Vehicle-mounted high-precision positioning method based on virtual 3D speed adaptive constraint

Abstract

The invention provides a vehicle-mounted high-precision positioning method based on virtual 3D speed adaptive constraint, and relates to the technical field of vehicle navigation. The method comprises the following steps: under the condition that a navigation system GNSS (Global Navigation Satellite System) signal of a vehicle is interrupted, acquiring measurement data of an inertial measurement unit in the vehicle navigation system, constructing a regression model of a virtual 3D speed of the vehicle, establishing a mapping relation between the measurement data and the virtual 3D speed of the vehicle, and generating the virtual 3D speed of the vehicle; constructing an observation model according to the virtual 3D speed, and adaptively optimizing a random model by evaluating the regression precision of the virtual 3D speed; and determining the positioning information of the vehicle according to the observation model and the random model. According to the method, the accuracy of the positioning information provided for the vehicle can be improved under the condition that GNSS positioning is unavailable due to a complex environment.

Description

Technical Field

[0001] The present application relates to the field of vehicle navigation technology, and in particular to a vehicle-mounted high-precision positioning method based on virtual 3D speed adaptive constraints. Background Art

[0002] With the development of technologies such as autonomous driving and intelligent transportation, people's demand for continuous, reliable and high-precision vehicle positioning is increasing. The global navigation satellite system (GNSS) and the inertial navigation system (INS) have natural complementary advantages, which to a certain extent make up for the defects in the positioning performance of a single system and show good performance in navigation applications. In theory, the GNSS / INS combined system can provide continuous and high-precision positioning services in all weather, all day and all airspace. However, when passing through complex environments such as high-rise buildings, tunnels, and viaducts, the GNSS signal will be blocked or even completely lost, resulting in unavailable GNSS positioning results. At this time, the combined system will be forced to enter the pure INS mode. Due to the lack of GNSS observation information to suppress INS error drift, the positioning result will significantly decrease over time and cannot provide accurate navigation information for the driver. Therefore, there is a problem in the existing technology that the vehicle positioning information is inaccurate when the environment is complex and GNSS positioning is unavailable. Summary of the invention

[0003] The present application provides a vehicle-mounted high-precision positioning method based on virtual 3D velocity adaptive constraints, which can improve the accuracy of positioning information provided to the vehicle when GNSS positioning is unavailable in a complex environment.

[0004] In a first aspect, an embodiment of the present application provides a vehicle-mounted high-precision positioning method based on virtual 3D velocity adaptive constraints, which may include: When the GNSS signal of the navigation system of the vehicle is interrupted, obtaining measurement data of an inertial measurement unit in the vehicle, constructing a regression model of a virtual 3D speed of the vehicle, establishing a mapping relationship between the measurement data and the virtual 3D speed of the vehicle, and generating a virtual 3D speed of the vehicle; An observation model is constructed according to the virtual 3D velocity, and a random model is adaptively optimized by evaluating the regression accuracy of the virtual 3D velocity; The positioning information of the vehicle is determined according to the observation model and the random model.

[0005] In the above implementation, the virtual 3D speed of the vehicle is generated by measuring data, and then an observation model is constructed based on the virtual 3D speed to provide the vehicle's positioning information when the navigation signal is interrupted. This can improve the stability of the on-board navigation and improve the accuracy of the positioning information provided to the vehicle when GNSS positioning is unavailable in a complex environment.

[0006] In some embodiments, when a GNSS signal of a navigation system of the vehicle is interrupted, acquiring measurement data of an inertial measurement unit in the vehicle and generating a virtual 3D speed of the vehicle according to the measurement data may include: Processing the observed data into a plurality of numerical sequences based on a sliding window; wherein the numerical sequences have a preset length and a preset depth and are arranged in one dimension; The plurality of numerical sequences are used as inputs of a preset hybrid network model to output a virtual 3D speed of the vehicle according to the preset hybrid network model.

[0007] In the above implementation, the complex observation data can be cut into multiple numerical sequences with fixed length and depth through the sliding window. These numerical sequences have a certain continuity in time or space, which can reflect the dynamic changes of the vehicle's motion state, so that the subsequent network model can process data more efficiently. By preprocessing the numerical sequence, noise and redundant information can be reduced, data quality can be improved, and the virtual 3D speed of the vehicle can be generated more accurately, thereby improving the accuracy of the positioning information provided to the vehicle.

[0008] In some embodiments, constructing an observation model according to the virtual 3D velocity may include: determining the observed velocity based on a velocity and attitude generated by an inertial navigation system; The observation model is constructed according to the virtual 3D velocity and the observation velocity.

[0009] The observation model includes: in, Characterizes the difference between the observed speed calculated by the inertial navigation system in the vehicle and the virtual 3D speed, is the observed speed calculated by the inertial navigation system, is the virtual 3D speed, is the observation coefficient matrix, X is the state variable vector, which contains all the error states that need to be estimated by the vehicle's integrated navigation system, Characterize the target observation noise covariance matrix under adaptive constraints.

[0010] In the above implementation, by directly executing 3D speed constraints on the vehicle in the observation domain without the need to compensate for the installation angle, compared with the traditional constraint method, the accuracy loss of constraint information and the system cost are reduced, and the complex vehicle motion state and road conditions can be better coped with. The stability of the vehicle kinematic constraints is further improved, thereby improving the accuracy of the positioning information provided to the vehicle when GNSS positioning is unavailable in complex environments.

[0011] In some embodiments, constructing an observation model according to the virtual 3D velocity and adaptively optimizing a random model by evaluating the regression accuracy of the virtual 3D velocity may also include: Determine the initial observation noise covariance matrix; The observation noise covariance matrix is ​​updated according to the difference between the observation speed and the virtual 3D speed to determine the target observation noise covariance matrix when the updating is completed.

[0012] In the above implementation, by constructing an adaptive model, the influence of observation noise changes and outlier observations on the filtering estimation results can be effectively suppressed, the filtering ability of the combined system for various noises in complex environments can be improved, and the accumulation of INS errors can be effectively suppressed, so as to further improve the accuracy of the positioning information provided to the vehicle when GNSS positioning is unavailable in complex environments, and provide continuous high-precision seamless positioning services for vehicle-mounted navigation.

[0013] In some embodiments, determining the positioning information of the vehicle according to the observation model and the random model may include: estimating an error state of the inertial navigation system based on the observation model; The error state is fed back to the current state estimation of the inertial navigation system to update the state of the inertial navigation system and obtain the positioning information of the vehicle.

[0014] In the above implementation, the error state of the inertial navigation system can be accurately determined through the observation model, which helps to timely discover and compensate for errors within the system, avoid the accumulation of errors and the divergence of positioning accuracy, and further improve the accuracy of the positioning information provided to the vehicle when GNSS positioning is unavailable in complex environments, thereby providing continuous, high-precision, seamless positioning services for on-board navigation.

[0015] In some embodiments, the preset length of the numerical sequence is a sampling rate of the inertial measurement unit.

[0016] In a second aspect, an embodiment of the present application provides a vehicle-mounted high-precision positioning device based on virtual 3D velocity adaptive constraints, which may include: A generating module, used for obtaining measurement data of an inertial measurement unit in the vehicle when a GNSS signal of a navigation system of the vehicle is interrupted, constructing a regression model of a virtual 3D velocity of the vehicle, establishing a mapping relationship between the measurement data and the virtual 3D velocity of the vehicle, and generating a virtual 3D velocity of the vehicle; A construction module, used to construct an observation model according to the virtual 3D velocity, and adaptively optimize the random model by evaluating the regression accuracy of the virtual 3D velocity; A positioning module is used to determine the positioning information of the vehicle according to the observation model and the random model.

[0017] In a third aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the above description when executing the computer program.

[0018] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the muscle function assessment method as described in the above description is implemented.

[0019] Compared with the prior art, the beneficial effects of the present application are: generating a virtual 3D speed of the vehicle by measuring data, and then constructing an observation model based on the virtual 3D speed to provide the vehicle's positioning information when the navigation signal is interrupted, which can improve the stability of the on-board navigation and improve the accuracy of the positioning information provided to the vehicle when GNSS positioning is unavailable in a complex environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of the steps of a vehicle-mounted high-precision positioning method based on virtual 3D velocity adaptive constraints provided in an embodiment of the present application.

[0021] Figure 2 A schematic diagram of a vehicle-mounted high-precision positioning device based on virtual 3D velocity adaptive constraints provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] The present application is further described in detail below in conjunction with test examples and specific implementation methods. However, this should not be understood as the scope of the above subject matter of the present application being limited to the following embodiments, and all technologies implemented based on the content of the present application belong to the scope of protection of the present application.

[0023] Unless otherwise specified, in the description of the specific embodiments of the present application, the terms indicating the orientation or position relationship such as "up", "down", "left", "right", "center", "inside", "outside", "side", etc. are all expressions based on the orientation or position relationship shown in the drawings, or the orientation or position relationship when the product / equipment / device is usually used. These terms of orientation or position relationship are only for the convenience of describing the scheme of the present application or simplifying the description in the specific embodiments to facilitate the technicians to quickly understand the scheme, rather than indicating or implying that a specific device / component / element must have a specific orientation, or be constructed and operated in a specific position relationship, and therefore cannot be understood as a limitation on the present application.

[0024] In the description of the embodiments of the present application, the technical terms "first", "second", etc. only distinguish one entity or operation from another entity or operation, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0025] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0026] During the research, the applicant found that the 3D speed constraint composed of the forward speed constraint provided by the odometer (ODO) and the non-holonomic constraints (NHC) proved to be the most practical vehicle auxiliary positioning solution during GNSS interruption, which can effectively suppress the INS error divergence and significantly improve the positioning accuracy. However, the odometer assistance needs to obtain speed information from the controller local area network or additional wheel encoding information, which will greatly increase the cost and complexity of the system. NHC believes that if there is no side slip or jump during the driving process of the vehicle, the lateral and vertical speeds will be very close to 0. It is a virtual speed observation of land vehicles and can be achieved without any additional sensors. However, to fully realize the potential of NHC constraints, it is necessary to accurately compensate for the IMU installation angle. If ignored or improperly handled, it will reduce the effect of NHC constraints and even have an adverse effect on navigation accuracy. However, the IMU installation angle is difficult to measure accurately. Some scholars have proposed to use Kalman filtering to estimate it. Although the estimation accuracy can meet the positioning requirements, frequent disassembly leads to the need to re-estimate and calibrate the installation angle before each positioning solution. Moreover, the vehicle will inevitably experience complex movements such as skidding and bouncing during driving, which will also make the traditional NHC assumption no longer accurate.

[0027] Example 1 Based on this, an embodiment of the present application provides a vehicle-mounted high-precision positioning method based on virtual 3D speed adaptive constraints.

[0028] Please see Figure 1 , Figure 1 A schematic diagram of the steps of a vehicle-mounted high-precision positioning method based on virtual 3D speed adaptive constraints provided in an embodiment of the present application. The vehicle-mounted high-precision positioning method based on virtual 3D speed adaptive constraints may include: S1. When the GNSS signal of the vehicle's navigation system is interrupted, obtain measurement data of an inertial measurement unit in the vehicle, construct a regression model of the vehicle's virtual 3D speed, establish a mapping relationship between the measurement data and the vehicle's virtual 3D speed, and generate the vehicle's virtual 3D speed.

[0029] S2. constructing an observation model according to the virtual 3D velocity, and adaptively optimizing the random model by evaluating the regression accuracy of the virtual 3D velocity.

[0030] S3. Determine the positioning information of the vehicle according to the observation model and the random model.

[0031] The method provided in the embodiment of the present application can be applied to a GNSS / INS combination system, hereinafter referred to as a combined system. Navigation signal interruption is a condition that characterizes the interruption of the GNSS signal. An inertial measurement unit (IMU) is a device that measures the three-axis attitude angle (or angular rate) and acceleration of a vehicle. It can usually be composed of three single-axis accelerometers and three single-axis gyroscopes, and these sensors are orthogonal to each other to provide measurement data. The measurement data of the inertial measurement unit may include linear acceleration and rotational angular rate in three directions.

[0032] In an embodiment of the present application, when the navigation signal is interrupted, motion information such as the acceleration and angular velocity of the vehicle can be obtained from the inertial measurement unit, and the virtual 3D speed of the vehicle can be directly predicted from the observation value domain through these motion information. Then, the observation model of the vehicle is constructed based on the virtual 3D speed. Without considering the IMU installation angle and the wheel odometer, it helps to convert the measurement data of the inertial measurement unit into speed information that is closer to the actual situation, thereby providing constraint information for vehicle positioning and obtaining reliable and high-precision positioning information.

[0033] In the above implementation, the virtual 3D speed of the vehicle is generated by measuring data, and then an observation model is constructed based on the virtual 3D speed to provide the vehicle's positioning information when the navigation signal is interrupted. This can improve the stability of the on-board navigation and improve the accuracy of the positioning information provided to the vehicle when GNSS positioning is unavailable in a complex environment.

[0034] Example 2 This embodiment is a specific implementation scheme for generating a virtual 3D speed of a vehicle in the above embodiment 1, and the scheme may include: S11. Processing the observed data into multiple numerical sequences based on a sliding window; wherein the numerical sequences have a preset length and a preset depth and are arranged in one dimension.

[0035] In the embodiment of the present application, since a single IMU observation value is not sufficient to mine the dynamic characteristics of the data, a sliding window can be used to slice all input data into multiple sequences. Assuming that the sampling rate of the IMU data set is aHZ, each input sequence can be regarded as a 1D vector with a length of a and a depth of 6. A 1D vector is a sequence of values ​​arranged in only one dimension. In the embodiment of the present application, the 6 IMU observation values ​​(depth of 6) of each sampling point are organized into a 1D vector, that is, an array containing 6 elements.

[0036] S12. Using the multiple numerical sequences as inputs of a preset hybrid network model to output a virtual 3D speed of the vehicle according to the preset hybrid network model.

[0037] To improve the prediction accuracy, the IMU observations can be normalized and connected to six channels as the input of the preset hybrid network model. The input and output models of the preset hybrid network model can be expressed as: in, and They respectively represent the IMU measurement values ​​after normalization, F represents the nonlinear regression model learned by the hybrid network, and are the lateral, forward and vertical regression speeds of the vehicle, that is, the virtual 3D speed of the vehicle in the embodiment of the present application.

[0038] In the embodiment of the present application, the hybrid network model is mainly composed of three parts: feature extraction layer, encoding layer and output layer. The three one-dimensional convolutional layers and the subsequent maximum pooling layer can be regarded as feature extraction components. The convolutional layer can extract the original features of the input data, and the maximum pooling layer selectively extracts key features and reduces the calculation parameters. Each one-dimensional convolutional network is processed with a linear ReLU activation function to enhance the model's ability to handle complex nonlinear relationships. The encoding layer draws on the structure of the Transformer and contains two sub-layers: a multi-head attention layer and a feedforward layer. Each sub-layer is followed by a residual connection and a regularization layer. The multi-head attention layer actually integrates a self-attention mechanism. The output of each attention head It can be expressed as: Where i=1,2,… ; Q is the query vector; K is the key vector; V is the value vector; , and is the weight matrix of the linear transformation, which is used to map the input sequence to the query, key and value spaces respectively; is the dimension of the key vector K. When scaling in the softmax function, use as the denominator to avoid When is large, the gradient of the softmax function becomes too small. Softmax is a commonly used activation function used to convert the original score into a probability distribution. In the embodiment of the present application, it is used to calculate the matching degree between each key vector K and the query vector Q. is the key matrix The transpose of . Indicates the use of query Q i , key K i Sum value V i Compute the attention function of the i-th attention head.

[0039] During calculation, the input sequence is transformed linearly to obtain the corresponding , , matrix, thus converting the input Q i , K i and V i Vector from d model Dimension is mapped to d q ,d k and d v The dimension of the linear transformation matrix of dimension . Where R is a set of real numbers, , , Q i , K i and V i The linear transformation matrix, d model Represents the feature dimension of each frame vector, d q ,d k and d v Q i , K i and V i The dimension, usually d q =d k =d v =d model / n head The self-attention mechanism is implemented using a scaled dot product layer. After multiple parallel calculations, the attention calculation results of all heads are concatenated, and then a linear transformation layer is used to integrate the information captured by different heads: Among them, Concat is the concatenation function, is the parameter matrix of the final linear transformation of the multi-head attention mechanism, indicating that this linear transformation matrix maps the nd-dimensional input (concatenated from the outputs of multiple heads) to d model dimensional output. nd is the number of "heads" in the multi-head attention. The merged features are input into the feedforward network, which is the core component of each Transformer encoding unit and consists of two fully connected layers with a ReLU activation function inserted in between. Then, the summed result is normalized through the layer to ensure that the data maintains a uniform scale and distribution before the next processing step, thereby improving the training stability and efficiency of the model. Finally, it is mapped to the three-dimensional velocity of the carrier through a one-dimensional convolutional network and two fully connected layers. In order to prevent overfitting of the model, a regularization layer is added before the fully connected layer.

[0040] In the above implementation, the complex observation data can be cut into multiple numerical sequences with fixed length and depth through the sliding window. These numerical sequences have a certain continuity in time or space, which can reflect the dynamic changes of the vehicle's motion state, so that the subsequent network model can process data more efficiently. By preprocessing the numerical sequence, noise and redundant information can be reduced, data quality can be improved, and the virtual 3D speed of the vehicle can be generated more accurately, thereby improving the accuracy of the positioning information provided to the vehicle.

[0041] Example 3 This embodiment is a specific implementation of constructing an observation model according to the virtual 3D velocity in the above embodiment 1. The step of constructing an observation model according to the virtual 3D velocity may include: S21. Determine the observation speed according to the speed and attitude calculated by the inertial navigation system.

[0042] S22, constructing an observation model according to the virtual 3D speed and the observation speed.

[0043] During the GNSS signal interruption, the state model of the combined system can still be established using the INS error equation, and the vehicle state variable X can be expressed as: In the embodiment of the present application, the geographic coordinate system of east-north-sky is adopted, wherein the superscript n represents the navigation coordinate system, , , They represent the attitude angle error vector, velocity error vector and position error vector of INS respectively. and They represent the bias errors of the gyroscope and accelerometer respectively.

[0044] Without considering the installation angle compensation, the observation information is directly provided by the 3-D velocity predicted by the hybrid network, that is, in Indicates the real 3D speed of the vehicle, is the speed prediction error, which represents the error between the 3D speed predicted by the hybrid network and the actual speed. Represents the vehicle's velocity component in the east (or lateral) direction. In a geographic coordinate system, the east direction is usually parallel to the longitude line. Represents the vehicle's velocity component in the north (or longitudinal) direction. In a geographic coordinate system, north usually points to the geographic North Pole. It represents the velocity component of the vehicle in the vertical direction (sky direction), and is usually used to describe the lifting and lowering motion of the vehicle.

[0045] Similarly, when the installation angle is not considered, the velocity calculated by INS is projected onto the V system and can be expressed as: in, The velocity calculated by INS is projected to the V system (vehicle coordinate system). is the attitude matrix, which is used to transform the vector in the navigation coordinate system (n system) to the carrier coordinate system (b system). The identity matrix I is a 3x3 matrix whose elements on the diagonal are all 1 and the rest are all 0. Represented by the attitude angle error vector The antisymmetric matrix is ​​used to describe the attitude angle error to the INS velocity vector impact. represents the vehicle velocity vector in the navigation coordinate system (n system), Indicated by The antisymmetric matrix formed is is the gyroscope measurement value in the carrier coordinate system, For the NHC lever arm, it can usually be measured accurately. Represents the angular velocity vector of the carrier relative to the inertial system The antisymmetric matrix.

[0046] The observation equation of the virtual 3D velocity constraint can be expressed as: in, Characterize the difference between the observed velocity and the virtual 3D velocity calculated by the inertial navigation system in the vehicle; is the observed velocity calculated from the inertial navigation system; is the virtual 3D speed; is the observation coefficient matrix, which is constructed based on the error equation of INS and the measurement model of the speed sensor; X is the state variable vector, which contains all the error states that need to be estimated by the vehicle's integrated navigation system. Characterize the target observation noise covariance matrix under adaptive constraints.

[0047] The observation coefficient matrix is ​​expressed as follows: in, It represents a 1-row, 3-column zero matrix, i.e. a matrix whose elements are all 0, used to fill certain parts of the matrix to ensure that the dimensions of the matrix match.

[0048] During the research, the applicant found that in the navigation system, the installation angle usually refers to the angular deviation between the installation direction of the sensor (such as a gyroscope, accelerometer) and the reference direction of the carrier (such as a vehicle). If these deviations are not compensated, they will lead to navigation errors.

[0049] In the embodiment of the present application, the observation information is directly provided by the virtual 3D speed predicted by the hybrid network. This speed information is relative to the vehicle itself, rather than to a specific sensor installation direction. Therefore, even if the sensor has an installation angle deviation, this deviation will not affect the speed information predicted by the hybrid network. Among the state variables, the present application focuses on the attitude angle error vector, velocity error vector, position error vector of the INS, and the zero bias error of the gyroscope and accelerometer. These state variables do not directly involve the installation angle of the sensor. When constructing the observation equation, the present application uses the difference between the velocity calculated by the INS and projected to the vehicle coordinate system and the virtual 3D velocity predicted by the network. This difference is based on the motion state of the vehicle itself, rather than the installation direction of the sensor.

[0050] In the above implementation, by directly executing 3D speed constraints on the vehicle in the observation domain without the need to compensate for the installation angle, compared with the traditional constraint method, the accuracy loss of constraint information and the system cost are reduced, and the complex vehicle motion state and road conditions can be better coped with. The stability of the vehicle kinematic constraints is further improved, thereby improving the accuracy of the positioning information provided to the vehicle when GNSS positioning is unavailable in complex environments.

[0051] Example 4 This embodiment is the step of adaptively optimizing the random model by evaluating the regression accuracy of the virtual 3D velocity in Embodiment 1, reducing the influence of the observation error on the positioning result, and improving the robustness of the system, including: S23. Determine the initial observation noise covariance matrix.

[0052] S24. Update the observation noise covariance matrix according to the difference between the observation speed and the virtual 3D speed, so as to determine the target observation noise covariance matrix when the update is completed.

[0053] In view of the inevitable errors or even gross errors in virtual 3D velocity, the embodiment of the present application proposes an improved low-complexity measurement noise adaptive model. The adaptive factor can be calculated according to the prediction residual to realize the adaptive measurement noise matrix and enhance the robustness of the positioning system.

[0054] In the embodiment of the present application, the Kalman filtering algorithm is used to reduce the measurement noise. First, the prior observation noise covariance matrix R can be set according to the error state of the speed prediction. V0 : Among them, e predict It can be obtained through the accuracy of neural network training and the error of testing, but it can only reflect the overall accuracy and error of the prediction model, and the error value is constantly changing. Therefore, it is necessary to calculate the adaptive factor to adjust R in real time. V0 .

[0055] The specific steps for updating the state of the inertial navigation system by adding the error state to the current state estimate of the inertial navigation system are as follows: In Kalman filtering, the new information r k It is expressed as: Among them, H k is the observation matrix, represents the predicted value of the state at time k at time k-1, e k represents the observation noise at time k, which is used to describe the uncertainty or error in the observation value.

[0056] According to the covariance matrix propagation theory, the new information r k The initial observation noise covariance matrix can be expressed as: Among them, C k represents the new information r at time k k The covariance matrix of describes the uncertainty of the new information, that is, the uncertainty of the difference between the observed value and the predicted value. k / k-1 represents the uncertainty of the state prediction at time k at time k-1, is the transpose of the observation matrix, R vk represents the covariance matrix of the observation noise at time k.

[0057] In order to solve the problem that the error of speed prediction value is too large to affect the accuracy of filtering estimation, the Sage windowing method is usually used to smooth the estimation to obtain the actual new information covariance matrix, but the new information of historical epochs needs to be stored, which will increase the calculation burden and affect the efficiency of system operation. The window size needs to be determined by empirical values. Based on this, this application proposes an improved low-complexity estimation model. First, construct C k The equally weighted recursive model of : in, represents the new information covariance matrix estimated by the equally weighted recursive model at time k; Indicates new information k With its transposed matrix The product of the mathematical expectation of ; k is the time index, which is used to identify and distinguish data at different time points; is the weight coefficient in the equal-weighted recursive model, Represents the recursive part based on the estimated value of the innovation covariance matrix at the previous moment. This part updates the estimated value at the previous moment by multiplying it by a recursive coefficient, 1−1 / k, to reflect the new information at the current moment. Represents the contribution of the new information at the current moment to the estimated value of the covariance matrix.

[0058] In order to ensure stable estimation effect and appropriate adaptive ability, the equal weighting is replaced by exponential fading memory weighting: Among them, β k represents the exponential fading memory weighting coefficient at time k, called the fading factor, which is usually taken as b=0.9:0.99. The above fading memory weighted estimation model does not need to store the information of historical epochs and has a more stable estimation ability. Then the estimated value of the observation noise covariance matrix is: Due to the existence of the negative sign, the above model is prone to negative determination, causing filtering anomalies. The usual practice is to delete the second term to make it available. However, this method may lead to inaccurate error estimation, which brings difficulties and uncertainty to subsequent estimation. To this end, the following adaptive factor is constructed: Among them, tr(.) represents the trace operation of the matrix, that is, the accumulation of the elements on the diagonal of the matrix. Then, the noise covariance matrix of the speed constraint under different states is obtained by adaptively adjusting the prior noise through the adaptive factor: In the above implementation, by constructing an adaptive model, the influence of observation noise changes and outlier observations on the filtering estimation results can be effectively suppressed, the filtering ability of the combined system for various noises in complex environments can be improved, and the accumulation of INS errors can be effectively suppressed, thereby further improving the accuracy of the positioning information provided to the vehicle when GNSS positioning is unavailable in complex environments, providing continuous, high-precision, seamless positioning services for vehicle-mounted navigation.

[0059] Example 5 This embodiment is the above embodiment 1, and determining the vehicle positioning information according to the observation model and the random model may include: Determine the error state of the inertial navigation system based on the observation model; The error state is added to the current state estimate of the inertial navigation system to update the state of the inertial navigation system and obtain the positioning information of the vehicle.

[0060] In the above implementation, the error state of the inertial navigation system can be accurately determined through the observation model, which helps to timely discover and compensate for errors within the system, avoid error accumulation and system divergence, and further improve the accuracy of the positioning information provided to the vehicle when GNSS positioning is unavailable in complex environments, thereby providing continuous, high-precision, seamless positioning services for on-board navigation.

[0061] Example 6 Please see Figure 2 , Figure 2 A schematic diagram of a vehicle-mounted high-precision positioning device based on virtual 3D speed adaptive constraints provided in an embodiment of the present application. The vehicle-mounted high-precision positioning device 20 based on virtual 3D speed adaptive constraints may include: A generating module 21 is used for acquiring measurement data of an inertial measurement unit in the vehicle when a GNSS signal of a navigation system of the vehicle is interrupted, constructing a regression model of a virtual 3D velocity of the vehicle, establishing a mapping relationship between the measurement data and the virtual 3D velocity of the vehicle, and generating a virtual 3D velocity of the vehicle; A construction module 22 is used to construct an observation model according to the virtual 3D velocity, and adaptively optimize the random model by evaluating the regression accuracy of the virtual 3D velocity; The positioning module 23 is used to determine the positioning information of the vehicle according to the observation model and the random model.

[0062] In some embodiments, the generating module 21 may be specifically used for: Processing the observed data into a plurality of numerical sequences based on a sliding window; wherein the numerical sequences have a preset length and a preset depth and are arranged in one dimension; The plurality of numerical sequences are used as inputs of a preset hybrid network model to output a virtual 3D speed of the vehicle according to the preset hybrid network model.

[0063] In some embodiments, the building block 22 may be specifically used to: Determining the observation speed based on the speed and attitude of the inertial navigation system; The observation model is constructed based on the virtual 3D velocity and the observed velocity.

[0064] In some embodiments, the building block 22 may also be specifically used to: Determine the initial observation noise covariance matrix; The observation noise covariance matrix is ​​updated according to the difference between the observation velocity and the virtual 3D velocity to determine the target observation noise covariance matrix when the updating is completed.

[0065] In some embodiments, the positioning module 23 may be specifically used for: Determine the error state of the inertial navigation system based on the observation model; The error state is added to the current state estimate of the inertial navigation system to update the state of the inertial navigation system and obtain the positioning information of the vehicle.

[0066] It should be understood that the various modules of the vehicle-mounted high-precision positioning device based on virtual 3D velocity adaptive constraints provided in the above embodiments are only illustrated by the division of the functional modules in the above description. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0067] The functional modules in the above embodiments may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit, and the above integrated unit may be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the embodiments of the present application.

[0068] Based on the same application concept, an embodiment of the present application also provides a computer device, which may include a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the method described in the above description is implemented.

[0069] Based on the same application concept, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method described in the above description is implemented.

[0070] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A vehicle-mounted high-precision positioning method based on virtual 3D velocity adaptive constraints, characterized in that: include: When the GNSS signal of the navigation system of the vehicle is interrupted, obtaining measurement data of an inertial measurement unit in the vehicle, constructing a regression model of a virtual 3D speed of the vehicle, establishing a mapping relationship between the measurement data and the virtual 3D speed of the vehicle, and generating a virtual 3D speed of the vehicle; An observation model is constructed according to the virtual 3D velocity, and a random model is adaptively optimized by evaluating the regression accuracy of the virtual 3D velocity; The positioning information of the vehicle is determined according to the observation model and the random model.

2. The method according to claim 1, characterized in that In the case where the GNSS signal of the navigation system of the vehicle is interrupted, the measurement data of the inertial measurement unit in the vehicle is obtained, a regression model of the virtual 3D speed of the vehicle is constructed, a mapping relationship between the measurement data and the virtual 3D speed of the vehicle is established, and the virtual 3D speed of the vehicle is generated, including: Processing the observed data into a plurality of numerical sequences based on a sliding window; wherein the numerical sequences have a preset length and a preset depth and are arranged in one dimension; The plurality of numerical sequences are used as inputs of a preset hybrid network model to output a virtual 3D speed of the vehicle according to the preset hybrid network model.

3. The method according to claim 1, characterized in that The step of constructing an observation model according to the virtual 3D velocity and adaptively optimizing the random model by evaluating the regression accuracy of the virtual 3D velocity comprises: Determining the observation speed based on the speed and attitude of the inertial navigation system; The observation model is constructed according to the virtual 3D velocity and the observation velocity.

4. The method according to claim 1, characterized in that The observation model includes: in, Characterizes the difference between the observed speed calculated by the inertial navigation system in the vehicle and the virtual 3D speed, is the observed speed calculated by the inertial navigation system, is the virtual 3D speed, is the observation coefficient matrix, X is the state variable vector, which contains all the error states that need to be estimated by the vehicle's integrated navigation system, Characterize the target observation noise covariance matrix under adaptive constraints.

5. The method according to claim 4, characterized in that The step of constructing an observation model according to the virtual 3D velocity and adaptively optimizing the random model by evaluating the regression accuracy of the virtual 3D velocity also includes: Determine the initial observation noise covariance matrix; The observation noise covariance matrix is ​​updated according to the difference between the observation speed and the virtual 3D speed to determine the target observation noise covariance matrix when the updating is completed.

6. The method according to claim 5, characterized in that The determining the positioning information of the vehicle according to the observation model and the random model includes: determining an error state of the inertial navigation system according to the observation model; The error state is added to a current state estimate of the inertial navigation system to update the state of the inertial navigation system and obtain the positioning information of the vehicle.

7. The method according to claim 2, characterized in that The preset length of the numerical sequence is the sampling rate of the inertial measurement unit.

8. A high-precision vehicle positioning device based on virtual 3D velocity adaptive constraints, characterized in that: include: A generating module, used for obtaining measurement data of an inertial measurement unit in the vehicle when a GNSS signal of a navigation system of the vehicle is interrupted, constructing a regression model of a virtual 3D velocity of the vehicle, establishing a mapping relationship between the measurement data and the virtual 3D velocity of the vehicle, and generating a virtual 3D velocity of the vehicle; A construction module, used to construct an observation model according to the virtual 3D velocity, and adaptively optimize the random model by evaluating the regression accuracy of the virtual 3D velocity; A positioning module is used to determine the positioning information of the vehicle according to the observation model and the random model.

9. A computer device, characterized in that: The computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images