Data-driven low-precision map matching method, device, medium and equipment

Abstract

The application discloses a data-driven low-precision map matching method, and belongs to the field of map data processing. The method comprises the following steps: in each low-precision single package map, taking a randomly selected data element as a center point, a corresponding fixed range submap is generated; different kinds of data elements of multiple submaps are subjected to feature extraction to obtain feature information corresponding to the submaps; the feature information corresponding to two submaps corresponding to each center point in each two low-precision single package maps is subjected to automatic matching in sequence to obtain a matching score of different kinds of data elements between the two submaps, and an object corresponding to the most confident matching meeting geometric consistency is determined through the matching score; and according to the most confident matching, other objects corresponding to the most confident matching meeting geometric consistency adjacent to the most confident matching are obtained. The application guarantees the stability of matching, saves the equipment cost, and does not depend on the uniqueness of map elements.

Description

Technical Field

[0001] The present application relates to the field of map data processing, and in particular to a data-driven low-precision map matching method, device, medium and equipment. Background Art

[0002] In the prior art, the artificial rule method relies more on the uniqueness of map elements when processing map data. The rule method first finds the most unique object in the single-package map, then matches the unique objects, combines the matched objects together, and then finds the less unique objects in turn for matching and combination. If there are no unique elements in the single-package map, or the unique elements are missing, such as tunnel scenes, scenes without signboards or poles; the position deviation of the unique elements is large, such as the inappropriate establishment of the position of signboards and poles; in the screening and matching process, a lot of map data is discarded, all of these will lead to matching failures. The rule method relies on thresholds, and the thresholds need to be continuously adjusted for different scenarios, such as the range of a certain number of meters to be considered a unique object.

[0003] This solution can obtain high-precision maps through learning calculations using low-precision equipment, solving the problem of high cost of high-precision equipment and the reliance of existing technologies on the uniqueness of map elements. Summary of the Invention

[0004] In response to the problems in the existing technology of over-reliance on the uniqueness of map elements and high cost of obtaining high-precision maps, this application mainly provides a data-driven low-precision map matching method, device, medium and equipment.

[0005] A technical solution adopted in this application is to provide a data-driven low-precision map matching method, which includes:

[0006] Loop-segment multiple low-precision map packages containing backtracking routes for the same mission to obtain low-precision single-package maps, so that each low-precision single-package map does not contain overlapping data elements;

[0007] In each low-precision single-packet map, a submap of a fixed range is generated with a randomly selected data element as the center point;

[0008] Extract features from different data elements of multiple subgraphs to obtain feature information corresponding to the subgraphs;

[0009] The feature information of the two sub-maps corresponding to each center point in each of the two low-precision single-packet maps is automatically matched in sequence to obtain the matching scores of different data elements between the two sub-maps. The matching scores are used to determine the object with the most confident matching that meets the geometric consistency.

[0010] According to the most confident match, obtain objects corresponding to other most confident matches that are adjacent to the most confident match and meet geometric consistency;

[0011] Align multiple low-precision single-packet maps with the most confident matching objects one by one to obtain a high-precision single-packet map;

[0012] The high-precision single-package maps are integrated and optimized to obtain a high-precision map.

[0013] Another technical solution adopted by this application is to provide a data-driven low-precision map matching device, which includes:

[0014] A module for loop-segmenting multiple low-precision map packages containing backtracks in the same mission to obtain low-precision single-package maps, so that each low-precision single-package map does not contain overlapping data elements;

[0015] A module for generating a submap of a fixed range in each low-precision single-packet map, with a randomly selected data element as the center point;

[0016] A module for extracting features from different data elements of multiple subgraphs to obtain feature information corresponding to the subgraphs;

[0017] Automatically match the feature information of the two sub-images corresponding to each center point in each of the two low-precision single-packet maps in sequence, obtain the matching scores of different data elements between the two sub-images, and determine the module that most confidently matches the corresponding object in accordance with the matching scores;

[0018] A module for obtaining, based on the most confident match, objects corresponding to other most confident matches that are adjacent to the most confident match and meet geometric consistency;

[0019] A module for aligning multiple low-precision single-packet maps with the most confidently matched objects one by one to obtain a high-precision single-packet map;

[0020] A module used to fuse and optimize high-precision single-package maps to obtain high-precision maps.

[0021] Another technical solution adopted in this application is: providing a computer-readable storage medium storing computer instructions, which are operated to execute the data-driven low-precision map matching method in solution one.

[0022] Another technical solution adopted in the present application is: providing a computer device, which includes a processor and a memory, wherein the memory stores computer instructions, and the computer instructions are operated to execute the data-driven low-precision map matching method in solution one.

[0023] The beneficial effects achieved by the technical solution of the present application are: the present application designs a data-driven low-precision map matching method, device, medium and equipment. The present application guarantees the matching process of data-driven matching by ring cutting of low-precision single package map, and saves the equipment cost; the accuracy of matching is guaranteed by generating sub-maps; the most confident matching does not depend on the uniqueness of map elements. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor under the premise of not paying any creative labor.

[0025] Figure 1 is a schematic diagram of one specific embodiment of a data-driven low-precision map matching method of the present application;

[0026] Figure 2 is a schematic diagram of one specific embodiment of a data-driven low-precision map matching device of the present application.

[0027] Through the above drawings, the specific embodiments of the present application have been shown, and more detailed description will be given in the following. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application by any means, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0028] The preferred embodiments of the present application will be described in detail below with reference to the drawings, so that the advantages and features of the present application can be more easily understood by those skilled in the art, and the protection scope of the present application can be more clearly and definitely defined.

[0029] It should be noted that in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the elements defined by the statement "include" do not exclude the presence of other identical elements in the process, method, article or equipment including the elements.

[0030] Rules can be understood as rolebases, which are manually defined rules (thresholds). These thresholds specifically refer to, for example, artificially defined uniqueness (e.g., if there is only one sign within a 10-meter radius, then that sign is unique). A sign should match another sign 20 meters away. A 10-meter threshold is inappropriate in this case. Increasing it to 30 meters would result in a range that is too large, leading to matches with signs that should not be matched. Setting the threshold for unique elements relies entirely on human experience. Thresholds include distance, angle, and orientation values ​​that represent the space. Poor quality of single-package mapping can lead to significant deviations in the positions of unique elements. Setting a high threshold can introduce false matches, while setting a low threshold can result in more discarded matches.

[0031] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0032] Figure 1 A specific implementation of a data-driven low-precision map matching method of the present application is shown. Figure 1 In the specific implementation shown, the data-driven low-precision map matching method includes:

[0033] Step S101 , loop segmenting is performed on multiple low-precision map packages containing backtracking routes in the same task to obtain low-precision single-package maps, so that each low-precision single-package map does not contain overlapping data elements.

[0034] In this embodiment, a task can create multiple low-precision single-package maps. However, in order to prevent the same path from being mapped twice, the low-precision single-package map containing the backtracking route needs to be split at the backtracking route to produce two low-precision single-package maps that do not contain the backtracking route to facilitate subsequent matching.

[0035] It should be noted that a low-precision single-package map refers to a vector map that collects spatial and semantic information about traffic elements such as traffic signs and lane lines collected by a vehicle during a trip through a section of road.

[0036] In a specific example of the present application, a low-precision device can collect one trip of data to build a low-precision single-package map. The data collected in one trip all belong to the information in the low-precision single-package map, and the low-precision single-package maps built by multiple trips of collection are matched, fused and optimized. The input of digital driven matching (DDMatch) is the low-precision single-package map after single-package mapping. Because there are low-precision collection devices, such as GPS, their accuracy is not high and their computing power is limited, resulting in unreliable perception results. This method adopts a two-by-two pairing method, and the same object is only paired with one data element in the low-precision single-package map. If a low-precision single-package map has a return route, it is equivalent to walking through the same place twice, building two maps, and having two data elements. The same object will correspond to the two data elements shown in the low-precision single-package map.

[0037] exist Figure 1 In the specific implementation shown, the data-driven low-precision map matching method further includes:

[0038] Step S102 : In each low-precision single-packet map, a sub-map of a corresponding fixed range is generated with a randomly selected data element as the center point.

[0039] In this embodiment, there will be many sub-maps in a low-precision single-packet map. The low-precision single-packet map is of indefinite size, and the sub-maps are of fixed size. The purpose of the sub-maps is to facilitate matching. There is a large amount of overlap between multiple sub-maps for the purpose of subsequent redundant matching (redundancy mechanism) to ensure the robustness of the matching.

[0040] In a specific embodiment of the present application, in each low-precision single-packet map, a corresponding sub-map of a fixed range is generated with a randomly selected data element as the center point, including: distributing data elements with predetermined distance intervals in each low-precision single-packet map, and generating a corresponding sub-map of a fixed range with each data element as the center point, wherein the fixed range is the product of a preset length and a preset width, and the preset length and the preset width are greater than the predetermined distance interval.

[0041] In this embodiment, each low-precision single-packet map in a task is covered by center points separated by a predetermined distance, each data element may exist in dozens of sub-maps, and digital-driven matching is the matching between two sub-maps.

[0042] In one specific example of this application, all low-precision single-packet maps in a mission are covered by center points 10-20 meters apart. Each center point generates a 100m x 100m submap, where 100 > 10, resulting in a significant amount of overlap. A single center point may be present in more than a dozen submaps.

[0043] In a specific example of this application, digital drive (DDMatch) involves matching two submaps at the same location on two low-precision single-packet maps. The two low-precision single-packet maps are of different lengths. If a straight section in one of the low-precision single-packet maps suddenly turns, causing the trajectory to deviate from that of the other low-precision single-packet map, the matching accuracy will decrease. Therefore, the low-precision single-packet map must be broken up and multiple submaps generated for matching. The submaps are generated by randomly selecting center points, with no specific rules for which center points to choose.

[0044] exist Figure 1 In the specific implementation shown, the data-driven low-precision map matching method further includes:

[0045] Step S103 : extracting features from different types of data elements of the multiple subgraphs to obtain feature information corresponding to the subgraphs.

[0046] In this embodiment, one-to-one matching between different types of data elements requires extracting the features of the data elements themselves and their surrounding elements to facilitate subsequent matching.

[0047] In a specific embodiment of the present application, feature extraction is performed on different types of data elements in multiple subgraphs to obtain feature information corresponding to the subgraphs, including: performing feature extraction based on the surrounding elements and the own attributes of a data element respectively to obtain corresponding semantic feature information and perceptual feature information, and superimposing the semantic feature information and the perceptual feature information to obtain overall feature information corresponding to a data element; performing feature extraction based on the own attributes of another data element to obtain attribute feature information, and then obtaining feature information corresponding to different types of data elements in the same subgraph, wherein the feature information includes but is not limited to type, size, position coordinates, orientation, and / or height.

[0048] In this embodiment, a low-precision single-packet map is divided into multiple sub-maps, and features are extracted from different data elements in each sub-map. The extracted features can make the matching of data elements easier and more convenient.

[0049] In one specific example of the present application, the data elements in the subgraph are divided into two types, one is Landmarks, which is composed of points such as signs and traffic lights, and the other is Lines, which is composed of lines such as lane lines. Landmarks are extracted according to the information of other surrounding elements (types, positions (direction, depth)), and also according to their own attributes (x, y, type); Lines are composed of a series of sampling point information, and are extracted according to their own attributes (x, y, type). The current features of Landmarks include: (category, x, y, smbert); in the future, size, orientation, height from the ground, etc. will be added; in the experiment, smbert can indeed enhance the matching of landmarks, but the disadvantage is that it is time-consuming.

[0050] In Figure 1 In the specific embodiment shown in the specific embodiment, the data-driven low-precision map matching method further comprises:

[0051] In step S104, the feature information corresponding to the two subgraphs corresponding to each center point in each of the two low-precision single package maps is sequentially automatically matched to obtain the matching scores of different types of data elements between the two subgraphs, and the most confident matching corresponding object is determined through the matching scores.

[0052] In this embodiment, when the two low-precision single package maps are matched, multiple pairs of subgraphs are matched at the same time, and the matching scores of different types of data elements are obtained respectively. According to these matching scores, the most confident matching corresponding object is obtained through geometric consistency verification. This method does not depend on the uniqueness of map elements at all, and will not discard map data, and does not need to constantly adjust the rule threshold.

[0053] In one specific embodiment of the present application, the feature information corresponding to the two subgraphs corresponding to each center point in each of the two low-precision single package maps is sequentially automatically matched to obtain the matching scores of different types of data elements between the two subgraphs, including: the feature information corresponding to the two subgraphs corresponding to each center point in each of the two low-precision single package maps is simultaneously input into the attention graph neural network; the attention graph neural network simultaneously performs attention automatic learning on the feature information corresponding to the two subgraphs to obtain the matching scores of different types of data elements between the two subgraphs.

[0054] In this embodiment, the attention graph neural network is already trained, and has high operation efficiency. According to the attention mechanism, different attention scores are allocated to identify more important corresponding relationships.

[0055] In a specific embodiment of the present application, the object corresponding to the most confident match that meets the geometric consistency is determined by the matching score, including: when the matching score is consistent with the baseline threshold, the data element corresponding to the matching score meets the geometric consistency; and when the number of matching scores that are consistent with the baseline threshold accounts for more than a preset threshold in the matching score, the data element corresponding to the matching score is the object corresponding to the most confident match.

[0056] In this embodiment, the accuracy of matching is ensured by setting the benchmark threshold and the ratio.

[0057] In a specific example of the present application, each pair of subgraphs corresponds to a match, and a match will have multiple matching results; if 14 out of 15 matching results give the same match, it is the most confident match of the network. The match between map elements constructed for the same object in two low-precision single-package maps. One center point corresponds to one subgraph. The two subgraphs share a set of center points in the world coordinate system, and the center points of the two subgraphs are the same set of center points with the same number, that is, the center points are the same and the size is the same, but they are matched between corresponding subgraphs in different low-precision single-package maps. The 15 matching results may be because multiple subgraphs contain map elements of the same object, that is, the same center point appears in 15 subgraphs.

[0058] In a specific embodiment of the present application, before the feature information corresponding to the two sub-graphs corresponding to each center point in each two low-precision single-packet maps is simultaneously input into the attention graph neural network, it also includes: constructing pseudo-single-packet data with complete true values; using the pseudo-single-packet data to pre-train the attention graph neural network; and using the incomplete true values ​​on the production line to train the pre-trained attention graph neural network.

[0059] In this embodiment, the attention graph neural network can be used for matching through two steps of pre-training and re-training.

[0060] In a specific example of the present application, a pseudo single packet is newly generated and used to train the attention graph neural network for the first time, making up for the problem of missing true value data; the second training of the attention graph neural network is trained using the actual true value on the production line. The complete true value means that all elements in the single packet map have a true value match. The production line uses the rolebase method to generate true values, and these true values ​​are missing. The pseudo single packet is to construct a batch of new matches, and the pseudo single packet and the original single packet are a new single packet match.

[0061] exist Figure 1 In the specific implementation shown, the data-driven low-precision map matching method further includes:

[0062] Step S105 : Based on the most confident match, obtain objects corresponding to other most confident matches that are adjacent to the most confident match and meet geometric consistency.

[0063] In this embodiment, if a pair of matches is the same object, then geometric consistency is achieved. In the vicinity of the most confident match, the matching results of the uncertain objects are checked to ensure that all objects in the low-precision single packet map are not missed, thus ensuring the completeness of the match.

[0064] In a specific embodiment of the present application, based on the most confident match, objects corresponding to other most confident matches that are adjacent to the most confident match and meet the geometric consistency are obtained, including: based on the most confident match, using the first correlation threshold in the first level and the second correlation threshold in the second level, iteratively selecting objects corresponding to other most confident matches that are adjacent to the most confident match and meet the geometric consistency multiple times.

[0065] In this embodiment, each time a batch of most confident matches is found, the most confident matches among the uncertain objects are sequentially searched for near the batch of most confident matches. The two levels of screening and matching of the uncertain objects ensure the accuracy of the matching.

[0066] exist Figure 1 In the specific implementation shown, the data-driven low-precision map matching method further includes:

[0067] Step S106: align the objects that correspond to the most confident matches in the multiple low-precision single-packet maps one by one to obtain a high-precision single-packet map.

[0068] In this embodiment, the matching objects in each two low-precision single-packet maps are combined through multiple iterations of pose graph optimization to obtain high-precision single-packet maps, thereby improving the accuracy of the map.

[0069] exist Figure 1 In the specific implementation shown, the data-driven low-precision map matching method further includes:

[0070] Step S107: fuse and optimize the high-precision single-package map to obtain a high-precision map.

[0071] In this embodiment, the map accuracy can be improved by fusion optimization.

[0072] This application performs loop segmentation on the low-precision single-packet map obtained by mapping single packets collected by low-precision equipment, solving the problem of inaccurate backtracking matching in DDMatch. By splitting the low-precision single-packet map into multiple sub-maps, the two sub-maps corresponding to the same center point between each two low-precision single-packet maps are matched. The matched objects are optimized through multiple iterations of the pose graph. The individual low-precision single-packet maps are combined to obtain a high-precision single-packet map. The high-precision single-packet maps are then fused and optimized to obtain a high-precision map. This solution does not rely on the uniqueness of map elements or the thresholds set by rules. It not only saves equipment costs but also reduces computational complexity.

[0073] Figure 2 The specific implementation of a data-driven low-precision map matching device of the present application is shown. Figure 2 In the specific implementation shown, the data-driven low-precision map matching device mainly includes:

[0074] Module 201 is a module for performing loop segmentation on multiple low-precision map packages containing backtracking routes in the same mission to obtain low-precision single-package maps, so that each low-precision single-package map does not contain overlapping data elements;

[0075] Module 202 is a module for generating a corresponding sub-map of a fixed range with a randomly selected data element as the center point in each low-precision single-packet map;

[0076] Module 203 is a module for extracting features from different data elements of multiple subgraphs to obtain feature information corresponding to the subgraphs;

[0077] Module 204 is configured to automatically match the feature information corresponding to the two sub-images corresponding to each center point in each of the two low-precision single-packet maps in sequence, obtain a matching score for different data elements between the two sub-images, and determine the object with the highest confidence that the matching object meets the geometric consistency based on the matching score;

[0078] Module 205 is a module for obtaining objects corresponding to other most confident matches that are adjacent to the most confident match and meet geometric consistency based on the most confident match;

[0079] Module 206 is configured to align the objects that correspond to the most confident matches in the multiple low-precision single-packet maps one by one to obtain a high-precision single-packet map;

[0080] Module 207 is a module for fusing and optimizing high-precision single-package maps to obtain high-precision maps.

[0081] In this implementation, the present application solves the problem of poor matching of digital drive matching on return routes through loop segmentation. By splitting low-precision single-packet maps into multiple sub-maps for matching, this solution is more efficient and accurate, and solves the high cost of high-precision equipment. Through the most confident matching, this solution does not rely on the uniqueness of map elements. This solution also saves computing power.

[0082] The data-driven low-precision map matching device provided in this application can be used to execute the data-driven low-precision map matching method described in any of the above embodiments. Its implementation principles and technical effects are similar and will not be repeated here.

[0083] In a specific embodiment of the present application, each functional module in a data-driven low-precision map matching device of the present application can be directly in hardware, in a software module executed by a processor, or in a combination of the two.

[0084] The software modules may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from and write information to the storage medium.

[0085] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof. A general-purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other such configuration. In the alternative, the storage medium may be integral to the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In the alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0086] In another specific embodiment of the present application, a computer-readable storage medium stores computer instructions, and the computer instructions are operated to execute the data-driven low-precision map matching method in any embodiment.

[0087] In another specific embodiment of the present application, a computer device includes a processor and a memory, wherein the memory stores computer instructions, and the computer instructions are operated to execute the data-driven low-precision map matching method in any embodiment.

[0088] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0089] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0090] The above description is merely an embodiment of the present application and does not limit the patent scope of the present application. Any equivalent structural transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A data-driven low-precision map matching method, characterized in that: include: Loop-segment multiple low-precision map packages containing backtracking routes for the same task to obtain low-precision single-package maps, so that each low-precision single-package map does not contain overlapping data elements; In each of the low-precision single-packet maps, a corresponding sub-map with a fixed range is generated with a randomly selected data element as a center point, and there are overlapping data elements between multiple sub-maps; Extracting features from different data elements of the plurality of subgraphs to obtain feature information corresponding to the subgraphs; Automatically matching the feature information corresponding to the two sub-images corresponding to each center point in each of the two low-precision single-packet maps in sequence to obtain a matching score for different data elements between the two sub-images, and determining the object with the most confident matching that meets geometric consistency based on the matching score; According to the most confident match, obtaining objects corresponding to other most confident matches that are adjacent to the most confident match and meet the geometric consistency; Aligning the objects corresponding to the most confident matches in the plurality of low-precision single-packet maps one by one to obtain a high-precision single-packet map; The high-precision single-package maps are fused and optimized to obtain a high-precision map.

2. The data-driven low-precision map matching method according to claim 1, characterized in that: In each of the low-precision single-packet maps, a submap of a corresponding fixed range is generated with a randomly selected data element as the center point, including: Data elements with predetermined distance intervals are distributed in each of the low-precision single-packet maps, and a corresponding sub-map with a fixed range is generated with each data element as the center point, wherein the fixed range is the product of a preset length and a preset width, and the preset length and the preset width are greater than the predetermined distance interval.

3. The data-driven low-precision map matching method according to claim 1, wherein: The extracting features of different data elements of the plurality of subgraphs to obtain feature information corresponding to the subgraphs includes: Extract features based on surrounding elements and the data element's own attributes to obtain corresponding semantic feature information and perceptual feature information, and superimpose the semantic feature information and the perceptual feature information to obtain overall feature information corresponding to the data element; Feature extraction is performed based on the own attributes of another data element to obtain attribute feature information, and then feature information corresponding to different types of data elements in the same subgraph is obtained, wherein the feature information includes type, size, position coordinates, orientation and height.

4. The data-driven low-precision map matching method according to claim 1, wherein: Automatically matching the feature information corresponding to the two sub-maps corresponding to each center point in each of the two low-precision single-packet maps in sequence to obtain matching scores of different data elements between the two sub-maps includes: Inputting feature information corresponding to the two sub-maps corresponding to each of the center points in each of the two low-precision single-packet maps into the attention graph neural network at the same time; The attention graph neural network simultaneously performs automatic attention learning on the feature information corresponding to the two subgraphs to obtain matching scores of different data elements between the two subgraphs.

5. The data-driven low-precision map matching method according to claim 1, wherein: The obtaining, based on the most confident match, objects corresponding to other most confident matches that are adjacent to the most confident match and meet the geometric consistency includes: According to the most confident match, objects corresponding to other most confident matches that are adjacent to the most confident match and meet the geometric consistency are selected in multiple iterations using a first correlation threshold in the first level and a second correlation threshold in the second level.

6. The data-driven low-precision map matching method according to claim 1, characterized in that: Determining the object corresponding to the most confident match that meets geometric consistency based on the matching score includes: When the matching score is consistent with the reference threshold, the data element corresponding to the matching score meets the geometric consistency; And when the proportion of the number of matching scores that are consistent with the reference threshold in the matching scores is greater than a preset threshold, the data element corresponding to the matching score is the object that is most confidently matched.

7. The data-driven low-precision map matching method according to claim 4, characterized in that: Before simultaneously inputting the feature information corresponding to the two sub-maps corresponding to each of the center points in each of the two low-precision single-packet maps into the attention graph neural network, the method further includes: Construct pseudo single packet data with complete true value; Pre-training the attention graph neural network using the pseudo single packet data; Use the incomplete ground truth from the production line to train the pre-trained attention graph neural network.

8. A data-driven low-precision map matching device, characterized in that: include: A module for performing loop segmentation on multiple low-precision map packages containing backtracks in the same mission to obtain low-precision single-package maps, so that each of the low-precision single-package maps does not contain overlapping data elements; A module for generating, in each of the low-precision single-packet maps, a corresponding sub-map with a randomly selected data element as a center point, wherein there are overlapping data elements between multiple sub-maps; A module for extracting features from different types of data elements of the plurality of subgraphs to obtain feature information corresponding to the subgraphs; Automatically matching the feature information corresponding to the two sub-images corresponding to each center point in each of the two low-precision single-packet maps in sequence to obtain matching scores of different data elements between the two sub-images, and determining a module that most confidently matches the corresponding object according to the matching scores; A module for obtaining, based on the most confident match, objects corresponding to other most confident matches adjacent to the most confident match and meeting the geometric consistency; A module for aligning the objects corresponding to the most confident matches in the plurality of low-precision single-packet maps one by one to obtain a high-precision single-packet map; A module for fusing and optimizing the high-precision single-package map to obtain a high-precision map.

9. A computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are operated to execute the data-driven low-precision map matching method according to any one of claims 1-7.

10. A computer device comprising a processor and a memory, wherein the memory stores computer instructions, wherein the processor operates the computer instructions to execute the data-driven low-precision map matching method according to any one of claims 1 to 7.

Images