Multi-path multi-packet matching and fusion method and device based on loop segmentation judgment

Abstract

This application discloses a multi-path, multi-packet matching and fusion method based on loopback segmentation judgment, belonging to the field of map data processing. The method includes acquiring multiple low-precision map packets from the same task; determining whether each low-precision map packet contains backtracking paths and segmenting backtracking paths to obtain low-precision single-packet maps; in each low-precision single-packet map, generating a corresponding sub-map with a fixed range, using randomly selected data elements as center points; simultaneously performing feature extraction and automatic matching on the two sub-maps corresponding to each center point of every two adjacent low-precision single-packet maps to obtain matching scores for different data elements between the two sub-maps; and determining the object corresponding to the most certain match that meets geometric consistency based on the matching scores. This application can save equipment costs, overcome the reliance on the uniqueness of map elements in existing technologies, and the established low-precision single-packet maps facilitate subsequent matching.

Description

Technical Field

[0001] This application relates to the field of map data processing, and in particular to a multi-path multi-package matching and fusion method and apparatus based on loop closure segmentation judgment. Background Technology

[0002] In existing technologies, rule-based methods rely heavily on the uniqueness of map elements when processing map data. These methods first identify the most unique objects within a single map packet, then match these unique objects together, and so on, finding and matching less unique objects. Matching failures can occur if the single map packet lacks unique elements, or if unique elements are missing (e.g., in tunnel scenes or scenes without signs or poles); if unique elements are poorly positioned (e.g., improperly placed signs or poles); or if a large amount of map data is discarded during the matching process. Rule-based methods also rely on thresholds, requiring continuous adjustment for different scenarios, such as determining the range within which an object is considered unique.

[0003] This solution utilizes low-precision equipment to generate high-precision maps through learning and computation, thus solving the problems of high cost of high-precision equipment and reliance on the uniqueness of map elements in existing technologies. Summary of the Invention

[0004] To address the problem of excessive reliance on the uniqueness of map elements and high cost of obtaining high-precision maps in existing technologies, this application mainly provides a multi-path multi-packet matching and fusion method and apparatus based on loop closure segmentation judgment.

[0005] One technical solution adopted in this application is: providing a multi-path multi-packet matching and fusion method based on loopback segmentation judgment, which includes:

[0006] Obtain multiple low-precision map packages from the same task, determine whether each low-precision map package contains backtracking paths and segment backtracking paths 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-package map, a sub-map with a fixed range is generated with a randomly selected data element as the center point;

[0008] In each of the two adjacent low-precision single-packet maps, the two sub-maps corresponding to each center point are simultaneously subjected to feature extraction and automatic matching to obtain the matching score corresponding to different data elements between the two sub-maps. The object corresponding to the most certain match that meets geometric consistency is determined by the matching score.

[0009] Based on the most certain match, obtain the objects corresponding to other most certain matches that are geometrically consistent and are adjacent to the most certain match;

[0010] Align the objects with the most certain matching from multiple low-precision single-packet maps one by one to obtain a high-precision single-packet map.

[0011] By merging and optimizing high-precision single-package maps, a high-precision map is obtained.

[0012] Another technical solution adopted in this application is: providing a multi-path multi-packet matching and fusion device based on loopback segmentation judgment, which includes:

[0013] This module is used to obtain multiple low-precision map packages in the same task, determine whether each low-precision map package contains backtracking and segment backtracking, and obtain low-precision single-package maps so that each low-precision single-package map does not contain overlapping data elements.

[0014] A module used to generate a sub-map with a fixed range in each low-precision single-package map, with randomly selected data elements as the center point;

[0015] The module is used to simultaneously extract features and automatically match the two sub-maps corresponding to each center point in each of the two adjacent low-precision single-package maps, and obtain the matching score corresponding to different data elements between the two adjacent sub-maps. The module determines the object corresponding to the most certain match that meets geometric consistency through the matching score.

[0016] A module for obtaining objects that are geometrically consistent and adjacent to the most certain match, based on the most certain match;

[0017] This module is used to align the objects that are most confidently matched among multiple low-precision single-packet maps one by one to obtain a high-precision single-packet map.

[0018] This module is used to merge and optimize high-precision single-package maps to obtain high-precision maps.

[0019] Another technical solution adopted in this application is: providing a computer-readable storage medium storing computer instructions, which are operated to execute the multi-path multi-packet matching and fusion method based on loop-loop segmentation judgment in Solution 1.

[0020] Another technical solution adopted in this application is: providing a computer device, which includes a processor and a memory, the memory storing computer instructions, which are operated to execute the multi-path multi-packet matching and fusion method based on loop-loop segmentation judgment in Solution 1.

[0021] The beneficial effects of the technical solution of this application are as follows: This application designs a multi-path multi-packet matching and fusion method and device based on loop-loop segmentation judgment. This application judges and performs loop-loop segmentation on low-precision single-packet maps containing backtracking paths, so that each low-precision single-packet map will not be built repeatedly, and the low-precision single-packet map saves equipment costs, saves computing power through most certain matching, and does not depend on the uniqueness of map elements. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of a specific implementation of a multi-path multi-packet matching and fusion method based on loop closure segmentation judgment according to this application;

[0024] Figure 2 This is a schematic diagram of a specific implementation of a multi-path multi-packet matching and fusion device based on loopback segmentation judgment according to this application.

[0025] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0026] The preferred embodiments of this application will now be described in detail with reference to the accompanying drawings, so that the advantages and features of this application can be more easily understood by those skilled in the art, thereby providing a clearer and more definite definition of the scope of protection of this application.

[0027] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0028] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0029] Figure 1 This paper illustrates a specific implementation of a multi-path, multi-packet matching and fusion method based on loopback segmentation judgment, as proposed in this application. Figure 1 In the specific implementation shown, the multi-path multi-packet matching and fusion method based on loopback segmentation judgment includes:

[0030] Step S101: Obtain multiple low-precision map packages from the same task, determine whether each low-precision map package contains backtracking paths and segment backtracking paths to obtain low-precision single-package maps, so that each low-precision single-package map does not contain overlapping data elements.

[0031] In this implementation, multiple low-precision single-package maps can be created for the same task. However, in order to prevent the same path from being created twice and causing the map to fail to match the correct map element during matching, the low-precision single-package map containing backtracking needs to be split at the backtracking point to create two low-precision single-package maps that do not contain backtracking, which facilitates subsequent matching.

[0032] 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 by a vehicle traveling along a section of road.

[0033] In one specific embodiment of this application, determining whether a low-precision map packet contains backtracking and segmenting it to obtain a low-precision single-packet map includes: if the low-precision map contains backtracking, segmenting the low-precision map packet at the backtracking point to obtain a low-precision single-packet map that does not contain overlapping data elements, wherein the low-precision single-packet map includes a low-precision map packet that does not contain backtracking and a segmented low-precision map packet that contains backtracking.

[0034] In this embodiment, the same task contains multiple low-precision single-package maps. Some low-precision single-package maps contain backtracking paths, while others do not. In order to facilitate the matching of map elements, the low-precision single-package maps containing backtracking paths need to be segmented at the backtracking paths so that each object corresponds to one map element in a low-precision single-package map.

[0035] In a specific example of this application, a low-precision device can collect data in one pass to build a low-precision single-package map. All data collected in one pass belongs to the information within this low-precision single-package map. The low-precision single-package maps built from multiple passes are then matched, fused, and optimized. Because low-precision data collection devices, such as GPS, have low accuracy and limited computing power, the perception results are unreliable. This method uses a pairwise pairing approach; the same object is paired with only one type of data element in the low-precision single-package map. If a low-precision single-package map has a backtracking mechanism, it's equivalent to traversing the same location twice, building maps twice, and having two data elements. The same object will correspond to two data elements represented in the low-precision single-package map.

[0036] exist Figure 1 In the specific implementation shown, the multi-path multi-packet matching and fusion method based on loopback segmentation judgment further includes:

[0037] Step S102: In each low-precision single-package map, a sub-map with a fixed range is generated, centered on a randomly selected data element.

[0038] In this embodiment, a low-precision single-packet map contains many sub-maps. The low-precision single-packet map is of variable size, while 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, which is for subsequent redundant matching (redundancy mechanism) to ensure the robustness of the matching.

[0039] In a specific instance of this application, all low-precision single-package maps in a task are covered by center points spaced 10–20 meters apart; each center point can generate a submap with a fixed range of 100 meters by 100 meters; where 100 is much larger than 10, there is a large amount of overlap between any two adjacent submaps. A single center point may exist in more than a dozen submaps.

[0040] In one specific embodiment of this application, each center point in each of two adjacent low-precision single-package maps is simultaneously subjected to feature extraction and automatic matching of the two sub-maps to obtain matching scores for different data elements between the two sub-maps. This includes: simultaneously extracting features from the different data elements corresponding to the two sub-maps to obtain feature information for the two sub-maps; and inputting the feature information for the two sub-maps into an attention map neural network for automatic matching to obtain matching scores.

[0041] In this embodiment, the information of the data elements to be matched is extracted to lay the foundation for subsequent matching; different data elements in each sub-graph can be extracted simultaneously. When matching, the feature information of the data elements in the corresponding sub-graphs of the same center point in two low-precision single-package maps is matched simultaneously, which is efficient and convenient.

[0042] In one specific embodiment of this application, feature extraction is performed on different types of data elements corresponding to two sub-graphs simultaneously to obtain feature information corresponding to the two sub-graphs. The method further includes: in the same sub-graph, feature extraction is performed based on the surrounding elements and the attributes of a data element to obtain corresponding semantic feature information and perceptual feature information; the semantic feature information and perceptual feature information are superimposed to obtain the overall feature information corresponding to the data element; feature extraction is performed based on the attributes of another data element to obtain attribute feature information, thereby obtaining feature information corresponding to different types of data elements in the same sub-graph.

[0043] In this embodiment, the feature information includes, but is not limited to, type, size, location coordinates, orientation, and / or height; the semantic feature information includes, but is not limited to, the type, relative depth, and orientation of other elements surrounding the data element; the perceptual feature information and attribute feature information include the location coordinates of the data element. Dividing a low-precision single-package map into multiple sub-maps and extracting features from different data elements in each sub-map makes matching data elements easier and more convenient.

[0044] In one specific embodiment of this application, the feature information corresponding to two subgraphs is input into the attention graph neural network for automatic matching. The method further includes: the attention graph neural network automatically learning the attention between a data element and other data elements in the same subgraph, and automatically learning the attention between the same data element in two subgraphs.

[0045] In this embodiment, the attention graph neural network is pre-trained and has high computational efficiency. Different attention scores are assigned based on the attention mechanism to identify more important correspondences. The attention graph neural network automatically learns the attention between different types of data elements in the same subgraph, enabling it to better identify objects corresponding to different data elements in the same subgraph; the attention graph neural network also automatically learns the attention between the same type of data elements in two subgraphs, enabling it to better identify the correspondence between the two subgraphs.

[0046] In one specific embodiment of this application, before inputting the feature information corresponding to the two sub-graphs into the attention graph neural network for automatic matching, the method further includes: constructing pseudo single-packet data with complete ground truth; using the pseudo single-packet data to pre-train the attention graph neural network; and using the incomplete ground truth from the production line to train the pre-trained attention graph neural network.

[0047] In this embodiment, the attention graph neural network is enabled to perform matching between subgraphs through two training steps: pre-training and re-training.

[0048] In a specific instance of this application, the pseudo-single packet is newly generated and used for the first training of the attention map neural network to compensate for the lack of ground truth data; the second training of the attention map neural network is performed using actual ground truth data from the production line. Complete ground truth refers to every element in the low-precision single packet map having a ground truth match. Ground truth data is generated on the production line using the rolebase method, and this ground truth data is missing.

[0049] In one specific embodiment of this application, determining the object corresponding to the most certain match that meets geometric consistency by matching score includes: when the matching score is consistent with a benchmark threshold, the data element corresponding to the matching score meets geometric consistency; and when the proportion of the number of matching scores that are consistent with the benchmark threshold in the matching score is greater than a preset threshold, the data element corresponding to the matching score is the object corresponding to the most certain match.

[0050] In this embodiment, multiple matching results can be filtered out using a benchmark threshold. All multiple matching results meet the geometric consistency requirement. When the proportion of the benchmark threshold's number to the total number of matching scores reaches a preset threshold, it is considered the most certain match. This ensures the accuracy of the matching.

[0051] In a specific example of this application, each pair of subgraphs corresponds to one match, and one match can have multiple matching results. If 14 out of 15 matching results give the same match, then this is the most confident match from the network. Matching of map elements for the same object constructed in two low-precision single-packet maps. One center point corresponds to one subgraph. The two subgraphs share a common set of center points in a world coordinate system. The center points of the two subgraphs are the same set of center points with the same number of points, i.e., the center points are the same and the same size; the only difference is the matching between the corresponding subgraphs in different low-precision single-packet maps. 15 matching results may be because multiple subgraphs contain map elements of the same object, meaning that 15 subgraphs have the same center point.

[0052] exist Figure 1 In the specific implementation shown, the multi-path multi-packet matching and fusion method based on loopback segmentation judgment further includes:

[0053] In step S103, for each center point in each of the two adjacent low-precision single-packet maps, feature extraction and automatic matching are performed simultaneously on the two sub-maps corresponding to each center point. The matching scores of different data elements between the two sub-maps are obtained, and the objects corresponding to the most certain matching that meet the geometric consistency are determined by the matching scores.

[0054] In this implementation, when matching two low-precision single-package maps, multiple pairs of sub-maps are matched simultaneously, obtaining matching scores for different types of data elements. Geometric consistency is then verified based on these matching scores to determine the object with the highest certainty of a match. This method does not rely on the uniqueness of map elements, does not discard map data, and does not require continuous adjustment of rule thresholds.

[0055] exist Figure 1 In the specific implementation shown, the multi-path multi-packet matching and fusion method based on loopback segmentation judgment further includes:

[0056] Step S104: Based on the most certain match, obtain the objects corresponding to other most certain matches that are geometrically consistent and adjacent to the most certain match.

[0057] In this implementation, if a pair of matches represents the same object, it meets the geometric consistency requirement. Near the most certain match, the matching results of uncertain objects are checked to ensure that no object is missed in the low-precision single-package map, guaranteeing the completeness of the matching.

[0058] exist Figure 1 In the specific implementation shown, the multi-path multi-packet matching and fusion method based on loopback segmentation judgment further includes:

[0059] Step S105: Align the objects that are most certain to match in the multiple low-precision single-packet maps one by one to obtain a high-precision single-packet map.

[0060] In this embodiment, through multiple iterations of pose graph optimization, matching objects from every two low-precision single-packet maps are combined to obtain high-precision single-packet maps. This improves the accuracy of the maps.

[0061] exist Figure 1 In the specific implementation shown, the multi-path multi-packet matching and fusion method based on loopback segmentation judgment further includes:

[0062] Step S106: Merge and optimize the high-precision single-package maps to obtain a high-precision map.

[0063] In this embodiment, the map accuracy is improved through fusion optimization.

[0064] This application uses low-precision acquisition equipment to collect map data and establish low-precision single-package maps. To improve matching accuracy, the low-precision single-package maps with backtracking paths are segmented so that each low-precision single-package map does not contain duplicate map data elements. Each low-precision single-package map has a center point distribution with a certain interval. Multiple center points are randomly selected to generate multiple corresponding sub-maps with fixed ranges. Each low-precision single-package map has a variable length, while each sub-map has a fixed length. The fixed-length sub-maps facilitate subsequent matching between data elements. The data elements in each sub-map are categorized and their features are extracted to extract the information of the elements to be matched, laying the foundation for subsequent matching. Data elements between two sub-maps corresponding to the same center point in two low-precision single-package maps are matched to obtain a matching score. The most confidently matched object is determined by comparing the matching score with a benchmark threshold. The results of other uncertainly matched objects are checked in the vicinity of the first batch of most confidently matched objects. This scheme can save equipment costs, does not rely on the uniqueness of map elements, and can also ensure matching accuracy.

[0065] Figure 2 This paper illustrates a specific implementation of a multi-path, multi-packet matching and fusion device based on loopback segmentation judgment according to this application. Figure 2 In the specific implementation shown, the multi-path multi-packet matching and fusion device based on loopback segmentation judgment mainly includes:

[0066] Module 201 is used to obtain multiple low-precision map packages in the same task, determine whether each low-precision map package contains backtracking and segment backtracking to obtain a low-precision single-package map, so that each low-precision single-package map does not contain overlapping data elements.

[0067] Module 202 is used to generate a sub-map with a fixed range in each low-precision single-package map, with randomly selected data elements as the center points.

[0068] Module 203 is used to simultaneously extract features and automatically match the two sub-maps corresponding to each center point in each of the two adjacent low-precision single-packet maps, obtain the matching scores corresponding to different data elements between the two adjacent sub-maps, and determine the object corresponding to the most certain match that meets geometric consistency through the matching scores.

[0069] Module 204 is used to obtain the objects corresponding to other geometrically consistent most certain matches that are adjacent to the most certain match, based on the most certain match.

[0070] Module 205 is used to align the objects that are most confidently matched among multiple low-precision single-packet maps one by one to obtain a high-precision single-packet map.

[0071] Module 206 is used to merge and optimize high-precision single-package maps to obtain high-precision maps.

[0072] In this implementation, by performing loop-based segmentation on the low-precision single-packet map containing backtracking paths, each object has only one corresponding data element in the low-precision single-packet map, facilitating subsequent matching. By splitting the low-precision single-packet map into multiple sub-maps for matching, the efficiency and accuracy are improved, and the high cost associated with high-precision equipment is resolved. Furthermore, the most certain matching method eliminates reliance on the uniqueness of map elements. This solution also saves computational resources.

[0073] In one specific embodiment of this application, determining whether a low-precision map packet contains backtracking and segmenting it to obtain a low-precision single-packet map includes: if the low-precision map contains backtracking, segmenting the low-precision map packet at the backtracking point to obtain a low-precision single-packet map that does not contain overlapping data elements, wherein the low-precision single-packet map includes a low-precision map packet that does not contain backtracking and a segmented low-precision map packet that contains backtracking.

[0074] In this embodiment, the same task contains multiple low-precision single-package maps. Some low-precision single-package maps contain backtracking paths, while others do not. In order to facilitate the matching of map elements, the low-precision single-package maps containing backtracking paths need to be segmented at the backtracking paths so that each object corresponds to one map element in a low-precision single-package map.

[0075] The multi-path multi-packet matching and fusion device based on loop closure segmentation judgment provided in this application can be used to execute the multi-path multi-packet matching and fusion method based on loop closure segmentation judgment described in any of the above embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0076] In one specific embodiment of this application, the functional modules of the multi-path multi-packet matching and fusion device based on loopback segmentation judgment can be directly in hardware, in software modules executed by a processor, or in a combination of both.

[0077] Software modules may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in this art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium.

[0078] The processor can be a Central Processing Unit (CPU), or 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 can be a microprocessor, but alternatively, it can be any conventional processor, controller, microcontroller, or state machine. The processor can 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 incorporating a DSP core, or any other such configuration. Alternatively, the storage medium can be integrated with the processor. The processor and storage medium can reside in an ASIC. The ASIC can reside in the user terminal. Alternatively, the processor and storage medium can reside as discrete components in the user terminal.

[0079] In another specific embodiment of this application, a computer-readable storage medium stores computer instructions that are operated to perform the multi-path multi-packet matching and fusion method based on loopback segmentation judgment in any embodiment.

[0080] In another specific embodiment of this application, a computer device includes a processor and a memory, the memory storing computer instructions that are operated to execute the multi-path multi-packet matching and fusion method based on loopback segmentation judgment in any embodiment.

[0081] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0082] The units described as separate components may or may not be physically separate. 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 the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0083] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A multi-path multi-packet matching and fusion method based on loop closure segmentation judgment, characterized in that, include: Multiple low-precision map packages from the same task are obtained. Each low-precision map package is then evaluated for backtracking, and the backtracking is segmented to obtain a single low-precision map package. This ensures that each single low-precision map package does not contain overlapping data elements. The evaluation process for determining whether each low-precision map package contains backtracking and segmenting the backtracking to obtain a single low-precision map package includes: If the low-precision map contains the backtrack, the low-precision map package is segmented at the backtrack to obtain the low-precision single-package map that does not contain overlapping data elements. The low-precision single-package map includes the low-precision map package that does not contain the backtrack and the segmented low-precision map package that contains the backtrack. In each of the low-precision single-package maps, a sub-map with a fixed range is generated with a randomly selected data element as the center point. Each of the low-precision single-package maps contains multiple sub-maps, and the multiple sub-maps overlap. For each of the center points in each of the two adjacent low-precision single-package maps, the two sub-maps corresponding to each center point are simultaneously subjected to feature extraction and automatic matching to obtain the matching score corresponding to different data elements between the two sub-maps. The object corresponding to the most certain match that meets geometric consistency is determined by the matching score. Based on the most certain match, obtain the objects corresponding to other most certain matches that are adjacent to the most certain match and conform to the geometric consistency. Align the objects corresponding to the most certain matches in the multiple low-precision single-packet maps one by one to obtain a high-precision single-packet map; The high-precision single-package maps are merged and optimized to obtain a high-precision map.

2. The multi-path multi-packet matching and fusion method based on loopback segmentation judgment as described in claim 1, characterized in that, In each of the two adjacent low-precision single-package maps, the two sub-maps corresponding to each center point are simultaneously subjected to feature extraction and automatic matching to obtain matching scores for different data elements between the two sub-maps, including: Simultaneously, feature extraction is performed on the different types of data elements corresponding to the two subgraphs to obtain the feature information corresponding to the two subgraphs respectively; The feature information corresponding to the two subgraphs is input into the attention map neural network for automatic matching to obtain a matching score.

3. The multi-path multi-packet matching and fusion method based on loop closure segmentation judgment as described in claim 2, characterized in that, The step of simultaneously extracting features from different types of data elements corresponding to the two subgraphs to obtain feature information corresponding to the two subgraphs also includes: In the same subgraph, feature extraction is performed based on the surrounding elements and the attributes of a data element to obtain corresponding semantic feature information and perceptual feature information. The semantic feature information and the perceptual feature information are superimposed to obtain the overall feature information corresponding to the data element. Feature extraction is performed based on the 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.

4. The multi-path multi-packet matching and fusion method based on loopback segmentation judgment as described in claim 2, characterized in that, The step of inputting the feature information corresponding to the two sub-graphs into the attention map neural network for automatic matching further includes: The attention graph neural network automatically learns the attention between a data element and other data elements in the same subgraph, as well as the attention between the same data element in two subgraphs.

5. The multi-path multi-packet matching and fusion method based on loopback segmentation judgment as described in claim 2, characterized in that, Before inputting the feature information corresponding to the two subgraphs into the attention map neural network for automatic matching, the method further includes: Construct pseudo-single-packet data with complete truth values; The attention map neural network is pre-trained using the pseudo-single packet data; The pre-trained attention graph neural network is trained using incomplete ground truth values ​​from the production line.

6. The multi-path multi-packet matching and fusion method based on loopback segmentation judgment as described in claim 1, characterized in that, The step of determining the object corresponding to the most certain match that meets geometric consistency through the matching score includes: When the matching score is consistent with the benchmark threshold, the data element corresponding to the matching score meets the geometric consistency. Furthermore, when the proportion of the number of matching scores that are consistent with the benchmark threshold is greater than the preset threshold, the data element corresponding to the matching score is the object that is most confidently matched.

7. A multi-path multi-packet matching and fusion device based on loopback segmentation judgment, characterized in that, include: A module for acquiring multiple low-precision map packages in the same task, determining whether each low-precision map package contains a backtrack and segmenting the backtrack to obtain a low-precision single-package map, such that each low-precision single-package map does not contain overlapping data elements, wherein determining whether each low-precision map package contains a backtrack and segmenting the backtrack to obtain a low-precision single-package map includes: if the low-precision map contains the backtrack, then segmenting the low-precision map package at the backtrack to obtain a low-precision single-package map that does not contain overlapping data elements, wherein the low-precision single-package map includes the low-precision map package that does not contain the backtrack and the segmented low-precision map package that contains the backtrack; A module for generating a corresponding fixed-range sub-map in each of the low-precision single-package maps, with randomly selected data elements as the center point, wherein each of the low-precision single-package maps has multiple sub-maps and the multiple sub-maps overlap. The module is used to simultaneously extract features and automatically match the two sub-maps corresponding to each center point in each of the two adjacent low-precision single-package maps, obtain the matching score corresponding to different data elements between the two adjacent sub-maps, and determine the object corresponding to the most certain match that meets geometric consistency through the matching score; A module for obtaining, based on the most certain match, other objects that conform to the geometric consistency and are adjacent to the most certain match; A module for aligning the objects corresponding to the most certain matches in multiple low-precision single-packet maps one by one to obtain a high-precision single-packet map; This module is used to fuse and optimize the high-precision single-package maps to obtain a high-precision map.

8. A computer-readable storage medium storing computer instructions, characterized in that, The computer instructions were manipulated The method is designed to perform the multi-path multi-packet matching and fusion method based on loop-loop segmentation judgment as described in any one of claims 1-6.

9. A computer device comprising a processor and a memory, the memory storing computer instructions, wherein the processor operates the computer instructions to perform the multi-path multi-packet matching and fusion method based on loopback segmentation judgment as described in any one of claims 1-6.

Images