Multi-task odometer network-assisted inertial navigation positioning method fusing motion state

Abstract

The embodiment of the invention discloses a multi-task odometer network-assisted inertial navigation positioning method fusing a motion state, and relates to the technical field of navigation positioning, and the method comprises the following steps: obtaining measurement information output by an inertial measurement unit of a carrier; under the condition that GNSS signals are unavailable, measuring information is input into the trained multi-task odometer network, and the estimated forward speed and the motion state label of the carrier are obtained through processing; updating the estimated forward speed according to the estimated forward speed and the type of the motion state label, and constructing a three-dimensional speed vector in combination with non-integrity constraint; and taking the three-dimensional velocity vector as the input of adaptive Kalman filtering, updating the navigation error parameter, and outputting a navigation positioning result. According to the method, the navigation error of the carrier can be adaptively updated according to different motion states, the navigation error is updated after the motion states are fused, error divergence of INS can be greatly inhibited, and a real odometer can be replaced to obtain a more stable and more accurate positioning result.

Description

Technical Field

[0001] This application relates to the technical field of navigation and positioning, and particularly to a multi-task odometer network-assisted inertial navigation positioning method that integrates motion states. Background Art

[0002] The GNSS / INS integrated navigation system is a navigation technology that combines the Global Navigation Satellite System (GNSS) and the Inertial Navigation System (INS). This combination utilizes the advantages of both systems and can provide more accurate and reliable positioning information. Since GNSS signals are vulnerable and easily blocked or reflected by high-rise buildings, resulting in multipath effects, while INS can output all navigation information of the carrier's position, speed, and attitude without being restricted by the environment. Therefore, the GNSS / INS integrated navigation system can maintain high-precision navigation in a short period of time.

[0003] However, if the GNSS signal is blocked, for example, when the vehicle drives into areas such as tunnels or bridge holes, the GNSS / INS integrated navigation system only relies on the INS system for independent calculation. In the initial few seconds, the INS system can output high-precision position information. However, as time goes by, without the input of measurement information, the Kalman Filter (KF) cannot perform measurement updates, resulting in the rapid accumulation of INS calculation errors over time.

[0004] To improve the positioning accuracy and robustness of the GNSS / INS integrated navigation system in complex environments, currently, the Inertial Measurement Unit (IMU) data of the IMU and the navigation state data calculated by the INS collected when the GNSS signal is good can be used to train a Deep Neural Network (DNN). When the GNSS signal is interrupted, the trained network is used to predict the GNSS pseudo-position, etc., to suppress the divergence of INS errors. However, this method still has some defects. For example: 1. During the rapid change of the vehicle's driving speed, the estimation accuracy of the trained neural network will significantly decrease, that is, the trained neural network is difficult to adapt to the dynamic change of the vehicle speed in real time; 2. The accuracy is insufficient in the zero-speed detection problem, that is, based on the estimated speed, the vehicle is detected to be in a stationary state, but in fact, the vehicle has not yet decelerated to zero speed or has just started; 3. The constant noise model constructed based on the estimation error cannot well describe the uncertainty of the measurement noise.

[0005] Therefore, there is currently a lack of a method that can improve the accuracy and stability of the positioning results output by a GNSS / INS integrated navigation system in the event of a GNSS signal interruption. Summary of the Invention

[0006] An embodiment of the present application provides a multi-task odometer network-assisted inertial navigation positioning method that integrates motion states to solve the defects of the above-related technologies. The technical solutions are as follows: In a first aspect, an embodiment of the present application provides a multi-task odometer network-assisted inertial navigation positioning method that integrates motion states, including: Obtain the triaxial gyroscope information and triaxial accelerometer information output by the inertial measurement unit of the vehicle; In the case where the GNSS signal is unavailable, input the triaxial gyroscope information and the triaxial accelerometer information into a trained multi-task odometer network; The multi-task odometer network processes the triaxial gyroscope information and the triaxial accelerometer information to obtain the estimated forward speed and motion state label of the vehicle; Update the estimated forward speed according to the type of the estimated forward speed and the motion state label, and construct a three-dimensional velocity vector in combination with nonholonomic constraints; Use the three-dimensional velocity vector as the input of an adaptive Kalman filter, and update the navigation error parameters of the vehicle through the adaptive Kalman filter; Output a navigation positioning result in combination with the navigation error parameters; Wherein, the multi-task odometer network is trained based on the triaxial gyroscope information, triaxial accelerometer information, and forward speed measurement values of the vehicle when the GNSS signal is available.

[0007] In an optional solution of the first aspect, the process of obtaining the estimated forward speed and motion state label of the vehicle by the multi-task odometer network based on the triaxial gyroscope information and the triaxial accelerometer information includes: The multi-task odometer network extracts features from the triaxial gyroscope information and the triaxial accelerometer information and outputs underlying features; Extract non-linear features based on the underlying features, and extract linear features based on the triaxial gyroscope information and the triaxial accelerometer information; Combine the non-linear features and the linear features to obtain the estimated forward speed of the vehicle; Determine the probability distribution of each motion state type based on the underlying features, and obtain the motion state label of the vehicle according to the motion state type with the largest probability value.

[0008] In an alternative solution of the first aspect, updating the estimated forward speed according to the estimated forward speed and the type of the motion state label, and constructing a three-dimensional velocity vector by combining nonholonomic constraints includes: When the motion state type corresponding to the motion state label is a stationary state and the estimated forward speed is less than the zero-speed judgment threshold, perform zero-speed update on the estimated forward speed to set the value of the estimated forward speed to 0; Construct the three-dimensional velocity vector based on the estimated forward speed after zero-speed update and nonholonomic constraints.

[0009] In an alternative solution of the first aspect, updating the estimated forward speed according to the estimated forward speed and the type of the motion state label, and constructing a three-dimensional velocity vector by combining nonholonomic constraints includes: When the motion state type corresponding to the motion state label is a non-stationary state, perform compensation update on the estimated forward speed through a preset scale factor, and construct a three-dimensional velocity vector by combining the estimated forward speed after compensation update and the nonholonomic constraints.

[0010] In an alternative solution of the first aspect, the navigation error parameters include forward speed measurement noise and nonholonomic constraint measurement noise; Taking the three-dimensional velocity vector as the input of an adaptive Kalman filter and updating the navigation error parameters of the vehicle through the adaptive Kalman filter includes: Taking the three-dimensional velocity vector as the input of an adaptive Kalman filter and performing measurement update through the adaptive Kalman filter; Update the forward speed measurement noise through Sage Husa filtering; Update the nonholonomic constraint measurement noise at the current moment based on the estimated forward speed and the nonholonomic constraint measurement noise at the previous moment.

[0011] In an alternative solution of the first aspect, the steps for constructing the training set for training the multi-task odometer network include: When the GNSS signal is available, obtain the three-axis gyroscope information, three-axis accelerometer information output by the vehicle's inertial measurement unit, and the forward speed measurement value output by the wheel odometer; Determine the motion state type of the vehicle according to the forward speed measurement value at the same moment and the z-axis output in the three-axis gyroscope information, and generate a sample motion state label; Taking the triaxial gyroscope information and triaxial accelerometer information at the same moment as the sample input, and the forward speed measurement value and the sample motion state label as the sample labels, a training set for training the multi-task odometer network is constructed based on the sample inputs and sample labels at multiple same moments.

[0012] In an alternative scheme of the first aspect, the multi-task odometer network is trained based on the constructed training set. The training steps of the multi-task odometer network include: The multi-task odometer network receives the sample inputs in the training set, so that the multi-task odometer network processes the sample inputs in the sample set to obtain corresponding forward speed estimation values and estimated motion state labels respectively. A first loss function is constructed according to the difference between the forward speed estimation value and the forward speed measurement value in the corresponding sample label, and a second loss function is constructed according to the estimated motion state label and the sample motion state label in the corresponding sample label. Based on the first loss function and the second loss function, it is determined whether the multi-task odometer network converges, and the parameters of the multi-task odometer network after convergence are determined to obtain the trained multi-task odometer network.

[0013] In a second aspect, an embodiment of the present application further provides a multi-task odometer network-assisted inertial navigation positioning device integrating motion states, including: A data acquisition module for acquiring triaxial gyroscope information and triaxial accelerometer information output by an inertial measurement unit of a vehicle. A calculation module for inputting the triaxial gyroscope information and the triaxial accelerometer information into the trained multi-task odometer network when the GNSS signal is unavailable. The calculation module is further configured to process the triaxial gyroscope information and the triaxial accelerometer information through the multi-task odometer network to obtain the estimated forward speed and motion state label of the vehicle. The calculation module is further configured to update the estimated forward speed according to the type of the estimated forward speed and the motion state label, and construct a three-dimensional velocity vector in combination with non-holonomic constraints. A navigation module for using the three-dimensional velocity vector as an input of an adaptive Kalman filter to update the navigation error parameters of the vehicle through the adaptive Kalman filter. The navigation module is further configured to output a navigation positioning result in combination with the navigation error parameters. Wherein, the multi-task odometer network is trained based on the triaxial gyroscope information, triaxial accelerometer information and forward speed measurement value of the vehicle when the GNSS signal is available.

[0014] In a third aspect, an embodiment of the present application further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method provided in the first aspect or any implementation manner of the first aspect of the embodiments of the present application is implemented.

[0015] In a fourth aspect, the present application further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method provided in the first aspect or any implementation manner of the first aspect of the embodiments of the present application is implemented.

[0016] The beneficial effects brought by the technical solutions provided in some embodiments of the present application at least include: A multi-task odometer network assisted inertial navigation positioning method integrating motion states provided by an embodiment of the present application estimates the estimated forward speed of a vehicle based on the parameters output by an inertial measurement unit through a multi-task odometer network during a GNSS signal interruption, and simultaneously detects and determines the motion state of the vehicle. The trained multi-task odometer network can greatly reduce the computational complexity and can accurately detect the motion state of the vehicle; In this way, the navigation error of the vehicle can be adaptively updated according to different motion states. Updating the navigation error after integrating the motion states can greatly suppress the error divergence of the INS and can obtain a more robust and accurate positioning result instead of a real odometer. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 is a schematic flowchart of a multi-task odometer network assisted inertial navigation positioning method integrating motion states provided by an embodiment of the present application; Figure 2 is a schematic structural diagram of a multi-task odometer network provided by an embodiment of the present application; Figure 3 is a box plot of speed estimation error statistics provided by an embodiment of the present application; Figure 4 is a graph of the change in northward and eastward position errors of various navigation schemes provided by an embodiment of the present application.

[0019] Figure 5 is a schematic structural diagram of a multi-task odometer network assisted inertial navigation positioning device integrating motion states provided by an embodiment of the present application; Figure 6 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0020] To make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be clearly and completely described below with reference to the accompanying drawings in the present application. Apparently, the described embodiments are some but not all of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0021] The terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned accompanying drawings are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or modules is not limited to the listed steps or modules, but optionally further includes steps or modules not listed, or optionally further includes other steps or modules inherent to these processes, methods, products or devices.

[0022] It should be noted that the terms "first" and "second" involved in the present application are only used to distinguish similar objects and do not represent a specific order for the objects. It can be understood that "first" and "second" can be interchanged in a specific order or sequence when permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order other than those described or illustrated herein.

[0023] The present application will be described in detail below with reference to specific embodiments.

[0024] Next, in combination with Figure 1 , a multi-task odometry network-assisted inertial navigation positioning method integrating motion states provided by an embodiment of the present application will be introduced. For details, please refer to Figure 1 , Figure 1 shows a schematic flow chart of a multi-task odometry network-assisted inertial navigation positioning method integrating motion states provided by an embodiment of the present application. As Figure 1 shown, the method includes the following steps: S101, obtaining triaxial gyroscope information and triaxial accelerometer information output by an inertial measurement unit of a vehicle; S102, when GNSS signals are unavailable, inputting the triaxial gyroscope information and the triaxial accelerometer information into a trained multi-task odometry network; processing, by the multi-task odometry network, based on the triaxial gyroscope information and the triaxial accelerometer information to obtain an estimated forward speed and a motion state label of the vehicle; S103, updating the estimated forward velocity according to the estimated forward velocity and the type of the motion state label, and constructing a three-dimensional velocity vector in combination with a non-holonomic constraint; S104, using the three-dimensional velocity vector as input of an adaptive Kalman filter, and updating the navigation error parameters of the vehicle through the adaptive Kalman filter; S105: Output a navigation positioning result in combination with the navigation error parameter.

[0025] It should be noted that the vehicles in the embodiments of the present application may include various types of vehicles, such as cars, trucks, etc., and the embodiments of the present application are not limited to this.

[0026] Specifically, in S101, the inertial measurement unit of the vehicle includes a three-axis gyroscope and a three-axis accelerometer, and the three-axis gyroscope information output by the three-axis gyroscope and the three-axis accelerometer information output by the three-axis accelerometer can be obtained, wherein the three-axis gyroscope information can be recorded as , the three-axis accelerometer information can be recorded as .

[0027] It is understandable that in related technologies, whether the GNSS signal is available can be determined by conditions such as whether the number of visible GNSS satellites in the vehicle's field of view is greater than the minimum number requirement and the satellite signal strength. This embodiment of the present application does not limit this.

[0028] It is understandable that in the related art, when the GNSS signal is available, the GNSS position information observed by the GNSS can be obtained. , obtain the position information output by the INS system by observing the position information , speed information and posture information , respectively 、 、 and As the input of the adaptive Kalman filter, the navigation error is estimated and the navigation error includes: position error , speed error and attitude error , the bias error of the INS system can also be calculated. The bias error includes: gyro bias error and accelerometer bias error The INS results and IMU measurements are corrected and compensated according to the estimated navigation error and bias error, and the navigation results are output. The corresponding error terms can be compensated to the output parameters of the corresponding components, such as the gyro bias error. The compensation is added to the three-axis gyroscope information output by the three-axis gyroscope, thereby correcting the error of the three-axis gyroscope.

[0029] When the GNSS signal is unavailable, the steps of S102 can be executed.

[0030] Specifically, the structure of the multi-task odometer network is as Figure 2 shown, including an input layer, a shared underlying feature extraction module, a speed estimation module, a motion state detection module, and an output layer. The multi-task odometer network outputs the estimated forward speed and the motion state label through two branches respectively. The calculation processes of the two branches are as follows: The first branch: Obtain the triaxial gyroscope information and triaxial accelerometer information through the input layer. The input layer transfers the triaxial gyroscope information and triaxial accelerometer information to the speed estimation module; the input layer also transfers the triaxial gyroscope information and triaxial accelerometer information to the underlying feature extraction module, and extracts the underlying features through multiple one-dimensional convolutional layers in the underlying feature extraction module (for example, set four one-dimensional convolutional layers), and also transfers the underlying features to the speed estimation module.

[0031] The speed estimation module includes a linear autoregressive module, and two encoding modules and a feed-forward module connected in sequence before and after. Among them, the linear autoregressive module receives the triaxial gyroscope information and triaxial accelerometer information to perform linear feature extraction and obtain linear features. The underlying features are input to the first encoding module of the speed estimation module to initially extract the global temporal dependence information, and then passed through a max pooling layer to halve the sequence length, and input to the second encoding module to extract the deep temporal dependence, and then sent to a feed-forward module including an adaptive max pooling layer and a fully connected layer to complete the non-linear feature extraction and obtain non-linear features. After that, the linear features and non-linear features are added and the estimated forward speed of the vehicle is output through the output layer.

[0032] The second branch: The underlying features are input to the motion state detection module. The motion state detection module includes an encoding module, a long short-term memory layer, and a fully connected layer connected in sequence. The underlying features are processed by the encoding module, the long short-term memory layer, and the fully connected layer respectively, specifically including: Extract the spatio-temporal attention information through the encoding module, and allocate the attention weights to obtain the output of the encoding block. Then, after transferring the output of the encoding block to the long short-term memory layer and then passing through a fully connected layer, a three-dimensional vector is obtained. Using the softmax function to process , the probability distribution of each motion state type can be obtained, and the motion state label of the vehicle can be obtained according to the motion state type with the largest probability value: , where , represents the ordinal number of the element in the three-dimensional vector .

[0033] Exemplarily, three types of motion state can be set, denoted as , the labels for non - stationary states include: when label = 1, it indicates that the vehicle is detected to be in a straight - line state, when label = 2, it indicates that the vehicle is turning, and the label for the stationary state includes: label = 0 indicates that the vehicle is detected to be in a stationary state.

[0034] In some embodiments, according to the different types of motion state labels, S103 specifically includes: In the case where the motion state type corresponding to the motion state label is the stationary state and the estimated forward speed is less than the zero - speed judgment threshold, S1031 is executed, including: S1031, perform zero - speed update on the estimated forward speed to set the value of the estimated forward speed to 0, and construct the three - dimensional velocity vector based on the zero - speed updated estimated forward speed and the non - holonomic constraint.

[0035] Specifically, when it is jointly detected by the speed estimation module and the motion state detection module that the carrier is in a stationary state, that is, the motion state type corresponding to label is the stationary state, and the estimated forward speed output by the speed estimation module is less than the zero - speed judgment threshold at this time, that is at this time, perform zero - velocity update (ZUPT, Zero Velocity Update), let , and construct the three - dimensional velocity vector by combining with the non - holonomic constraint NHC. Among them, represents the zero - speed judgment threshold, which is usually less than or equal to 0.1 m / s. In this application, a more stringent threshold is adopted, let .

[0036] In the case where the motion state type corresponding to the motion state label is a non - stationary state, S1032 is executed, including: Compensate and update the estimated forward speed through a preset scale factor, and construct a three - dimensional velocity vector by combining the compensated and updated estimated forward speed and the non - holonomic constraint.

[0037] Exemplarily, the input of the multi - task odometer network of this application is the gyroscope data and accelerometer data of the inertial measurement unit. The dimension of the input data is 6, and the length of the time series is 50. The output of the multi - task odometer network is the estimated forward speed and the vehicle motion state label label.

[0038] In the case where GNSS signals are unavailable, the trained multi-task odometry network receives the output data of the inertial measurement unit to estimate the forward speed of the vehicle in real time and simultaneously detect the vehicle motion state. When it is determined according to the label that the vehicle is in a straight or turning state, based on a preset scale factor compensate and update the estimated forward speed to obtain the compensated and updated estimated forward speed after compensation update: ; Furthermore, combine with NHC, set the lateral and ground speeds of the vehicle to zero, and construct a three-dimensional velocity vector .

[0039] S104, Use the three-dimensional velocity vector as the input of the adaptive Kalman filter to update the navigation error parameters of the vehicle through the adaptive Kalman filter; In the case where the motion state type corresponding to the motion state label is the stationary state and the estimated forward speed is less than the zero-speed judgment threshold, input the three-dimensional velocity vector obtained after performing zero-speed update in S1031 into the adaptive Kalman filter for filtering update to estimate the navigation error. By fusing the vehicle motion state, MT-ONet can assist INS to achieve more accurate and robust positioning in the case of GNSS interruption.

[0040] In the case where the motion state type corresponding to the motion state label is a non-stationary state, input the three-dimensional velocity vector obtained after performing compensation update in S1032 into the adaptive Kalman filter for measurement update.

[0041] Specifically, use the Sage Husa adaptive filter to adjust the forward speed measurement noise ; Furthermore, update the current nonholonomic constraint measurement noise based on the estimated forward speed and the nonholonomic constraint measurement noise at the previous moment, including: Estimate the NHC measurement noise according to the estimated forward speed output by the speed estimation module . The NHC measurement noise is composed of the lateral measurement noise and the ground measurement noise as follows: ; where diag() represents constructing a diagonal matrix using , the forward speed measurement noise is a scalar, and the NHC measurement noise is a diagonal matrix composed of the sum of the squares of and the sum of the squares of . ​

[0042] The lateral and vertical measurement noises of the NHC measurement noise at the previous moment are known as , and the NHC measurement noise at the subsequent moment is = , is a preset adaptive factor, which is used to control the magnitude of the NHC measurement noise, so as to avoid excessive noise when the vehicle speed is too fast or too small noise when the vehicle speed is close to zero.

[0043] S105. Output a navigation and positioning result by combining the navigation error parameter.

[0044] It can be understood that a typical Kalman filter algorithm includes two parts: state prediction and measurement update. In the measurement update part, by comparing the forward speed measurement noise and non-integrity constraint measurement noise at the previous moment, the weight of the output measurement value of the corresponding component is adjusted. When the measurement noise increases relative to the previous moment, the "trust" in the corresponding measurement information is reduced, that is, the weight is reduced, so as to maintain the robustness of the filtering result.

[0045] The measurement value output by the INS system can be corrected through the navigation error parameter, so that the filtering result can feedback a more accurate positioning result, so that the error divergence of the INS can be greatly suppressed even when there is no GNSS signal for a long time, and a more robust and accurate positioning result can be obtained instead of the real odometer.

[0046] In some embodiments, the steps for constructing a training set for training the multi-task odometer network include: When the GNSS signal is available, obtain the three-axis gyroscope information, three-axis accelerometer information output by the vehicle's inertial measurement unit, and the forward speed measurement value output by the wheeled odometer; Determine the type of the vehicle's motion state according to the forward speed measurement value at the same moment and the z-axis output in the three-axis gyroscope information, and generate a sample motion state label; Use the three-axis gyroscope information and three-axis accelerometer information at the same moment as the sample input, and the forward speed measurement value and sample motion state label as the sample label, and construct a training set for training the multi-task odometer network based on the sample input and sample label at multiple same moments.

[0047] In some embodiments, the multi-task odometer network is trained based on the constructed training set. The training steps of the multi-task odometer network include: Receive the sample input in the training set through the multi-task odometer network, so that the multi-task odometer network processes the sample input in the sample set to obtain the corresponding forward speed estimate value and estimated motion state label respectively; Construct a first loss function based on the difference between the forward speed estimated value and the forward speed measured value in the corresponding sample label, and construct a second loss function based on the estimated motion state label and the sample motion state label in the corresponding sample label; Determine whether the multi-task odometer network converges based on the first loss function and the second loss function, and determine the parameters of the multi-task odometer network after convergence to obtain a trained multi-task odometer network.

[0048] Exemplarily, the weighted sum of the first loss function and the second loss function can be minimized according to the Adam optimization algorithm. If the value of the weighted sum is less than a preset threshold, it is determined that the multi-task odometer network converges, thereby updating the parameters of the multi-task odometer network.

[0049] In some embodiments, as Figure 3 shown, Figure 3 is a box plot of the speed estimation errors of the multi-task odometer network provided in this application and other deep learning networks, including the multi-task odometer network (MT-ONet), LONet, OdoNet, TPA-LSTM, and LSTNet provided in this application.

[0050] It can be seen the box plots of the speed estimation error distributions of all models in three speed intervals. The short horizontal line inside the box represents the median, the top and bottom of the box represent the 75th and 25th percentiles respectively. The upper and lower edges represent the 90th percentile and the 10th percentile respectively. It can be seen that the multi-task odometer network provided in this application has the smallest median error, and the boxes are all smaller, the error distribution is more concentrated, and the performance is more stable.

[0051] In addition, in terms of speed estimation, the multi-task odometer network provided by this application has the highest speed estimation accuracy, with a more concentrated error distribution and more stable performance. In terms of zero-speed detection, the multi-task odometer network provided by this application has a higher precision rate and F1 score, indicating that its comprehensive performance is better. In addition, the multi-task odometer network provided by this application can also detect straight-line and turning, which is not available in other network models. Finally, in terms of the number of parameters and computational complexity, the multi-task odometer network provided by this application has a certain increase in the number of parameters compared with LONet, but it has the lowest computational complexity. Comparing the MT-ONet aided inertial navigation scheme with the traditional NHC aided inertial navigation scheme, the true odometer aided inertial navigation scheme, the OdoNet and LONet aided inertial navigation schemes, the MT-ONet aided inertial navigation SageHusa robust adaptive scheme, and the proposed MT-ONet aided inertial navigation robust adaptive scheme that fuses the motion state, the experimental results show that during the GNSS signal interruption, the MT-ONet aided inertial navigation robust adaptive scheme that fuses the motion state can significantly suppress the error divergence of the INS, is superior to the true odometer scheme, and can replace the true odometer to obtain a more robust and accurate positioning result.

[0052] In some embodiments, as Figure 4 shown, Figure 4 it exemplifies the northward and eastward position error variation diagrams of all navigation schemes. In the case of GNSS signal interruption, the navigation schemes for comparison are: 1. NHC scheme: only relying on NHC to suppress the INS error divergence; 2. ODO scheme: using NHC and the forward speed output by the true odometer combined with ZUPT to suppress the INS error divergence; 3. OdoNet scheme: the same as the ODO scheme, but using the forward speed estimated by OdoNet to replace the output of the true odometer to suppress the INS error; 4. LONet scheme: the same as the ODO scheme, but using the forward speed estimated by LONet to replace the output of the true odometer to suppress the INS error; 5. MT-ONet scheme: using the forward speed estimated by the multi-task odometer network provided by this application to replace the output of the true odometer to suppress the INS error; 6. SageMT-ONet scheme: the same as the MT-ONet scheme, based on the multi-task odometer network of the MT-ONet scheme, combined with SageHusa filtering to adaptively adjust the speed noise. 7. AdaMT-ONet scheme: the same as the MT-ONet scheme, and also using the methods of S101-S105 provided in the above embodiments to adaptively adjust the forward speed noise and the NHC speed noise.

[0053] During the 60s interruption, the AdaMT-ONet and ODO schemes maintained relatively low northward errors within the first 15 seconds. After 15 seconds of the interruption, the error of the AdaMT-ONet scheme began to increase and stabilized at 30 seconds of the interruption. During the last 30 seconds, AdaMT-ONet maintained relatively stable and low northward errors, significantly better than the ODO and NHC schemes, where the NHC scheme had almost the largest error. In terms of eastward error, the errors of all schemes increased within the first 40 seconds and reached the maximum at 40 seconds. After 40 seconds, the errors of all schemes except the ODO scheme began to decrease, and at this time, the eastward errors of the three schemes based on MT-ONet were the smallest. After 50 seconds, the AdaMT-ONet scheme had the smallest error. During the 120s interruption, the northward errors of the AdaMT-ONet and MT-ONet schemes were basically the lowest, better than the ODO scheme. Their eastward errors were very close to the ODO scheme. The SageMT-ONet scheme did not show an advantage, indicating that the Sage-Husa method did not bring a gain. There were three stops during the 180s interruption, lasting 13 seconds, 26 seconds, and 7 seconds respectively. Obviously, during the stops, the position errors of most schemes stopped diverging, except for the NHC scheme, whose eastward error still diverged during the long stop, indicating that its suppression effect on the divergence of inertial navigation errors was limited. In terms of northward error, the AdaMT-ONet and SageMT-ONet schemes always remained the lowest. In terms of eastward error, the AdaMT-ONet, MT-ONet, and ODO schemes were relatively close within the first 120 seconds. After that, the errors of the AdaMT-ONet and MT-ONet schemes were greater than the ODO scheme, and the errors of the AdaMT-ONet and NHC schemes were the lowest during the last 30 seconds. Generally speaking, during the three GNSS interruptions, the AdaMT-ONet scheme we proposed was better than the traditional NHC and ODO schemes, more robust than the OdoNet, LONet, and MT-ONet schemes without using the robust adaptive strategy, and had stronger adaptive ability than the method based on the Sage-Husa filter, and could maintain accuracy over time.

[0054] The following is the device embodiment of the present application, which can be used to execute the method embodiment of the present application. For the details not disclosed in the device embodiment of the present application, please refer to the method embodiment of the present application.

[0055] Next, please refer to Figure 5 , which is a schematic structural diagram of a multi-task odometer network assisted inertial navigation positioning device integrating motion states provided for an exemplary embodiment of the present application. The device can be implemented as all or part of a terminal through software, hardware, or a combination of both, and can also be integrated as an independent module on a server. The device 50 includes a data acquisition module 501, a calculation module 502, and a navigation module 503, where: The data acquisition module 501 is used to acquire the triaxial gyroscope information and triaxial accelerometer information output by the inertial measurement unit of the vehicle; The calculation module 502 is used to input the triaxial gyroscope information and the triaxial accelerometer information into the trained multi-task odometer network when the GNSS signal is unavailable; The calculation module 502 is further used to process the triaxial gyroscope information and the triaxial accelerometer information through the multi-task odometer network to obtain the estimated forward speed and motion state label of the vehicle; The calculation module 502 is further used to update the estimated forward speed according to the type of the estimated forward speed and the motion state label, and construct a three-dimensional velocity vector by combining non-holonomic constraints; The navigation module 503 is used to use the three-dimensional velocity vector as the input of the adaptive Kalman filter, and update the navigation error parameters of the vehicle through the adaptive Kalman filter; The navigation module 503 is further used to output a navigation positioning result in combination with the navigation error parameters; It should be noted that when the device 50 provided in the above embodiment executes a multi-task odometer network-assisted inertial navigation positioning method for fusing motion states, only the above-mentioned division of each functional module is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device provided in the above embodiment and the embodiment of a multi-task odometer network-assisted inertial navigation positioning method for fusing motion states belong to the same concept, and the implementation process thereof is detailed in the method embodiment, which will not be repeated here.

[0056] An embodiment of the present application further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the method in any of the above embodiments are implemented.

[0057] Please refer to Figure 6 , which is a structural block diagram of an electronic device provided by an embodiment of the present application.

[0058] As Figure 6 shown, the electronic device 600 includes a processor 601 and a memory 602.

[0059] In the embodiments of the present application, the processor 601 is the control center of the computer system, which can be the processor of a physical machine or the processor of a virtual machine. The processor 601 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. The processor 601 may be implemented in at least one of the following hardware forms: DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array).

[0060] The processor 601 may also include a main processor and a coprocessor. The main processor is a processor used to process data in the wake state, also known as the CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state.

[0061] The memory 602 may include one or more computer-readable storage media, and the computer-readable storage media may be non-transitory. The memory 602 may further include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash storage devices. In some embodiments of the present application, the non-transitory computer-readable storage media in the memory 602 are used to store at least one instruction, and the at least one instruction is used to be executed by the processor 601 to implement the method in the embodiments of the present application.

[0062] In some embodiments, the electronic device 600 further includes: a peripheral device interface 603 and at least one peripheral device 604. The processor 601, the memory 602, and the peripheral device interface 603 may be connected through a bus or signal lines. Each peripheral device 604 may be connected to the peripheral device interface 603 through a bus, signal lines, or a circuit board. Specifically, the peripheral device 604 includes: a display screen, a camera, and an audio circuit. The peripheral device interface 603 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 601 and the memory 602.

[0063] In some embodiments of the present application, the processor 601, the memory 602, and the peripheral device interface 603 are integrated on the same chip or circuit board; in some other embodiments of the present application, any one or two of the processor 601, the memory 602, and the peripheral device interface 603 may be implemented on a separate chip or circuit board. The embodiments of the present application do not make specific limitations on this.

[0064] The block diagram of the electronic device shown in the embodiments of the present application does not limit the electronic device 600. The electronic device 600 may include more or fewer components than those shown in the figure, combine some components, or adopt different component arrangements.

[0065] The embodiments of the present application also provide a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the method in any of the foregoing embodiments are implemented. Among them, the computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, and magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0066] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solutions, or the part that contributes to the related technologies, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disks, optical disks, etc., and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-task odometry network assisted inertial navigation positioning method integrating motion states, characterized in that, Including: Obtaining triaxial gyroscope information and triaxial accelerometer information output by an inertial measurement unit of a vehicle; When GNSS signals are unavailable, inputting the triaxial gyroscope information and the triaxial accelerometer information into a trained multi-task odometer network; Processing, by the multi-task odometer network, based on the triaxial gyroscope information and the triaxial accelerometer information to obtain an estimated forward speed and a motion state label of the vehicle; Updating the estimated forward speed according to the type of the estimated forward speed and the motion state label, and constructing a three-dimensional velocity vector by combining nonholonomic constraints; Using the three-dimensional velocity vector as an input of an adaptive Kalman filter, and updating navigation error parameters of the vehicle through the adaptive Kalman filter; Outputting a navigation positioning result by combining the navigation error parameters; Wherein, the multi-task odometer network is trained based on triaxial gyroscope information, triaxial accelerometer information, and forward speed measurement values of the vehicle when GNSS signals are available.

2. The multi-task odometer network assisted inertial navigation positioning method integrating motion states according to claim 1, wherein The processing, by the multi-task odometer network, based on the triaxial gyroscope information and the triaxial accelerometer information to obtain the estimated forward speed and the motion state label of the vehicle includes: Performing feature extraction on the triaxial gyroscope information and the triaxial accelerometer information by the multi-task odometer network, and outputting underlying features; Extracting non-linear features based on the underlying features, and extracting linear features based on the triaxial gyroscope information and the triaxial accelerometer information; Combining the non-linear features and the linear features to obtain the estimated forward speed of the vehicle; Determining a probability distribution of each motion state type based on the underlying features, and obtaining the motion state label of the vehicle according to the motion state type with the largest probability value.

3. The multi-task odometer network assisted inertial navigation positioning method integrating motion states according to claim 2, wherein The updating the estimated forward speed according to the type of the estimated forward speed and the motion state label, and constructing a three-dimensional velocity vector by combining nonholonomic constraints includes: When the motion state type corresponding to the motion state label is a stationary state and the estimated forward speed is less than a zero-speed judgment threshold, performing zero-speed update on the estimated forward speed to set the value of the estimated forward speed to 0; Constructing the three-dimensional velocity vector based on the estimated forward speed after zero-speed update and nonholonomic constraints.

4. A multi-task odometry network assisted inertial navigation positioning method integrating motion states according to claim 2, wherein The updating the estimated forward speed according to the type of the estimated forward speed and the motion state label, and constructing a three-dimensional velocity vector by combining nonholonomic constraints includes: When the motion state type corresponding to the motion state label is a non-stationary state, performing compensation update on the estimated forward speed through a preset scaling factor, and constructing a three-dimensional velocity vector by combining the estimated forward speed after compensation update and the nonholonomic constraints.

5. A multi-task odometer network assisted inertial navigation positioning method integrating motion states according to claim 4, characterized in that The navigation error parameters include forward speed measurement noise and nonholonomic constraint measurement noise; The using the three-dimensional velocity vector as an input of an adaptive Kalman filter, and updating navigation error parameters of the vehicle through the adaptive Kalman filter includes: Take the three-dimensional velocity vector as the input of the adaptive Kalman filter and perform measurement update through the adaptive Kalman filter; Obtain the forward velocity measurement noise through Sage Husa filter update; Update the current nonholonomic constraint measurement noise based on the estimated forward velocity and the nonholonomic constraint measurement noise at the previous moment.

6. A multi-task odometer network assisted inertial navigation positioning method integrating motion states according to any one of claims 1-5, characterized in that The construction steps of the training set for training the multi-task odometer network include: When the GNSS signal is available, obtain the three-axis gyroscope information, three-axis accelerometer information output by the vehicle's inertial measurement unit, and the forward velocity measurement value output by the wheel odometer; Determine the type of the vehicle's motion state according to the forward velocity measurement value at the same moment and the z-axis output in the three-axis gyroscope information, and generate a sample motion state label; Use the three-axis gyroscope information and three-axis accelerometer information at the same moment as the sample input, and the forward velocity measurement value and the sample motion state label as the sample label. Based on the sample inputs and sample labels at multiple same moments, construct a training set for training the multi-task odometer network.

7. An integrated motion state multi-task odometer network assisted inertial navigation positioning method according to claim 6, characterized in that, Train the multi-task odometer network based on the constructed training set. The training steps of the multi-task odometer network include: Receive the sample input in the training set through the multi-task odometer network, so that the multi-task odometer network processes the sample input in the sample set to obtain the corresponding estimated forward velocity and estimated motion state label respectively; Construct a first loss function according to the difference between the estimated forward velocity and the forward velocity measurement value in the corresponding sample label, and construct a second loss function according to the difference between the estimated motion state label and the sample motion state label in the corresponding sample label; Determine whether the multi-task odometer network converges based on the first loss function and the second loss function, and determine the parameters of the multi-task odometer network after convergence to obtain a trained multi-task odometer network.

8. A multi-task odometer network assisted inertial navigation positioning device integrating motion states, characterized in that, Include: A data acquisition module for obtaining the three-axis gyroscope information and three-axis accelerometer information output by the vehicle's inertial measurement unit; A calculation module for inputting the three-axis gyroscope information and the three-axis accelerometer information into the trained multi-task odometer network when the GNSS signal is not available; The calculation module is further configured to process the three-axis gyroscope information and the three-axis accelerometer information through the multi-task odometer network to obtain the estimated forward velocity and motion state label of the vehicle; The calculation module is further configured to update the estimated forward velocity according to the type of the estimated forward velocity and the motion state label, and construct a three-dimensional velocity vector in combination with nonholonomic constraints; A navigation module for taking the three-dimensional velocity vector as the input of the adaptive Kalman filter and updating the navigation error parameters of the vehicle through the adaptive Kalman filter; The navigation module is further configured to output a navigation positioning result in combination with the navigation error parameters; Wherein, the multi-task odometer network is trained based on the three-axis gyroscope information, three-axis accelerometer information and forward velocity measurement value of the vehicle when the GNSS signal is available.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A non-transitory computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images