Crowdsourcing data driven model and parameter self-learning navigation method and device

Abstract

The application relates to the technical field of combined navigation, in particular to a navigation method and device driven by crowd-sourced data and self-learned parameters, wherein the method comprises the following steps: acquiring crowd-sourced map data of a target navigation system and extracting at least one characteristic factor; updating the crowd-sourced map data to generate crowd-sourced map label data; constructing a preset variance self-learning model based on the at least one characteristic factor and the crowd-sourced map label data, optimizing position variance information of the target navigation system to obtain model predicted position variance; constructing a scaling factor according to the model predicted position variance, and generating a final positioning result of the target navigation system through the scaled model predicted position variance and a preset semi-compact combined architecture. The application can autonomously learn the position variance information, does not need manual intervention and high-precision reference true value equipment, can continuously optimize the system model through the update of the crowd-sourced data during high-precision navigation and positioning, and continuously improves the generalization ability and intelligent level of the system.

Description

Technical Field

[0001] The present application relates to the field of combined navigation technology, and in particular to a navigation method and device using a crowdsourced data-driven model and parameter self-learning. Background Art

[0002] With the development of artificial intelligence and the Internet of Things (IoT), unmanned systems, such as drones and autonomous vehicles, are becoming increasingly intelligent and networked. Positioning, as the foundation and core of environmental perception and decision-making in unmanned systems, requires critical accuracy and availability.

[0003] Among related technologies, the Global Navigation Satellite System (GNSS), as the most mature positioning technology, has been widely used in the field of vehicle navigation. However, due to the fragility and limitations of satellite radio signals, GNSS faces three major problems: blocking, interference, and deception. In complex scenarios such as urban canyons, tree-lined areas, elevated roads, and tunnels, its navigation and positioning performance rapidly degrades. As a relative positioning method, the Strapdown Inertial Navigation System (SINS) can provide full navigation information including position, velocity, and attitude around the clock and at high frequency. However, due to the errors of inertial measurement unit devices, its dead reckoning error will diverge over time. GNSS and SINS are highly complementary. The GNSS / SINS combined navigation formed by the complementary advantages of the two has the characteristics of high accuracy, strong reliability, wide adaptability, and good dynamics. Therefore, it has become an important technical means for navigation and positioning of unmanned systems.

[0004] However, the GNSS / SINS combined navigation system in related technologies mainly relies on model-driven positioning, which makes it difficult to fully extract common features in large-scale data and build complex models. Most algorithms focus on GNSS gross error detection, and there is insufficient research on the consistency between GNSS position variance and actual error. In addition, algorithm training schemes usually rely on high-precision reference systems to generate labels. As a result, the accuracy of navigation and positioning results is lacking and certain costs are increased, which needs to be solved urgently. Summary of the Invention

[0005] The present application provides a navigation method and device with a crowdsourced data-driven model and parameter self-learning, so as to solve the problems in the related art that the GNSS / SINS combined navigation system mainly relies on model-driven when achieving positioning, making it difficult to fully extract common features in large-scale data and difficult to build complex models. In addition, most algorithms mainly focus on GNSS gross error detection, and lack research on the consistency between GNSS position variance and actual error. In addition, the algorithm training scheme also usually relies on high-precision reference systems to generate labels. As a result, the accuracy of the navigation positioning results is lacking and a certain cost is increased.

[0006] The first aspect of the present application provides a navigation method of crowdsourced data-driven model and parameter self-learning, comprising the following steps: obtaining crowdsourced map data of a target navigation system and extracting at least one characteristic factor of the crowdsourced map data; screening map results that meet a preset reliability based on the at least one characteristic factor, and updating the crowdsourced map data according to the map results to generate crowdsourced map label data that meets a preset accuracy; constructing a preset variance self-learning model based on the at least one characteristic factor and the crowdsourced map label data, and optimizing the position variance information of the target navigation system through the preset variance self-learning model to obtain the model predicted position variance of the target navigation system; constructing a scaling factor based on the model predicted position variance to scale the model predicted position variance using the scaling factor to obtain the scaled model predicted position variance, and generating the final positioning result of the target navigation system through the scaled model predicted position variance and a preset semi-tight combination architecture.

[0007] Optionally, in one embodiment of the present application, extracting at least one characteristic factor of the crowdsourced map data includes: post-processing and solving the crowdsourced map data to obtain processed crowdsourced map data; and extracting the at least one characteristic factor based on the processed crowdsourced map data using a preset correlation coefficient.

[0008] Optionally, in one embodiment of the present application, the screening of map results that meet a preset reliability based on the at least one characteristic factor, and updating the crowdsourced map data according to the map results to generate crowdsourced map label data that meet a preset accuracy include: using simulated positioning errors to determine the accuracy characteristics and observation update interval of the inertial sensor of the target navigation system; screening map results that meet the preset reliability based on the accuracy characteristics, the observation update interval and the at least one characteristic factor; and correcting the divergent error of the map results to generate crowdsourced map label data that meet the preset accuracy.

[0009] Optionally, in one embodiment of the present application, the preset variance self-learning model is constructed based on the at least one feature factor and the crowdsourced map label data, including: expanding the channel of the at least one feature factor, and using temporal convolution to aggregate the feature information of the at least one feature factor to obtain local features with time dependence; learning contextual information with dependencies from the local features, and based on the context information, using an attention mechanism to focus on the feature information to generate position variance information; using the position variance information and the crowdsourced map label data to train and optimize the initial variance self-learning model to obtain the preset variance self-learning model.

[0010] Optionally, in one embodiment of the present application, the final positioning result of the target navigation system is generated by using the scaled model predicted position variance and the preset semi-tight combination architecture, including: updating the data information of the strapdown inertial navigation system according to the map result measurement; fusing the data information and the scaled model predicted position variance through the preset semi-tight combination structure to obtain the final positioning result of the target navigation system.

[0011] The second aspect of the present application provides a navigation device with a crowdsourced data-driven model and parameter self-learning, including: an extraction module for obtaining crowdsourced map data of a target navigation system and extracting at least one characteristic factor of the crowdsourced map data; an update module for screening map results that meet a preset reliability based on the at least one characteristic factor, and updating the crowdsourced map data according to the map results to generate crowdsourced map label data that meets a preset accuracy; a construction module for constructing a preset variance self-learning model based on the at least one characteristic factor and the crowdsourced map label data, and optimizing the position variance information of the target navigation system through the preset variance self-learning model to obtain the model predicted position variance of the target navigation system; a positioning module for constructing a scaling factor according to the model predicted position variance, so as to scale the model predicted position variance using the scaling factor to obtain the scaled model predicted variance, and generating the final positioning result of the target navigation system through the scaled model predicted position variance and a preset semi-tight combination architecture.

[0012] Optionally, in one embodiment of the present application, the extraction module includes: a processing unit for post-processing and solving the crowdsourced map data to obtain processed crowdsourced map data; and an extraction unit for extracting the at least one characteristic factor based on the processed crowdsourced map data using a preset correlation coefficient.

[0013] Optionally, in one embodiment of the present application, the update module includes: a determination unit for determining the accuracy characteristics and observation update interval of the inertial sensor of the target navigation system using simulated positioning errors; a screening unit for screening map results that meet a preset reliability based on the accuracy characteristics, the observation update interval and the at least one characteristic factor; and a correction unit for correcting the divergent error of the map result to generate crowdsourced map label data that meets the preset accuracy.

[0014] Optionally, in one embodiment of the present application, the construction module includes: an expansion unit for expanding the channel of the at least one feature factor, and using temporal convolution to aggregate the feature information of the at least one feature factor to obtain a local feature with time dependence; a learning unit for learning contextual information with dependencies from the local features, and focusing the feature information based on the context information using an attention mechanism to generate position variance information; an optimization unit for training and optimizing an initial variance self-learning model using the position variance information and the crowdsourced map label data to obtain the preset variance self-learning model.

[0015] Optionally, in one embodiment of the present application, the positioning module includes: an updating unit for updating the data information of the strapdown inertial navigation system according to the map result measurement; a fusion unit for fusing the data information and the scaled model predicted position variance through the preset semi-tight combination structure to obtain the final positioning result of the target navigation system.

[0016] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the crowdsourcing data-driven model and parameter self-learning navigation method as described in the above embodiment.

[0017] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, it implements the above crowdsourcing data-driven model and parameter self-learning navigation method.

[0018] The fifth aspect of the present application provides a computer program product, including a computer program, which, when executed, is used to implement the above crowdsourcing data-driven model and parameter self-learning navigation method.

[0019] The embodiment of the present application can generate certain characteristic factors by acquiring crowdsourced map data of the target navigation system, and obtain crowdsourced map label data while updating the data information of the relevant navigation system to train a certain variance self-learning model. Finally, the model prediction variance generated by the variance self-learning model and a certain semi-tight combination architecture are used to fuse the updated data information to generate a high-precision positioning result. Thus, autonomous learning of variance information is achieved without the need for manual intervention and high-precision reference truth equipment. While achieving high-precision navigation and positioning, the system model is continuously optimized through the update of crowdsourced data, continuously improving the generalization ability and intelligence level of the system. Thus, the GNSS / SINS integrated navigation system in the related art mainly relies on model driving when achieving positioning, making it difficult to fully extract common features in large-scale data and difficult to achieve the construction of complex models. In addition, most algorithms mainly focus on GNSS gross error detection, and insufficient research on the consistency between GNSS position variance and actual error is conducted. In addition, the algorithm training scheme also usually relies on high-precision reference system to generate labels. As a result, the accuracy of the navigation and positioning results is lacking and a certain cost is increased.

[0020] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0022] Figure 1 A flowchart of a crowdsourced data-driven model and parameter self-learning navigation method provided according to an embodiment of the present application;

[0023] Figure 2 A flowchart of feature candidate factor screening according to one embodiment of the present application;

[0024] Figure 3 This is a flow chart of the characteristic factor correlation analysis of one embodiment of the present application;

[0025] Figure 4 This is a flowchart of self-generation of tag data according to one embodiment of the present application;

[0026] Figure 5 This is a schematic diagram of the architecture of a variance self-learning model according to one embodiment of the present application;

[0027] Figure 6 This is a flowchart of adaptive navigation based on semi-tight combination according to one embodiment of the present application;

[0028] Figure 7A flowchart of an intelligent navigation method using a crowdsourced data-driven model and parameter self-learning according to an embodiment of the present application;

[0029] Figure 8 A schematic diagram of the structure of a navigation device with a crowdsourced data-driven model and parameter self-learning according to an embodiment of the present application;

[0030] Figure 9 A schematic diagram of the structure of an electronic device provided according to an embodiment of the present application.

[0031] Reference numerals:

[0032] 10-Crowdsourced data driven model and parameter self-learning navigation device: 100-extraction module, 200-update module, 300-construction module and 400-positioning module; 901-memory, 902-processor and 903-communication interface. DETAILED DESCRIPTION

[0033] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0034] The following describes a crowdsourced data driven model and parameter self-learning navigation method and device of an embodiment of the present application with reference to the accompanying drawings. In view of the fact that the GNSS / SINS integrated navigation system in the related art mentioned in the above background technology mainly relies on model driving when achieving positioning, it is difficult to fully extract common features in large-scale data and it is not easy to realize the construction of complex models. In addition, most algorithms mainly focus on GNSS gross error detection, and insufficient research is conducted on the consistency between GNSS position variance and actual error. In addition, the algorithm training scheme also usually relies on high-precision reference system to generate labels. As a result, the accuracy of the navigation positioning results is lacking and the cost is increased. The present application provides a crowdsourced data driven model and parameter self-learning navigation method, in which certain feature factors can be generated by obtaining crowdsourced map data of the target navigation system, and crowdsourced map label data can be obtained while updating the data information of the relevant navigation system to train a certain variance self-learning model. Finally, the model prediction variance generated by the variance self-learning model and a certain semi-tight combination architecture are used to fuse the updated data information to generate a high-precision positioning result. This enables autonomous learning of variance information without the need for human intervention or high-precision reference truth equipment. While achieving high-precision navigation and positioning, the system model is continuously optimized through crowdsourced data updates, continuously improving the system's generalization and intelligence capabilities. This addresses the problem in related technologies where GNSS / SINS integrated navigation systems rely primarily on model-driven methods for positioning, making it difficult to fully extract common features from large-scale data and construct complex models. Furthermore, most algorithms focus primarily on GNSS gross error detection, with insufficient research on the consistency between GNSS position variance and actual error. Furthermore, algorithm training schemes often rely on high-precision reference systems to generate labels, resulting in a lack of accuracy in navigation and positioning results and increased costs.

[0035] Before explaining the crowdsourcing data-driven model and parameter self-learning navigation method in the embodiments of the present application, the global navigation satellite system and the strapdown inertial navigation system involved in the embodiments of the present application are first explained.

[0036] The Global Navigation Satellite System (GNSS), currently the most mature positioning technology, has been widely used in vehicle navigation. However, due to the inherent fragility and limitations of satellite radio signals, GNSS faces three major challenges: blocking, interference, and spoofing. Its navigation and positioning performance rapidly degrades in complex scenarios such as urban canyons, tree-lined areas, elevated highways, and tunnels.

[0037] As a relative positioning method, the Strapdown Inertial Navigation System (SINS) can provide comprehensive navigation information, including position, velocity, and attitude, at high frequency and around the clock. However, due to inertial measurement unit (IMU) device errors, its dead-reckoning errors can diverge over time. GNSS and SINS are highly complementary. Combining the strengths of these two, GNSS / SINS integrated navigation boasts high accuracy, strong reliability, wide adaptability, and excellent dynamics, making it a crucial technology for navigation and positioning of unmanned systems.

[0038] As a core element in the process of heterogeneous information fusion, the stochastic model determines the weight of each type of information and significantly affects the accuracy of navigation and positioning results. However, due to factors such as nonlinearity, non-Gaussianity, unmodeled errors, and observational gross errors, complex scenarios can lead to a mismatch between the actual errors and variances of observation information and state information, a phenomenon known as variance distortion. Variance distortion can lead to an unreasonable weighting of current and historical observation information, and cause a high number of true rejection and false acceptance errors in quality control. It also leads to spurious covariances in positioning results, ultimately affecting the accuracy of the navigation and positioning system. Stochastic modeling for SINS is mature. In the GNSS field, a model-driven approach is used to analyze and detect gross errors in observation data through empirical thresholds and adaptive filtering algorithms. However, this leads to incomplete analysis and inability to achieve optimal GNSS / SINS fusion.

[0039] Therefore, the embodiment of the present application proposes a navigation method of crowdsourcing data-driven model and parameter self-learning.

[0040] Specifically, Figure 1 A flowchart of a crowdsourced data-driven model and parameter self-learning navigation method provided in an embodiment of the present application.

[0041] like Figure 1 As shown, the crowdsourced data driven model and parameter self-learning navigation method includes the following steps:

[0042] In step S101 , crowdsourced map data of a target navigation system is acquired, and at least one characteristic factor of the crowdsourced map data is extracted.

[0043] It is understood that crowdsourced map data here refers to large-scale map data collected through crowdsourcing, such as GNSS, SINS, odometer, and CAN bus data. Characteristic factors here refer to characteristic factors related to the observation environment and solution quality of crowdsourced map data, such as signal-to-noise ratio and the number of fixed satellites.

[0044] It should be noted that the target navigation system here refers to an intelligent navigation system that uses crowdsourced map data to drive the model and whose parameters can be self-learned. In the embodiment of this application, it is mainly but not limited to a GNSS navigation system; of course, in actual applications, it can also be adjusted to other navigation systems according to actual conditions. This is only an illustrative explanation without specific limitations.

[0045] In some embodiments, due to the diversity of error sources, a single feature cannot fully and accurately characterize the true error of the target navigation system in actual dynamic scenarios. Currently, there are many parameter indicators that can indirectly reflect the accuracy of the positioning results of the target navigation system, but not all parameter indicators show a clear correlation with the positioning results. Therefore, the present application can obtain crowdsourced map data of the target navigation system and extract at least one characteristic factor of the crowdsourced map data to determine whether some new map data can be updated to the target navigation system.

[0046] For example, the embodiment of the present application can collect GNSS and SINS raw observation data through a crowdsourcing client, including but not limited to pseudorange, phase, Doppler, acceleration and angular velocity data. Furthermore, the embodiment of the present application can also upload these crowdsourced map data to a cloud server through the HTTP / HTTPS protocol to perform more refined data processing and extract characteristic factors.

[0047] Optionally, in one embodiment of the present application, extracting at least one characteristic factor of crowdsourced map data includes: post-processing and solving the crowdsourced map data to obtain processed crowdsourced map data; and extracting at least one characteristic factor based on the processed crowdsourced map data using a preset correlation coefficient.

[0048] In some embodiments, when extracting at least one characteristic factor from crowdsourced map data, to improve the accuracy of the extraction results, embodiments of the present application may first perform certain post-processing on the crowdsourced map data. Post-processing, as used herein, refers to further processing, analysis, and transformation of the raw observation data using a series of mathematical methods, algorithms, or specific processing flows after collection to obtain results that better meet practical needs. Because raw observation data is often obtained directly from observation equipment, it may contain noise, errors, and the like, or its format and content may not meet data usage requirements.

[0049] After obtaining the processed crowdsourced map data, the embodiment of the present application can extract at least one characteristic factor using a preset correlation coefficient. The preset correlation coefficient here can be understood as a coefficient that can measure (calculate) the correlation between the characteristic factor and the observation environment and solution quality, for example, the Spearman's correlation coefficient. The Spearman correlation coefficient, also known as the Spearman rank correlation coefficient, is a non-parametric statistic that can be used to measure the strength and direction of the monotonic relationship between two variables.

[0050] For example, Figure 2 This is a flowchart of feature candidate factor screening according to one embodiment of the present application. Figure 2 As shown in the figure, after uploading raw GNSS and SINS observation data to the cloud server, the cloud server can perform GNSS post-processing. For example, the cloud server can, but is not limited to, model GNSS observations using a dynamic RTK positioning model, select the satellite with the highest elevation angle from different satellite navigation systems as the reference, exclude satellite systems with fewer than four visible satellites, and match common-view satellites between the base station and the rover using satellite PRN numbers.

[0051] Furthermore, the cloud can also extract features related to the observation environment through Kalman filtering: parameters such as signal-to-noise ratio (C / N0) and multipath effect (MP), as well as features related to solution quality: position precision dilution (PDOP), position standard deviation (STD), ambiguity fixation ratio value, fixed number of satellites (NFSAT) and other parameters.

[0052] For example, the calculation formula of the signal-to-noise ratio can be, but is not limited to, expressed as follows:

[0053] C / N0=10log(C / kTB),

[0054] Where C is the signal strength, N0 is the received signal carrier noise intensity, k is the Boltzmann constant, T is the temperature, and B is the equivalent noise bandwidth.

[0055] Multipath refers to the fact that signals from a satellite to a receiver travel multiple propagation paths, causing interference at the receiver and ultimately affecting GNSS positioning accuracy. In this embodiment, the multipath error of three frequency points can be extracted using a geometry-free and ionosphere-free combination. The formula can be, but is not limited to, the following:

[0056]

[0057] Among them, MP i is the geometry-free and ionosphere-free observation value at the i-th frequency point, P i , Li are the pseudo-range and phase observation value on the corresponding frequency point respectively, and a represents the correlation coefficient between the two values β represents In the embodiments of the present application, the multipath value of each frequency point can be calculated by the moving average method, i.e. subtracting the multipath average value in the moving window from the instantaneous value at the current epoch.

[0058] The remaining characteristic information such as PDOP, position standard deviation (STD), ambiguity fixing Ratio value, number of fixed satellites (NFAST) and the like can be obtained based on the relevant mainstream algorithm.

[0059] Next, the embodiments of the present application can, but are not limited to, use the Spearman coefficient to extract the correlation between each characteristic factor and the GNSS position error. Figure 3 The flowchart of the characteristic factor correlation analysis of an embodiment of the present application is shown in Figure 3 The expression of the Spearman coefficient ρ can be as follows:

[0060]

[0061] Wherein, F and L represent the characteristic factor and the GNSS position error respectively. F and L are sorted, and the sorted parameter queue is denoted as R(F) and R(L), which are called the rank of F and L. cov(R(F), R(L)) is the covariance of the rank, σ R(F) and σ R(L) The value range of the correlation coefficient ρ in the embodiments of the present application can be set as (-1, 1), and when ρ takes a negative value, it indicates a negative correlation.

[0062] Since the characteristic factors such as PDOP, STD, Ratio, NFAST have a single numerical value at each epoch, and C / N0 and MP have multiple numerical values at each epoch, for example, the C / N0 of each received satellite signal and the MP of each received frequency point can be calculated at each epoch. In order to facilitate the correlation analysis of C / N0, MP and GNSS position error, the characteristic values of C / N0 and MP can be extracted by constructing a membership matrix to uniquely represent the C / N0 and MP information of the current epoch. Taking the characteristic value extraction of C / N0 as an example, the membership matrix M t can be expressed as follows:

[0063]

[0064] Wherein, is the membership function. The membership function of the signal-to-noise ratio can be calculated by the following formula:

[0065]

[0066] in, represents the signal-to-noise ratio of satellite j at time t. is the maximum absolute difference between the signal-to-noise ratios of all satellites at time t. The closer the signal-to-noise ratios of two satellites are, the larger the corresponding membership function will be, and the higher the weight will be in the fusion process.

[0067] Then, the embodiment of the present application can calculate the eigenvector V corresponding to the maximum eigenvalue of the membership matrix by the power iteration method. n ) T The contribution weight of each satellite's signal-to-noise ratio to the current epoch's signal-to-noise ratio eigenvalue can be calculated from its corresponding eigenvalue, and the formula can be expressed as follows:

[0068]

[0069] Finally, the signal-to-noise ratio eigenvalue λ of the current epoch is calculated based on the above formula t For correlation analysis, the formula can be expressed as follows:

[0070]

[0071] It should be noted that the MP eigenvalue of the current epoch can also be obtained by constructing the membership matrix as described above, and the calculation method of other characteristic factors can also be set or adjusted according to actual conditions. This is only an example and is not a specific limitation.

[0072] Step S102: Screening map results that meet a preset reliability based on at least one characteristic factor, and updating crowdsourced map data according to the map results to generate crowdsourced map label data that meets a preset accuracy.

[0073] It is understood that the preset reliability here can be understood as a preset reliability limit. Only when certain data in the crowdsourced map data meets this reliability limit can it be further utilized, such as updating the crowdsourced map data. In the embodiments of this application, the relevant data that meets this reliability can be referred to as, but is not limited to, map results. The preset accuracy here also refers to a preset accuracy limit. Only when certain data in the crowdsourced map data meets a certain reliability limit and this accuracy limit can valid crowdsourced map label data be generated.

[0074] In other embodiments, considering that the crowdsourcing collection method has the characteristics of device isomorphism and data scale, after extracting at least one characteristic factor of the map data, the present application can also filter out map results that meet a certain reliability based on at least one characteristic factor to update the crowdsourcing map data. By mining device characteristics and combining processing strategies, crowdsourcing map label data that meets a certain accuracy can be self-generated.

[0075] For example, the present application can screen out reliable GNSS position data based on at least one obtained characteristic factor to perform GNSS / SINS measurement updates, and use FBC smoothing technology to achieve autonomous generation of high-precision crowdsourced map label data.

[0076] This process is further explained below.

[0077] Optionally, in one embodiment of the present application, map results that meet a preset reliability are screened based on at least one characteristic factor, and crowdsourced map data is updated according to the map results to generate crowdsourced map label data that meet a preset accuracy, including: using simulated positioning errors to determine the accuracy characteristics and observation update interval of the inertial sensor of the target navigation system; screening map results that meet the preset reliability based on the accuracy characteristics, observation update interval and at least one characteristic factor; and correcting the divergent errors of the map results to generate crowdsourced map label data that meet the preset accuracy.

[0078] In actual implementation, when selecting map results that meet a certain reliability level, this application can first use simulated positioning errors to determine the observation update interval of the target navigation system's inertial sensor observation data. Then, map results that meet a certain reliability level are selected based on accuracy characteristics, observation update interval, and at least one characteristic factor. Finally, the relevant errors of the map results are corrected to obtain crowdsourced map label data that meets a certain accuracy level.

[0079] For example, this application can analyze the accuracy characteristics of the inertial sensor of the GNSS navigation system through the cloud to determine the observation update interval. Then, based on the obtained accuracy characteristics, observation update interval, and at least one characteristic factor, reliable GNSS position data is selected for GNSS / SINS measurement updates, and FBC smoothing technology is used to achieve autonomous generation of high-precision crowdsourced map label data. The specific process can be expressed as follows:

[0080] Since inertial sensors have the ability to maintain high-precision posture during short-term recursion, and this is especially true for inertial sensors with higher precision levels, the embodiments of the present application can sparsely select GNSS observation information from only some epochs for measurement updates during GNSS / SINS combined solution. Of course, these epochs should have high precision and high reliability, so that the inertial navigation divergence error can be accurately corrected.

[0081] Figure 4 This is a flowchart of the self-generation of label data in one embodiment of the present application. Figure 4As shown, for large-scale homogeneous data collected through crowdsourcing, the present embodiment can analyze the common characteristics of inertial sensors. By simulating the positioning error of integrated navigation under different observation update intervals, the largest observation update interval Δt can be selected as much as possible while ensuring consistent position accuracy level for label data generation.

[0082] The GNSS result quality Q obtained by cloud processing is graded based on the observation environment-related parameter factors and solution quality-related parameter factors extracted in other embodiments. The observation environment-related parameter factors include, but are not limited to, the number of common view satellites (NCSAT) and PDOP; the solution quality-related parameter factors include, but are not limited to, the ambiguity fixation state (IAS), the number of fixed satellites (NFSAT), the ambiguity fixation ratio, the result standard deviation (STD), and the result forward and backward error (SEP).

[0083] For example, the number of common-view satellites indicates the number of satellites that can be monitored jointly by the base station and the rover in the current epoch; the ambiguity fixation determines whether the result of the current epoch is a fixed solution or a floating-point solution; the forward and backward mutual difference of the result indicates the second norm of the difference between the forward filtering result and the backward filtering result, which can be expressed as follows:

[0084]

[0085] Among them, ΔX, ΔY, and ΔZ are the differences between the forward and backward results in the X, Y, and Z directions. The remaining parameter factors can be obtained based on relevant mainstream algorithms.

[0086] Then, in the embodiment of the present application, thresholds can be set for the above parameter factors to achieve grading of GNSS result quality Q, for example:

[0087] Q = 1 represents a high-precision, reliable fixed solution. It should simultaneously meet the following conditions: NCSAT is greater than 15 satellites; IAS is an ambiguity-fixed solution; NFSAT is greater than 7 satellites; and the ambiguity-fixing ratio is greater than 5.

[0088] Q = 2 represents a fixed solution with average accuracy and average reliability. It should meet the following conditions: NCSAT is greater than 10 satellites; IAS is an ambiguity fixed solution; NFSAT is greater than 4 satellites; the ambiguity fixation ratio is greater than 3; and SEP is less than 0.5m.

[0089] Q=3 represents a floating-point solution with general precision. It should simultaneously meet the following conditions: NCSAT greater than 10 satellites; PDOP less than 2.5; STD less than 2 meters; SEP less than 1 meter.

[0090] Q = 4 represents a solution with low accuracy and reliability. Results that do not meet the above conditions are classified here.

[0091] Forward filtering updates are performed based on the observation update interval and the GNSS result quality Q. All points that satisfy Q = 1 are used for measurement updates. When the measurement update interval between the previous and next epochs is greater than Δt, points with Q = 2, Q = 3, and Q = 4 are used to fill in the gaps. During the filling process, the currently selected point can be used as the starting moment, and Δt can be taken as the new time period to ensure that the interval between points is always less than Δt. For individual time periods, if the interval within the time period is large and the accuracy of all points within the time period is poor, the observation update variance value of the selected midpoint is appropriately increased to avoid erroneous observation updates from causing serious deviations in the inertial navigation recursive value.

[0092] Based on forward filtering, FBC smoothing technology is used to suppress the divergence of inertial guidance errors by utilizing the kinematic constraints of the state between epochs, and to generate high-precision label data based on past, current, and future measurements. The calculation formula of FBC can be expressed as follows:

[0093]

[0094] in, and P k / N Represent the smoothed state vector and state error covariance matrix respectively; and P k Represents the state vector and state error covariance matrix after filtering update; A k represents the smoothing gain; Φ k represents the state transition matrix; It is the forecast status The error covariance matrix of , the table below N represents the total number of epochs. Through the above formula, the forward filtering result and the reverse filtering result can be smoothed to obtain the forward and reverse smoothed state vectors X Fk and X Bk , and the corresponding state error covariance matrix is ​​P Fk and P Bk .

[0095] Finally, the weighted fusion of the forward and directional results is achieved based on the variance information to obtain the optimal solution result X Ck and its covariance information P Ck As label data, the formula can be expressed as follows:

[0096]

[0097] Step S103: constructing a preset variance self-learning model based on at least one characteristic factor and crowdsourced map label data, and optimizing the position variance information of the target navigation system through the preset variance self-learning model to obtain the model-predicted position variance of the target navigation system.

[0098] It can be understood that the preset variance self-learning model can be understood as a pre-constructed model capable of continuously realizing variance information self-learning. The variance information can be understood as position variance information of the navigation system.

[0099] As a possible implementation manner, the application can construct a certain variance self-learning model to realize autonomous learning of the target navigation system variance information, without manual intervention and high-precision reference true value equipment. In the implementation of high-precision navigation and positioning, the model is continuously optimized through the update of crowd-sourced data, the generalization ability and intelligent level of the target navigation system are continuously improved, and the position variance information of the target navigation system is further optimized.

[0100] For example, the application can but not limited to use the generated at least one feature factor and crowd-sourced map label data, combine the sensitivity of CNN (convolutional neural network) to local features, the memory of LSTM (long short-term memory network) to time series, and the focusing of attention mechanism to feature information to construct a certain variance self-learning model to realize self-learning of variance information. The feature factor and the crowd-sourced map label data are mainly used to train the variance self-learning model.

[0101] Then, the application embodiment can use the constructed variance self-learning model to generate the model predicted position variance of the target navigation system. Further, with the continuous optimization of the variance self-learning model itself, the generated position variance information of the target navigation system, i.e., the model predicted position variance, will also be continuously optimized.

[0102] Next, the construction process of the variance self-learning model in the application embodiment is further explained.

[0103] Optionally, in an embodiment of the application, based on the at least one feature factor and the crowd-sourced map label data, the preset variance self-learning model is constructed, and the method further includes: expanding the channel of the at least one feature factor, and using time sequence convolution to aggregate feature information of the at least one feature factor to obtain local features with time dependence; learning context information with dependent relationship from the local features to generate position variance information by focusing feature information using attention mechanism based on the context information; and training and optimizing the initial variance self-learning model using the position variance information and the crowd-sourced map label data to obtain the preset variance self-learning model.

[0104] Based on the related description of other embodiments, it can be understood that the application can but not limited to use the generated at least one feature factor and crowd-sourced map label data, combine the sensitivity of CNN (convolutional neural network) to local features, the memory of LSTM (long short-term memory network) to time series, and the focusing of attention mechanism to feature information to construct a certain variance self-learning model to realize self-learning of variance information. Figure 5This is a schematic diagram of the architecture of the variance self-learning model of an embodiment of the present application. Figure 5 As shown, the specific construction process can be, but is not limited to, expressed as follows:

[0105] To fit the mapping relationship between the characteristic factors and the GNSS position variance, we first construct a CNN-LSTM network model based on the attention mechanism. The parameter configuration of each module can be expressed as follows:

[0106] One-dimensional convolution module (1D-CNN): This module contains two convolution operations, each of which uses ReL u Function is used as activation function. The input feature data structure is L T ×L F , where L F is the number of GNSS feature factors, LT is the length of the time series used. Convolution is modeled as a sliding window process, and its height is the sliding window length L w , the height of the convolution kernel is the same as the number of feature factors, that is, L W ×L F During feature extraction, the convolution kernel slides in the time dimension to achieve temporal processing.

[0107] Bidirectional LSTM module (BiLSTM): This module builds an aligned two-layer model based on LSTM. One layer propagates from front to back, and the other propagates from back to front. Through forward and backward bidirectional reasoning of features, it learns contextual information with dependencies from local features. Its internal update method can be expressed as follows:

[0108]

[0109] Among them, the input gate uses the Sigmoid activation function σ(); W i and W c are the input gate and memory gate weight matrices, b i and b c are the biases for the input gate and memory gate respectively. t-1 The hidden state of the previous moment can be obtained by the memory state C of the previous moment t-1 and output status o t-1 Get h t-1 =o t-1 *tanh(C t-1 ).

[0110]

[0111] Among them, the forget gate uses the Sigmoid activation function; W f is the forget gate weight matrix, b fBias for the forget gate. Memory cell state C t Update based on the output of the forget gate.

[0112]

[0113] Where the output gate uses a sigmoid activation function and combines the forward propagation hidden state h t-1 and the backward propagation hidden state h' t-1 to generate a prediction output; W o is the output gate weight matrix, b o is the output gate bias, and the hidden state is updated based on this.

[0114] Further, the attention mechanism (Attention) in the embodiments of the present application mainly obtains attention weights through autonomous prompts and non-autonomous prompts generated by the data itself, and generates an attention score function using the BiLSTM output vector q and the corresponding key k i . Wherein the attention weights are mainly generated by the softmax function, but not limited to. For the i-th input vector, its attention weight can be represented by the following formula:

[0115]

[0116] Where a is the attention weight, s(q, k i ) is the attention score function. It should be noted that there are multiple models of the score function, and an additive model is selected as an example for illustrative purposes, but is not a specific limitation:

[0117] s(q, k i ) = v T tanh(Wq + Uk i ),

[0118] Where W, U, and v are learnable parameters that can be optimized and adjusted through model training. Finally, the information selection mechanism is used to integrate all features, which can be represented as follows:

[0119]

[0120] The embodiments of the present application can train the variance self-learning network model based on the generated feature factors and the crowd-sourced map label data, and can continuously update the data set through crowd-sourcing for incremental learning of the model, continuously optimize and adjust the parameters of the variance self-learning model, improve its generalization ability, and finally realize self-learning of the GNSS position variance without human intervention, thereby greatly improving the intelligent level of the present application.

[0121] Step S104, constructing a scaling factor based on the model predicted position variance, and using the scaling factor to scale the model predicted position variance to obtain the scaled model predicted variance, and generating the final positioning result of the target navigation system through the scaled model predicted position variance and the preset semi-tight combination architecture.

[0122] It can be understood that the preset semi-tight combination architecture here can be understood as a pre-set intermediate architecture state between the loosely coupled architecture and the tightly coupled architecture. It can combine the flexibility and maintainability of the loosely coupled architecture with the high efficiency and real-time processing capabilities of the tightly coupled architecture. By optimizing the dependencies and communication methods between components, the "semi-tight combination" architecture can improve the scalability and maintainability of the system while maintaining the overall performance of the system. For example, in a satellite navigation and inertial navigation combination system, the semi-tight combination architecture can give full play to the high-precision positioning information of satellite navigation and the continuous and autonomous navigation capabilities of inertial navigation, and is suitable for scenarios such as vehicle navigation and aerospace that have high requirements for navigation accuracy and reliability.

[0123] During the actual implementation process, it is considered that although the variance self-learning model can learn the position variance information, it is difficult to regress the non-diagonal elements of the covariance matrix, and the correlation between the parameters is lost. Therefore, after obtaining the model-predicted position variance of the target navigation system through the constructed variance self-learning model, the present application can also construct a scaling factor based on the model-predicted variance to scale the model-predicted variance using the scaling factor, thereby achieving the correlation between the parameters by adjusting the non-diagonal elements through the scaling factor. Ultimately, the data information of the target navigation system and the data information of other auxiliary navigation systems after adaptive fusion optimization can be achieved through the scaled model prediction variance and a certain semi-tight combination architecture, and finally the final positioning result of the target navigation system is generated.

[0124] Specifically, the embodiment of the present application can use the prediction results of the variance self-learning model, that is, the model-predicted position variance, to replace the original GNSS position variance information, and construct a scaling factor to adjust the non-diagonal elements to maintain the correlation between parameters. Then, based on the semi-tight combination architecture, the fusion optimization is performed, that is, the updated GNSS and SINS information is measured to obtain a high-precision positioning result. Figure 6 This is a flowchart of an adaptive navigation based on semi-tight combination according to an embodiment of the present application, as shown in FIG. Figure 6 As shown, the specific process can be expressed as follows:

[0125] The cloud sends the incrementally updated variance self-learning model to the client of the target navigation system. The client extracts characteristic factors in real time and uses the variance self-learning model to predict the GNSS position variance.

[0126] Among them, the GNSS position variance predicted by the variance self-learning model only includes the main diagonal information. At this time, the scaling factor λ can be extracted according to the off-diagonal elements in the original GNSS position variance = (λ xx ,λ yy ,λ zz ) T , based on the position variance predicted by the model, the corresponding off-diagonal elements are restored to achieve the transfer of parameter correlation. The formula can be expressed as follows:

[0127]

[0128] in, are the off-diagonal elements of the variance matrix predicted by the network model, are the non-diagonal elements of the original GNSS variance matrix, i, j∈(x, y, z).

[0129] The scaling factor λ can be extracted by the following formula:

[0130]

[0131] After obtaining the scaling factor, embodiments of the present application can use the scaling factor to scale the model prediction variance, resulting in a scaled model prediction variance, thereby maintaining the correlation between parameters. Furthermore, based on the scaled model prediction variance and a certain semi-tight combination architecture, the updated position variance information of the target navigation system and the updated position variance information of other auxiliary navigation systems are integrated to obtain a high-precision positioning result.

[0132] Next, the process of how to fuse the updated position variance information of the target navigation system and the updated position variance information of other auxiliary navigation systems based on the scaled model prediction variance and a certain semi-tight combination architecture in the embodiment of the present application is further explained.

[0133] Optionally, in one embodiment of the present application, the final positioning result of the target navigation system is generated by using the scaled model predicted position variance and a preset semi-tight combination architecture, including: updating the data information of the strapdown inertial navigation system according to the map result measurement; fusing the data information and the scaled model predicted position variance through a preset semi-tight combination structure to obtain the final positioning result of the target navigation system.

[0134] Based on the relevant descriptions of other embodiments, it can be understood that when the embodiment of the present application generates crowdsourced map label data, it uses at least one characteristic factor to screen out map results that meet a certain reliability, and then updates the crowdsourced map data.

[0135] As a possible implementation method, when the crowdsourced map data of the target navigation system is updated, the embodiment of the present application can also measure and update the SINS data information. Then, a certain semi-tight combination architecture is used to fuse the measured and updated SINS data information with the scaled model prediction variance to obtain the final high-precision positioning result.

[0136] Specifically, Figure 6 As shown, the embodiment of the present application can use the scaled model prediction variance matrix to replace the original GNSS position variance, and use the semi-tight combination architecture to fuse the updated GNSS information and SINS information. Among them, the semi-tight combination navigation parameters include but are not limited to position r e , speed v e , attitude φ, accelerometer bias b a and gyro bias b g , where e represents the ECEF coordinate system. The error differential equation of the navigation state can be expressed as follows:

[0137]

[0138] Among them, δg e is the gravity error vector, is the Earth's rotational angular velocity, ξ r ,ξ v and ξ φ are the random walk noise of position, velocity and attitude in ECEF system, b represents the IMU coordinate system, fe is the output of accelerometer in ECEF system, Represents the rotation matrix from the IMU coordinate system to the ECEF coordinate system.

[0139] Based on the position information obtained by GNSS and SINS respectively, and using the placement relationship between the two, the following position and velocity observation equations can be constructed for observation update:

[0140]

[0141] in, l b is the position vector of GNSS in the IMU coordinate system, and All of them can be calculated through the mechanical arrangement at the current moment.

[0142] In addition, the semi-tight combination architecture in the embodiment of the present application can also use the short-term dead reckoning results of SINS to assist in quality control and ambiguity resolution in GNSS filtering, and achieve high-precision real-time navigation positioning through the two-way transmission and sharing of information between the two systems.

[0143] After completing the above work, the target navigation system can realize autonomous learning of variance information without human intervention and high-precision reference true value equipment. Through the update and iteration of crowdsourcing data, the generalization ability of the system can be continuously improved, and intelligent navigation and positioning with variance self-learning can be realized.

[0144] The following is a specific example to illustrate the present application in detail.

[0145] Figure 7 This is a flowchart of an intelligent navigation method based on a crowdsourced data driven model and parameter self-learning according to an embodiment of the present application. Figure 7 As shown:

[0146] (1) Collect large-scale GNSS, SINS and other data through crowdsourcing and upload them to the cloud;

[0147] (2) Perform GNSS post-processing in the cloud and use the Spearman correlation coefficient to extract characteristic factors related to the observation environment and solution quality;

[0148] (3) Reliable GNSS results are screened based on characteristic factors to update GNSS / SINS measurements, and high-precision label data is generated using FBC smoothing technology;

[0149] (4) Combining the sensitivity of CNN to local features, the memory of LSTM to time series, and the focus of attention mechanism on feature information, a deep learning model is constructed to achieve self-learning of variance information;

[0150] (5) The client replaces the original GNSS variance with the model prediction result and adjusts the off-diagonal elements based on the scaling factor to maintain their correlation. Finally, a semi-tight combination architecture is used to fuse the optimized GNSS and SINS to obtain a high-precision positioning result.

[0151] By updating the above crowdsourcing data, the system model is continuously optimized, the generalization ability and intelligence level of the system are continuously improved, and intelligent navigation and positioning with variance self-learning is realized.

[0152] According to the crowdsourced data-driven model and parameter self-learning navigation method proposed in the embodiment of the present application, a certain characteristic factor can be generated by obtaining crowdsourced map data of the target navigation system, and crowdsourced map label data is obtained while updating the data information of the relevant navigation system to train a certain variance self-learning model. Finally, the model prediction variance generated by the variance self-learning model and a certain semi-tight combination architecture are integrated with the updated data information to generate a high-precision positioning result. Thus, autonomous learning of variance information is achieved without manual intervention and high-precision reference truth equipment. While achieving high-precision navigation and positioning, the system model is continuously optimized through the update of crowdsourced data, continuously improving the generalization ability and intelligence level of the system. Thus, the GNSS / SINS integrated navigation system in the related art mainly relies on model driving when achieving positioning, making it difficult to fully extract common features in large-scale data and difficult to build complex models. In addition, most algorithms mainly focus on GNSS gross error detection, and lack research on the consistency between GNSS position variance and actual error. In addition, the algorithm training scheme also usually relies on high-precision reference system to generate labels, thus lacking the accuracy of navigation and positioning results and increasing certain costs.

[0153] Next, a navigation device with a crowdsourcing data driven model and parameter self-learning proposed in accordance with an embodiment of the present application will be described with reference to the accompanying drawings.

[0154] Figure 8 It is a structural diagram of a crowdsourced data driven model and parameter self-learning navigation device according to an embodiment of the present application.

[0155] like Figure 8 As shown, the crowdsourced data driven model and parameter self-learning navigation device 10 includes: an extraction module 100, an update module 200, a construction module 300 and a positioning module 400.

[0156] The extraction module 100 is configured to obtain crowdsourced map data of a target navigation system and extract at least one characteristic factor of the crowdsourced map data;

[0157] An updating module 200 is configured to screen map results that meet a preset reliability based on at least one characteristic factor, and update the crowdsourced map data according to the map results to generate crowdsourced map label data that meets a preset accuracy;

[0158] A construction module 300 is configured to construct a preset variance self-learning model based on at least one characteristic factor and crowdsourced map label data, and optimize the position variance information of the target navigation system using the preset variance self-learning model to obtain a model-predicted position variance of the target navigation system;

[0159] The positioning module 400 is used to construct a scaling factor based on the model predicted position variance, so as to scale the model predicted position variance using the scaling factor to obtain the scaled model predicted position variance, and generate the final positioning result of the target navigation system through the scaled model predicted position variance and a preset semi-tight combination architecture.

[0160] Optionally, in one embodiment of the present application, the extraction module 100 includes: a processing unit and an extraction unit.

[0161] The processing unit is used for post-processing and solving the crowdsourced map data to obtain processed crowdsourced map data;

[0162] The extraction unit is configured to extract at least one characteristic factor based on the processed crowdsourced map data using a preset correlation coefficient.

[0163] Optionally, in one embodiment of the present application, the updating module 200 includes: a determining unit, a screening unit, and a correcting unit.

[0164] Wherein, the determining unit is used to determine the accuracy characteristics and observation update interval of the inertial sensor of the target navigation system using the simulated positioning error;

[0165] a screening unit, configured to screen map results that meet a preset reliability based on an accuracy characteristic, an observation update interval, and at least one characteristic factor;

[0166] The correction unit is used to correct the divergent errors of the map results to generate crowdsourced map label data that meets the preset accuracy.

[0167] Optionally, in one embodiment of the present application, the construction module 300 includes: an expansion unit, a learning unit, and an optimization unit.

[0168] The expansion unit is used to expand the channel of at least one feature factor and use temporal convolution to aggregate feature information of at least one feature factor to obtain a local feature with time dependence;

[0169] The learning unit is used to learn contextual information with dependencies from local features, and to generate position variance information based on the contextual information by focusing on feature information using an attention mechanism;

[0170] The optimization unit is used to train and optimize the initial variance self-learning model using the location variance information and crowdsourced map label data to obtain a preset variance self-learning model.

[0171] Optionally, in one embodiment of the present application, the positioning module 400 includes: an updating unit and a fusion unit.

[0172] The updating unit is used to update the data information of the strapdown inertial navigation system according to the map result measurement;

[0173] The fusion unit is used to fuse the data information and the scaled model prediction position variance through a preset semi-tight combination structure to obtain the final positioning result of the target navigation system.

[0174] It should be noted that the aforementioned explanation of the crowdsourcing data driven model and parameter self-learning navigation method embodiment is also applicable to the crowdsourcing data driven model and parameter self-learning navigation device of this embodiment, and will not be repeated here.

[0175] According to the crowdsourced data-driven model and parameter self-learning navigation device proposed in the embodiment of the present application, a certain characteristic factor can be generated by obtaining crowdsourced map data of the target navigation system, and crowdsourced map label data is obtained while updating the data information of the relevant navigation system to train a certain variance self-learning model. Finally, the model prediction variance generated by the variance self-learning model and a certain semi-tight combination architecture are integrated with the updated data information to generate a high-precision positioning result. Thus, autonomous learning of variance information is achieved without manual intervention and high-precision reference truth equipment. While achieving high-precision navigation and positioning, the system model is continuously optimized through the update of crowdsourced data, continuously improving the generalization ability and intelligence level of the system. Thus, the GNSS / SINS integrated navigation system in the related art mainly relies on model driving when achieving positioning, making it difficult to fully extract common features in large-scale data and difficult to build complex models. In addition, most algorithms mainly focus on GNSS gross error detection, and the consistency between GNSS position variance and actual error is insufficient. In addition, the algorithm training scheme also usually relies on high-precision reference system to generate labels. As a result, the accuracy of navigation and positioning results is lacking and a certain cost is increased.

[0176] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0177] A memory 901 , a processor 902 , and a computer program stored in the memory 901 and executable on the processor 902 .

[0178] When the processor 902 executes the program, the crowdsourcing data driven model and parameter self-learning navigation method provided in the above embodiment is implemented.

[0179] Furthermore, the electronic device further includes:

[0180] The communication interface 903 is used for communication between the memory 901 and the processor 902 .

[0181] The memory 901 is used to store computer programs that can be run on the processor 902 .

[0182] The memory 901 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0183] If the memory 901, the processor 902, and the communication interface 903 are implemented independently, the communication interface 903, the memory 901, and the processor 902 can be connected to each other via a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0184] Optionally, in a specific implementation, if the memory 901, the processor 902 and the communication interface 903 are integrated on a chip, the memory 901, the processor 902 and the communication interface 903 can communicate with each other through an internal interface.

[0185] The processor 902 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0186] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned crowdsourcing data-driven model and parameter self-learning navigation method.

[0187] An embodiment of the present application also provides a computer program product, including a computer program, which can run computer instructions. When the computer instructions are executed by a processor, the navigation method of the crowdsourcing data-driven model and parameter self-learning provided in the embodiment of the present application is implemented.

[0188] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0189] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this application, "N" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0190] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing a custom logical function or process step, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a different order than shown or discussed, including performing functions in a substantially simultaneous manner or in a reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0191] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or N wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program can be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing it in other suitable ways as necessary, and then storing it in a computer memory.

[0192] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented using hardware, as in another embodiment, it can be implemented using any one or a combination of the following technologies known in the art: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0193] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0194] In addition, the functional units in the various embodiments of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into a module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0195] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A crowdsourced data driven model and parameter self-learning navigation method, characterized in that: The following steps are involved: Acquiring crowdsourced map data of a target navigation system, and extracting at least one characteristic factor of the crowdsourced map data; Screening map results that meet a preset reliability based on the at least one characteristic factor, and updating the crowdsourced map data according to the map results to generate crowdsourced map label data that meets a preset accuracy; Based on the at least one characteristic factor and the crowdsourced map label data, constructing a preset variance self-learning model, and optimizing the position variance information of the target navigation system through the preset variance self-learning model to obtain a model-predicted position variance of the target navigation system; A scaling factor is constructed according to the model-predicted position variance to scale the model-predicted position variance using the scaling factor to obtain a scaled model-predicted variance, and a final positioning result of the target navigation system is generated using the scaled model-predicted position variance and a preset semi-tight combination architecture, wherein the expression of the preset semi-tight combination architecture is: in, , , is the position vector of the global navigation satellite system GNSS in the IMU coordinate system, 、 and All of these can be derived through mechanical arrangement at the current moment; is the gravity error vector, is the Earth's rotation angular velocity, 、 and are the random walk noises of position, velocity and attitude in the ECEF system, represents the IMU coordinate system, is the output of the accelerometer in the ECEF system, Represents the rotation matrix from the IMU coordinate system to the ECEF coordinate system, Indicates location, Indicates speed, To express attitude, represents the accelerometer bias, Indicates gyro zero bias; Among them, the preset variance self-learning model is constructed based on the at least one feature factor and the crowdsourced map label data, including: expanding the channel of the at least one feature factor, and using temporal convolution to aggregate the feature information of the at least one feature factor to obtain local features with time dependence; learning contextual information with dependencies from the local features, and focusing the feature information based on the context information using an attention mechanism to generate position variance information; using the position variance information and the crowdsourced map label data to train and optimize the initial variance self-learning model to obtain the preset variance self-learning model.

2. The method according to claim 1, characterized in that The extracting at least one characteristic factor of the crowdsourced map data includes: Post-processing and solving the crowd-sourced map data to obtain processed crowd-sourced map data; Based on the processed crowdsourced map data, the at least one characteristic factor is extracted using a preset correlation coefficient.

3. The method according to claim 1, characterized in that The screening of map results that meet a preset reliability based on the at least one characteristic factor, and updating the crowdsourced map data according to the map results to generate crowdsourced map label data that meets a preset accuracy, includes: The simulated positioning error is used to determine the accuracy characteristics and observation update interval of the inertial sensor of the target navigation system; screening a map result that meets a preset reliability based on the accuracy characteristic, the observation update interval, and the at least one characteristic factor; The divergence error of the map result is corrected to generate crowdsourced map label data that meets a preset accuracy.

4. The method according to claim 1, wherein The method of generating a final positioning result of the target navigation system by using the scaled model predicted position variance and a preset semi-tight combination architecture includes: Measuring and updating data information of a strapdown inertial navigation system according to the map result; The data information and the scaled model predicted position variance are fused through the preset semi-tight combination architecture to obtain a final positioning result of the target navigation system.

5. A navigation device with crowdsourced data driven model and parameter self-learning, characterized in that: include: an extraction module, configured to obtain crowdsourced map data of a target navigation system and extract at least one characteristic factor of the crowdsourced map data; an updating module, configured to screen map results that meet a preset reliability based on the at least one characteristic factor, and update the crowdsourced map data according to the map results to generate crowdsourced map label data that meets a preset accuracy; a construction module, configured to construct a preset variance self-learning model based on the at least one characteristic factor and the crowdsourced map label data, and optimize the position variance information of the target navigation system through the preset variance self-learning model to obtain a model-predicted position variance of the target navigation system; A positioning module is configured to construct a scaling factor based on the model-predicted position variance, scale the model-predicted position variance using the scaling factor, obtain a scaled model-predicted position variance, and generate a final positioning result of the target navigation system using the scaled model-predicted position variance and a preset semi-tight combination architecture, wherein the expression of the preset semi-tight combination architecture is: in, , , is the position vector of the global navigation satellite system GNSS in the IMU coordinate system, 、 and All of these can be derived through mechanical arrangement at the current moment; is the gravity error vector, is the Earth's rotation angular velocity, 、 and are the random walk noises of position, velocity and attitude in the ECEF system, represents the IMU coordinate system, is the output of the accelerometer in the ECEF system, Represents the rotation matrix from the IMU coordinate system to the ECEF coordinate system, Indicates location, Indicates speed, To express attitude, represents the accelerometer bias, Indicates gyro bias; Among them, the construction module includes: an expansion unit, which is used to expand the channel of the at least one feature factor and use temporal convolution to aggregate the feature information of the at least one feature factor to obtain a local feature with time dependence; a learning unit, which is used to learn contextual information with dependencies from the local features, and to generate position variance information based on the context information by using the attention mechanism to focus on the feature information; an optimization unit, which is used to train and optimize the initial variance self-learning model using the position variance information and the crowdsourced map label data to obtain the preset variance self-learning model.

6. The device according to claim 5, characterized in that The extraction module comprises: a processing unit, configured to post-process and solve the crowdsourced map data to obtain processed crowdsourced map data; An extraction unit is configured to extract the at least one characteristic factor based on the processed crowdsourced map data using a preset correlation coefficient.

7. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the crowdsourced data driven model and parameter self-learning navigation method according to any one of claims 1 to 4.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the crowdsourced data driven model and parameter self-learning navigation method as described in any one of claims 1 to 4.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed, it is used to implement the crowdsourcing data driven model and parameter self-learning navigation method as described in any one of claims 1 to 4.

Images