagnetic intensity range; a generative adversarial network discriminator is designed, which specifically includes an input layer, a convolutional layer and an output layer. The input layer is used to receive the geomagnetic feature map and the historical geomagnetic feature map, which are spliced ​​into 2-channel input along the channel dimension. The three convolutional layers are combined with the activation function LeakyReLU, and the output layer is flattened and then the fully connected layer is used to output the discrimination result of the geomagnetic feature map. When it is judged to be false, the probability loss is extracted by binary cross entropy, and the weight parameters of the generator are updated through the back propagation algorithm according to the probability loss. When it is judged to be true, the geomagnetic feature map is output.

[0057] S2: Based on the geomagnetic feature map, tight coupling matching is performed, and the similarity of the geomagnetic sequence is obtained through the sliding window dynamic time warping algorithm to generate a fused positioning trajectory;

[0058] S2.1: Tightly couple the geomagnetic feature map with the six-degree-of-freedom motion data to generate an initial positioning trajectory;

[0059] Through linear interpolation, the timestamps of the generation time of the geomagnetic feature map and the sampling time of the vehicle inertial measurement unit are aligned and matched in both directions; the global coordinate system of the geomagnetic feature map generated by the generative adversarial network and the body coordinate system of the vehicle inertial measurement unit are unified through quaternion attitude solution to generate a rotation matrix, which is updated by the angular velocity integral of the vehicle inertial measurement unit. The geomagnetic feature map generated by the generative adversarial network is projected to the body coordinate system through the rotation matrix to generate the theoretical projection value of the geomagnetic feature map, and the theoretical geomagnetic characteristic value is obtained through bilinear interpolation. The difference between the geomagnetic observation value and the theoretical geomagnetic characteristic value is used as the residual vector; the three-axis acceleration is converted to the global coordinate system through the real-time rotation matrix, and the gravity acceleration component is deducted. Based on Newton's law of kinematics, the three-axis acceleration in the global coordinate system is numerically integrated to predict the position and speed of the vehicle in the global coordinate system. and attitude angle; based on the error state Kalman filter, an error state vector including position error, velocity error and attitude angle error is defined, and the predicted vehicle position, velocity and attitude are used as state prediction values, while the residual vector is used as observation data; according to the discrimination result generated by the generative adversarial network discriminator, the discrimination result is converted into an observation noise weight coefficient through an exponential decay function, and the amplitude of the observation noise covariance matrix is ​​dynamically adjusted (the higher the value of the discrimination result, the smaller the noise covariance and the greater the weight of the observation data); based on the covariance matrix of the state prediction value and the adjusted observation noise covariance matrix, the Kalman gain matrix is ​​calculated through the Kalman gain formula, and the observation data and the state prediction value are weightedly fused using the Kalman gain matrix to update the error state vector and the covariance matrix of the error state vector; the corrected vehicle global position, velocity and attitude are generated as the initial positioning trajectory.

[0060] S2.2: Obtain the similarity of geomagnetic sequences through the sliding window dynamic time warping algorithm and generate a fused positioning trajectory;

[0061] Based on the geomagnetic feature map, a continuous geomagnetic intensity sequence is extracted at fixed intervals and combined with high-precision map data to form a reference geomagnetic sequence library. The geomagnetic vector in the vehicle coordinate system is collected in real time using a three-axis magnetometer. The quaternion attitude matrix is ​​calculated based on the attitude of the initial positioning trajectory, and the geomagnetic vector is converted to the global coordinate system and cached in chronological order as a real-time geomagnetic sequence. The sliding window length and step size are set according to the speed of the initial positioning trajectory, and local candidate sequences are retrieved from the reference geomagnetic sequence library based on the global position of the vehicle in the initial positioning trajectory.

[0062] Within a sliding window, the real-time geomagnetic sequence is aligned with the starting points of local candidate sequences in the reference geomagnetic sequence library using the timestamp of the initial positioning trajectory as the reference to construct a cost matrix. The Euclidean distance of each pair of points in the cost matrix is ​​obtained, and a dynamic programming algorithm is used to search for the minimum cumulative cost path within a slope constraint. The slope constraint is dynamically adjusted based on the velocity covariance of the initial positioning trajectory. The cumulative cost path is normalized to the average distance to generate a sequence similarity score.

[0063] The sequence similarity scores are sorted, and the position trajectories corresponding to the Top-K (first K best candidates) candidate reference sequences are selected as candidate trajectories. The weights of the candidate trajectories are adjusted based on the position residuals of the initial positioning trajectory. The coordinates of the candidate trajectories are weighted averaged according to the weights to obtain the fused position. The fused position is used as the observation value of the error state Kalman filter, and the position prediction of the initial positioning trajectory is used as the state prediction value. The prediction-observation residual is calculated and the state estimate is updated. The Mahalanobis distance robustness mechanism is used to eliminate abnormal candidate trajectories based on the covariance matrix of the initial positioning trajectory, and the fused position, velocity, and posture of the current sliding window are generated as the optimized fused positioning trajectory.

[0064] S3: See Figure 3 , extract the road network topology connection relationship from the high-precision map, construct the weighted road connection graph Laplacian matrix, and combine it with the fused positioning trajectory to obtain the topology constraint correction coefficient through the graph convolution network;

[0065] S3.1: Extract road network topology from HD maps and construct a weighted Laplacian matrix of the road connection graph.

[0066] Lane-level geometric data is parsed from the HD map using a standard HD map format parsing tool. The starting points, end points, and intermediate key points of all lane centerlines are extracted as lane centerline nodes for the road connection diagram. Each lane centerline node contains latitude and longitude coordinates and elevation information in the global coordinate system. Lane attributes are attached to each lane centerline node, including lane type (for example, main road, ramp, or emergency lane), lane width, speed limit, and permitted driving direction (for example, straight ahead, left turn, or right turn).

[0067] Define topological connection relationships based on lane attributes of lane centerline nodes and generate steering constraint weights;

[0068] Furthermore, the lane centerline nodes extracted from the high-precision map are used as the basic lane centerline nodes of the road connection diagram, and adjacent edges are established between consecutive intermediate key points belonging to the same lane; for lane centerline nodes of different lanes, when there are physical connection points of lane lines or intersection lane line extension intersections, adjacent edges are established across lanes; the legality of the adjacent edges is judged according to the allowed driving direction in the lane attributes, and when the lane attributes associated with the lane centerline node include a no-turn sign, the adjacent edges in the corresponding direction are deleted; when the lane attribute is a variable lane, the existence and direction constraints of the adjacent edges are dynamically adjusted according to the current time period or traffic signal status, and the traffic is legal. Lane length is generated by calculating the geometric distance between adjacent centerline nodes in the same lane. Time cost is calculated based on the speed limit value in the lane attributes to generate traffic efficiency. When lane attributes prohibit a specific turn, the turning constraint weight is set to the maximum weight and marked as impassable. For variable lanes or signal-controlled sections, time period rules or signal light status are obtained through the real-time traffic data interface. When allowing traffic, the turning constraint weight is generated by linearly superposing the lane length and traffic efficiency. A road connection graph adjacency relationship set containing lane length, traffic efficiency, and turning constraint weight is generated, providing a topological basis for constructing the weighted road connection graph Laplace matrix.

[0069] It should be noted that the specific steps for generating the steering constraint weight by linearly superimposing the lane length and the traffic efficiency when allowing passage are as follows: normalizing the lane length to obtain the lane length coefficient, normalizing the traffic efficiency to obtain the traffic efficiency coefficient, and adding the lane length coefficient and the traffic efficiency coefficient in a fixed ratio to generate the steering constraint weight.

[0070] Based on the steering constraint weights, a sparse matrix and a diagonal matrix are constructed, and the Laplacian matrix of the road connection graph is generated;

[0071] Specifically, according to the steering constraint weights in the defined topological connection relationship, the steering constraint weights of the corresponding positions in the sparse matrix are filled for each pair of lane centerline nodes with adjacent edges, and the unconnected lane centerline node pairs are retained as zero values; the diagonal elements of the diagonal matrix are generated by accumulating the steering constraint weights of each row of the sparse matrix; and the sparse matrix is ​​subtracted from the diagonal matrix to obtain the road connection graph Laplacian matrix.

[0072] S3.2: Combine the fused positioning trajectory and obtain the topology constraint correction coefficient through the graph convolutional network;

[0073] Furthermore, the fused localization trajectory is matched with the lane centerline nodes of the road connectivity graph using nearest neighbor matching to generate path segments consisting of consecutive matching nodes. A symmetric normalized adjacency matrix is ​​generated based on the road connectivity graph Laplacian matrix, and adjacency weights are assigned based on lane steering constraints and lane lengths. A node feature matrix is ​​generated by concatenating the node spatial coordinates, lane attributes, and nearest neighbor matching results. This matrix, along with the normalized adjacency matrix, is fed into a pre-trained graph convolutional network. A two-layer feature propagation operation within the graph convolutional network aggregates the topological constraint information of neighboring nodes and generates topological constraint correction coefficients.

[0074] It should be noted that the training process of the graph convolutional network involves generating a symmetric normalized adjacency matrix based on the road connectivity graph Laplacian matrix as the topological relationship input. A two-layer graph convolutional network structure is used to propagate node features through the adjacency matrix and perform nonlinear transformations. The network parameters are optimized using a mean squared error loss function, enabling the graph convolutional network to accurately learn the topological constraints encoded by the road connectivity graph Laplacian matrix. Ultimately, a graph convolutional network model is obtained that can output topological constraint correction coefficients.

[0075] S4: A joint optimization function is constructed based on the topological constraint correction coefficient, and the asymmetric Gauss-Newton iterative algorithm is used to solve the optimal positioning coordinates;

[0076] S4.1: Perform second-order differences on the coordinates of adjacent points in the fused positioning trajectory to obtain a trajectory smoothing term. Multiply the spatial distance between the fused positioning trajectory point and the corresponding lane centerline node in the Laplacian matrix of the road connection map by the topological constraint correction coefficient to obtain a topological fit term. The deviation between the fused positioning trajectory point and the actual projection value of the geomagnetic feature map is used as the observation consistency term, and a joint optimization function is generated through weighted fusion.

[0077] It should be noted that in the joint optimization function, the trajectory smoothing term ensures the motion continuity of the fused positioning trajectory, the topology fitting term ensures that the trajectory conforms to the road topology connection relationship, and the observation consistency term maintains the consistency of the trajectory and the sensor observation data. The input variable of the joint optimization function is the position coordinate sequence of the fused positioning trajectory points, and the optimization goal is to minimize the output value of the joint optimization function.

[0078] S4.2: Use the asymmetric Gauss-Newton iterative algorithm to solve the optimal positioning coordinates;

[0079] Furthermore, mathematical derivatives are performed on the trajectory smoothing term, topology fit term, and observation consistency term in the joint optimization function. The trajectory smoothing term obtains a derivative matrix by performing second-order differences on the coordinates of adjacent trajectory points. The topology fit term obtains a block diagonal matrix by performing the spatial distance derivatives between the trajectory points and the corresponding nodes of the Laplacian matrix of the road connectivity graph. The observation consistency term obtains a diagonal matrix by performing partial derivatives on the projection values ​​of the geomagnetic feature map. These derivatives are then combined according to the weight ratio of the joint optimization function to form a complete Jacobian matrix structure. The derivatives of the trajectory smoothing term form a tridiagonal block, the derivatives of the topology fit term form a block diagonal structure, and the derivatives of the observation consistency term form a diagonal block. Simultaneously, the residual vector of the joint optimization function at the current trajectory point coordinate is extracted, and the Jacobian matrix and the residual vector are combined to construct a system of linear equations. This linear system is numerically solved using sparse matrix factorization to obtain the incremental update of the trajectory coordinates. During the iterative update process, the weight distribution of the topological constraint correction coefficients in the Jacobian matrix is ​​adjusted based on the node connectivity of the road connection graph Laplacian matrix. The derivative matrix coefficients of the observation items are dynamically scaled based on the output of the geomagnetic feature map discriminator. Each iteration executes a step search algorithm, determining the optimal update step size through cubic polynomial interpolation. The coordinate increments are then superimposed on the coordinates of the current trajectory point according to the optimal step size. The iteration is terminated when the change in the joint optimization function value between two consecutive iterations is less than the preset accuracy requirement or the maximum number of allowed iterations is reached. The final output is the optimal positioning coordinate sequence that minimizes the joint optimization function.

[0080] It should be noted that test data under typical scenarios are collected, including positioning trajectory samples in various environments such as urban roads and highways; the asymmetric Gauss-Newton iterative algorithm is run on the positioning trajectory samples, and the changes in the joint optimization function value in each iteration are recorded; the relative change rate index of the joint optimization function value is used as the function value change rate, and the change law of the function value change rate with the number of iterations is analyzed to draw the convergence curve; the change amount when the function value change rate tends to be stable is selected as the benchmark value, and twice the benchmark value is used as the preset accuracy requirement to ensure that both the positioning accuracy requirements are met and excessive iterations are avoided.

[0081] See also Figure 2This embodiment also provides a vehicle navigation and positioning system when the navigation signal is lost, including: a data acquisition module, used to construct a space-time tensor field when the navigation signal is lost, predict the time window of the navigation signal loss based on the space-time tensor field, and collect road environment data, and generate a geomagnetic feature map of the current road environment through an adversarial generative network; a matching positioning module, used to perform tight coupling matching based on the geomagnetic feature map, and obtain the similarity of the geomagnetic sequence through a sliding window dynamic time warping algorithm to generate a fused positioning trajectory; a positioning correction module, used to extract the road network topological connection relationship from the high-precision map, construct a weighted road connection graph Laplacian matrix, and combine the fused positioning trajectory to obtain the topological constraint correction coefficient through a graph convolutional network; a joint optimization module, used to construct a joint optimization function based on the topological constraint correction coefficient, and use an asymmetric Gauss-Newton iterative algorithm to solve the optimal positioning coordinates.

[0082] This embodiment also provides a computer device, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the vehicle navigation and positioning method when the navigation signal is lost as proposed in the above embodiment.

[0083] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0084] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the vehicle navigation and positioning method when the navigation signal is lost as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0085] In summary, the present invention solves the problems of static and poor adaptability of existing geomagnetic modeling by constructing a space-time tensor field to predict the time window of navigation signal loss, and combining it with a generative adversarial network to dynamically generate geomagnetic feature maps. At the same time, a graph convolutional network is introduced to generate dynamic topological constraint correction coefficients, and an asymmetric Gauss-Newton algorithm is used for joint optimization and solution, overcoming the shortcomings of traditional methods in terms of rigid topological modeling and unstable nonlinear optimization, thereby achieving high-precision and high-robustness continuous vehicle positioning in GNSS signal loss scenarios.

[0086] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A vehicle navigation and positioning method when navigation signals are lost, characterized by: include: Construct a space-time tensor field when the navigation signal is lost, predict the time window when the navigation signal is lost based on the space-time tensor field, collect road environment data, and generate a geomagnetic feature map of the current road environment through a generative adversarial network; Based on the geomagnetic feature map, tight coupling matching is performed, and the similarity of the geomagnetic sequence is obtained through the sliding window dynamic time warping algorithm to generate a fused positioning trajectory; Extract the road network topology from the HD map, construct a weighted road connection graph Laplacian matrix, combine it with the fused positioning trajectory, and obtain the topology constraint correction coefficient through the graph convolutional network. The specific steps are as follows: Analyze lane-level geometry data from high-precision maps and extract lane centerline nodes; The topological connection relationship is defined by the lane attributes of the lane centerline nodes, and the lane length and traffic efficiency are extracted. The steering constraint weight is generated through linear superposition; Based on the steering constraint weights, a sparse matrix is ​​constructed. A diagonal matrix is ​​generated by accumulating the steering constraint weights of each row of the sparse matrix. The road connection graph Laplace matrix is ​​generated by combining the Laplace matrix formula. The symmetric normalized adjacency matrix is ​​obtained based on the Laplacian matrix of the road connection graph, and combined with the fused positioning trajectory, the topological constraint correction coefficient is generated through the graph convolutional network; According to the fused positioning trajectory, topological constraint correction coefficient and lane centerline node, a joint optimization function is constructed through the objective function, and the asymmetric Gauss-Newton iterative algorithm is used to solve the optimal positioning coordinates.

2. The vehicle navigation and positioning method when the navigation signal is lost according to claim 1, characterized in that: By continuously monitoring the carrier-to-noise ratio of the satellite navigation signal, when the carrier-to-noise ratio is lower than the carrier-to-noise ratio threshold, it is determined that the navigation signal is lost.

3. The vehicle navigation and positioning method when the navigation signal is lost according to claim 2, wherein: The steps of constructing the space-time tensor field when the navigation signal is lost and predicting the time window when the navigation signal is lost according to the space-time tensor field are as follows: Collect satellite geometry, ionospheric delay gradient, and multipath interference distribution, align and normalize them, and then fuse them into a space-time tensor field through tensor product. Tucker decomposition is used to reduce the dimension of the space-time tensor field, and a state-space model is established through dynamic Bayesian prediction to predict the signal loss window.

4. The vehicle navigation and positioning method when the navigation signal is lost according to claim 3, wherein: The road environment data includes six-degree-of-freedom data, traffic density, and high-precision map road types; The six-degree-of-freedom data, traffic density, and high-precision map road types are spliced ​​into feature vectors, and the geomagnetic feature map is obtained through the adversarial generative network generator.

5. The vehicle navigation and positioning method when the navigation signal is lost according to claim 4, characterized in that: The method performs tight coupling matching based on the geomagnetic feature map, obtains the geomagnetic sequence similarity through the sliding window dynamic time warping algorithm, and generates a fused positioning trajectory. The specific steps are as follows: Tightly couple the geomagnetic feature map with the six-degree-of-freedom motion data to generate the initial positioning trajectory; Extract continuous geomagnetic intensity sequences from geomagnetic feature maps and combine them with high-precision maps to form a reference geomagnetic sequence library; Retrieve local candidate sequences from the reference geomagnetic sequence library based on the initial positioning trajectory, and generate sequence similarity scores using a dynamic programming algorithm; Candidate trajectories are obtained based on the sequence similarity score, and abnormal candidate trajectories are eliminated to generate a fused positioning trajectory.

6. The vehicle navigation and positioning method when the navigation signal is lost according to claim 1, characterized in that: The asymmetric Gauss-Newton iterative algorithm is used to solve the optimal positioning coordinates. The specific steps are as follows: The trajectory smoothing term, topology fitting term and observation consistency term in the joint optimization function are constructed as residual vector components respectively; Based on the residual components, the Jacobian matrix is ​​constructed through partial derivatives, and the trajectory increment is solved through sparse LU decomposition to generate the optimal positioning coordinates.

7. A vehicle navigation and positioning system when a navigation signal is lost, based on the vehicle navigation and positioning method when a navigation signal is lost according to any one of claims 1 to 6, characterized in that: include: The data acquisition module is used to construct a space-time tensor field when the navigation signal is lost, predict the time window when the navigation signal is lost based on the space-time tensor field, collect road environment data, and generate a geomagnetic feature map of the current road environment through a generative adversarial network; The matching and positioning module is used to perform tight coupling matching based on the geomagnetic feature map, obtain the similarity of the geomagnetic sequence through the sliding window dynamic time warping algorithm, and generate a fused positioning trajectory; The positioning correction module is used to extract the road network topology connection relationship from the high-precision map, construct the weighted road connection graph Laplacian matrix, and combine it with the fused positioning trajectory to obtain the topology constraint correction coefficient through the graph convolution network; The joint optimization module is used to construct a joint optimization function based on the topological constraint correction coefficient and use the asymmetric Gauss-Newton iterative algorithm to solve the optimal positioning coordinates.

Images