Map incremental data checking method, device and equipment and storage medium

Abstract

The application provides a map incremental data verification method and device, equipment and a storage medium. After the map incremental data is obtained, incremental spatio-temporal index data is constructed based on the map incremental data. Then, the incremental spatio-temporal index data is input into a trained incremental verification model to obtain a prediction result of each incremental map element. The prediction result of each incremental map element is verified from multiple verification dimensions to obtain a verification result of each incremental map element. The multiple verification dimensions include a spatial dimension, an attribute dimension, a type dimension and a feature dimension. Based on the verification result indicating that the incremental map element passes the verification, the base map of the target area is updated. According to the embodiment of the application, the spatial, attribute, type and feature four-dimensional consistency verification is performed on each incremental map element, and the review verification accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of high-precision map technology, and more specifically, to a method, apparatus, device, and storage medium for verifying incremental map data. Background Technology

[0002] As core foundational data supporting vehicle environmental perception and decision-making, the accuracy and reliability of autonomous driving maps directly impact the safety and stability of autonomous driving systems. With the rapid development of autonomous driving technology, the accuracy requirements for map data are constantly increasing; even minor data deviations or errors can seriously affect vehicle driving safety. Therefore, autonomous driving maps must undergo rigorous review and verification before data release to ensure the accuracy, completeness, and consistency of the map data. How to efficiently and accurately review and verify map data has become an important research direction in the field of autonomous driving map technology. Summary of the Invention

[0003] In view of this, this application provides a method, apparatus, device and storage medium for verifying incremental map data, so as to at least solve the problems existing in the related technologies.

[0004] Specifically, this application is implemented through the following technical solution: This application provides a method for verifying incremental map data, including: Acquire incremental map data; the incremental map data includes multiple incremental map elements; Based on the incremental map data, incremental spatiotemporal index data is constructed. The incremental spatiotemporal index data includes global index data and spatiotemporal index data. The global index data includes the index identifier of the target area covered by the incremental map data. The spatiotemporal index data includes time index data and spatial index data. The time index data is used to record the timestamp of each incremental map element. The spatial index data includes the metadata, element identifier, and vector point chain of each incremental map element. The vector point chain is used to represent the shape of the incremental map element. The incremental spatiotemporal index data is input into the trained incremental verification model to obtain the prediction result for each incremental map element. The prediction result for each incremental map element is verified from multiple verification dimensions to obtain the verification result for each incremental map element; the multiple verification dimensions include spatial dimension, attribute dimension, type dimension and feature dimension. Based on the verification results indicating the incremental map elements that have passed verification, the base map of the target area is updated.

[0005] This application also provides a device for verifying incremental map data, including: The data acquisition module is used to acquire incremental map data; the incremental map data includes multiple incremental map elements. An index building module is used to build incremental spatiotemporal index data based on the incremental map data. The incremental spatiotemporal index data includes global index data and spatiotemporal index data. The global index data includes the index identifier of the target area covered by the incremental map data. The spatiotemporal index data includes time index data and spatial index data. The time index data is used to record the timestamp of each incremental map element. The spatial index data includes the metadata, element identifier, and vector point chain of each incremental map element. The vector point chain is used to represent the shape of the incremental map element. The model validation module is used to input the incremental spatiotemporal index data into the trained incremental validation model to obtain the prediction result for each incremental map element. The multi-dimensional verification module is used to verify the prediction result of each incremental map element from multiple verification dimensions to obtain the verification result of each incremental map element; the multiple verification dimensions include spatial dimension, attribute dimension, type dimension and feature dimension. The map update module is used to update the base map of the target area based on the verification results indicating the incremental map elements that have passed verification.

[0006] This application also provides a computer device, including 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 steps of the map incremental data verification method described in any of the foregoing embodiments.

[0007] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the map incremental data verification method described in any of the foregoing embodiments.

[0008] This application also provides a computer program product, including a computer program that, when run by a processor, performs the steps of any of the possible map incremental data verification methods described above.

[0009] The technical solutions provided by the embodiments of this application may include the following beneficial effects: In this embodiment, an automated image review chain from data input to decision output is formed based on a two-layer vectorized index of "incremental spatiotemporal index data", model reasoning, and four-dimensional verification (spatial dimension, attribute dimension, type dimension, and feature dimension). Compared with the technical solution of manual image review, it can improve the efficiency of image review and thus achieve "fast", "stable", and "accurate" multi-dimensional verification.

[0010] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this specification. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating an exemplary embodiment of a map incremental data verification method. Figure 2 This is a flowchart illustrating an exemplary embodiment of the present application for constructing an incremental spatiotemporal index; Figure 3 This is a flowchart illustrating a multidimensional verification method as shown in an exemplary embodiment of this application; Figure 4 This is a flowchart illustrating a map incremental data verification process in an exemplary embodiment of this application; Figure 5 This is a flowchart illustrating the training process of an incremental verification model according to an exemplary embodiment of this application; Figure 6 This is a schematic diagram of the structure of a map incremental data verification device shown in an exemplary embodiment of this application; Figure 7 This is a schematic diagram of the structure of another map incremental data verification device shown in an exemplary embodiment of this application; Figure 8 This is a hardware structure diagram of a computer device illustrated in an exemplary embodiment of this application. Detailed Implementation

[0012] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0013] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0014] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0015] As core foundational data supporting vehicle environmental perception and decision-making, the accuracy and reliability of autonomous driving maps directly impact the safety and stability of autonomous driving systems. With the rapid development of autonomous driving technology, the accuracy requirements for map data are constantly increasing; even minor data deviations or errors can seriously affect vehicle driving safety. Therefore, autonomous driving maps must undergo rigorous review and verification before data release to ensure the accuracy, completeness, and consistency of the map data. How to efficiently and accurately review and verify map data has become an important research direction in the field of autonomous driving map technology.

[0016] However, under the dual pressure of data explosion and high-frequency updates, the traditional "human experience-based" map review and verification faces the following pain points: (1) Difficulty in associating local elements: During the incremental review process, changes in geometry, attributes, functions, and relationships of incremental review elements within a local area will cause corresponding changes in data at other map levels; (2) High frequency of incremental data updates: Changes at the daily or even hourly level require minute-level map review; (3) Poor semantic consistency across multiple levels: Incremental updates at the road level, lane level, and feature level are prone to breakage, resulting in "upper-level updates and lower-level distortion".

[0017] Based on the above research, this disclosure provides a method for verifying incremental map data. The method first acquires incremental map data, which includes multiple incremental map elements. Next, based on the incremental map data, incremental spatiotemporal index data is constructed. This incremental spatiotemporal index data includes global index data and spatiotemporal index data. The global index data includes the index identifier of the target area covered by the incremental map data. The spatiotemporal index data includes time index data and spatial index data. The time index data records the timestamp of each incremental map element, and the spatial index data includes the metadata, element identifier, and vector point chain of each incremental map element. The vector point chain represents the shape of the incremental map element. Then, the incremental spatiotemporal index data is input into a trained incremental verification model to obtain a prediction result for each incremental map element. The prediction result for each incremental map element is verified from multiple verification dimensions to obtain a verification result for each incremental map element. These multiple verification dimensions include spatial dimension, attribute dimension, type dimension, and feature dimension. Finally, based on the verification results indicating the verified incremental map elements, the base map of the target area is updated.

[0018] In this embodiment, an automated image review chain from data input to decision output is formed based on a two-layer vectorized index of "incremental spatiotemporal index data", model reasoning, and four-dimensional verification (spatial dimension, attribute dimension, type dimension, and feature dimension). Compared with the technical solution of manual image review, it can improve the efficiency of image review and thus achieve "fast", "stable", and "accurate" multi-dimensional verification.

[0019] Furthermore, a two-layer vectorized index for "incremental spatiotemporal index data" is constructed. Based on the two-layer design of global index (regional positioning) and local index (element positioning) of incremental spatiotemporal index data, it can locate incremental elements in any region and any time window in milliseconds, thereby realizing millisecond-level dynamic indexing and second-level version switching of incremental data, which can improve the efficiency of data processing.

[0020] In addition, the prediction results for each incremental map element are verified from the spatial dimension, attribute dimension, type dimension and feature dimension respectively. That is, four dimensions are covered in parallel to cover all key quality attributes of the element, thereby improving the accuracy of verification.

[0021] To facilitate understanding of this embodiment, a detailed description of the map incremental data verification method disclosed in this disclosure is provided first. The execution entity of the map incremental data verification method provided in this disclosure is generally a computer device. This computer device can be a server, which can be an independent physical server, a server cluster composed of multiple physical servers, or a distributed system. It can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, big data, and artificial intelligence platforms. In other embodiments, the computer device can also be a terminal device, which can be a mobile device, terminal, handheld device, computing device, vehicle-mounted device, etc.

[0022] In other embodiments, the method can also be applied to an implementation environment consisting of computer equipment and servers, or an implementation environment consisting of terminal equipment and servers. Furthermore, the method for verifying incremental map data can also be implemented by a processor calling computer-readable instructions stored in memory.

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0024] Please see the appendix Figure 1 The above is a flowchart illustrating a method for verifying incremental map data, as shown in an exemplary embodiment of this application. Figure 1 As shown, the map incremental data verification method in this embodiment may include the following steps S101~S105: S101: Obtain incremental map data; the incremental map data includes multiple incremental map elements.

[0025] Here, incremental map data can refer to data collected by sensors installed on mobile devices or data collected by roadside equipment.

[0026] In some implementations, the mobile device may be a data collection vehicle, and the number of mobile devices may be one or more, without limitation; the sensor may include a camera and / or radar.

[0027] In other implementations, the mobile device can also be the user's vehicle, that is, the user's vehicle collects road information through sensors during normal driving to obtain incremental map data.

[0028] Incremental map data refers to map element data that has been added, modified, or deleted relative to the approved base map. For example, newly collected lane lines, traffic signs, or the discovery that a guardrail has been removed.

[0029] Multiple incremental map elements refer to the smallest independent unit (or map element instance) in incremental map data. For example, multiple incremental map elements include lane lines, signs, and guardrails. Each incremental map element carries element attribute information, which includes shape information, attribute information, confidence level, and time information (or timestamp). Among them, the attribute information includes sign type, color, length, reflectivity level, etc.

[0030] S102: Based on the incremental map data, construct incremental spatiotemporal index data; the incremental spatiotemporal index data includes global index data and spatiotemporal index data, the global index data includes the index identifier of the target area covered by the incremental map data, the spatiotemporal index data includes time index data and spatial index data, the time index data is used to record the timestamp of each incremental map element, the spatial index data includes the metadata, element identifier and vector point chain of each incremental map element, and the vector point chain is used to represent the shape of the incremental map element.

[0031] Here, incremental spatiotemporal index data refers to the index data structure established by structuring the original incremental map data, with spatial location and timestamp as the joint primary key, including global index (for quick location of areas) and local index (for precise location of each element).

[0032] Global index data refers to the top-level index of incremental spatiotemporal index data, which is used to quickly locate the geographical area covered by incremental map data (such as a rectangular area or a road segment) according to spatial range, avoiding the need to search the entire data and improving retrieval efficiency.

[0033] Spatiotemporal index data refers to the underlying index of incremental spatiotemporal index data. A fine-grained index is built for each incremental map element, including time index data (recording timestamps and supporting version switching and rollback) and spatial index data (recording metadata, element identifiers, and vector point chains), supporting millisecond-level element-level retrieval.

[0034] Metadata can refer to the basic attributes of an element, including the road level (such as national highway, provincial highway, county road, expressway, first-class highway, etc.), lane information (lane width and / or number of lanes, such as two-way four lanes, one-way two lanes), function type (such as arterial road, secondary arterial road, branch road, ramp, etc.), administrative division (geographic administrative region, such as city, district, street, etc.), review batch (including review time and data batch, such as 2025-06-20-batch-04), and over-limit status, among which the over-limit status can be obtained by performing over-limit detection on the element. Over-limit detection can include at least one of the following: size over-limit detection, position over-limit detection, curvature over-limit detection, and change amplitude over-limit detection.

[0035] Element identifiers can take any identifier type, such as numbers, letters, text, or one or more combinations thereof, which will not be elaborated here.

[0036] Vector point chains refer to the sequence of coordinate points obtained by "curvature adaptive sampling" of the original geometry of incremental map elements. That is, the sampling density is reduced on straight segments and full density is maintained at sharp bends, thereby reducing the amount of data while ensuring geometric accuracy.

[0037] In this step, the raw, unordered incremental map data is processed into structured index data. The "global index" is responsible for quickly locating regions, while the "local index" is responsible for precision down to each individual element. Each element carries complete information (metadata, unique identifier, point chain shape, timestamp), supporting second-level version switching. This index structure is the foundation for millisecond-level retrieval and version management.

[0038] S103: Input the incremental spatiotemporal index data into the trained incremental verification model to obtain the prediction result for each incremental map element.

[0039] The incremental verification model refers to a neural network model embedded with a verification rule graph. The trained incremental verification model is used to infer the input incremental spatiotemporal index data, output the prediction result of each incremental map element, compare the prediction result with the verification rule graph, calculate the verification loss, and determine whether the verification is passed.

[0040] The prediction results include predicted vector point chains, predicted element attribute information, predicted type probability distribution, predicted engineering design parameters, and predicted correlation relationships. Here, the predicted element attribute information includes the value combinations of the incremental map element in various attribute fields, such as sign type, sign shape, sign color, location type, reflectivity level, etc. The predicted vector point chain refers to the original coordinate sequence output by the incremental verification model, which is used to characterize the shape and position of the incremental map element.

[0041] S104: For the prediction result of each incremental map element, the verification results of each incremental map element are verified from multiple verification dimensions to obtain the verification result of each incremental map element; the multiple verification dimensions include spatial dimension, attribute dimension, type dimension and feature dimension.

[0042] Verification dimensions refer to the perspectives from which incremental map elements are judged for quality. This application verifies from the following four dimensions: spatial dimension (whether the location has drifted), attribute dimension (whether the attributes are complete and correct), type dimension (whether the type identification is reliable), and feature dimension (whether the geometric shape is distorted). This ensures the quality of incremental data.

[0043] S105: Based on the verification results indicating the incremental map elements that have passed verification, update the base map of the target area.

[0044] Here, for each incremental map element, if the verification passes, it is added to the base map of the target area to update the base map and obtain the updated base map.

[0045] In this embodiment, an automated image review chain from data input to decision output is formed based on a two-layer vectorized index of "incremental spatiotemporal index data", model reasoning, and four-dimensional verification (spatial dimension, attribute dimension, type dimension, and feature dimension). Compared with the technical solution of manual image review, it can improve the efficiency of image review and thus achieve "fast", "stable", and "accurate" multi-dimensional verification.

[0046] Furthermore, a two-layer vectorized index for "incremental spatiotemporal index data" is constructed. Based on the two-layer design of global index (regional positioning) and local index (element positioning) of incremental spatiotemporal index data, it can locate incremental elements in any region and any time window in milliseconds, thereby realizing millisecond-level dynamic indexing and second-level version switching of incremental data, which can improve the efficiency of data processing.

[0047] In addition, the prediction results for each incremental map element are verified from the spatial dimension, attribute dimension, type dimension and feature dimension respectively. That is, four dimensions are covered in parallel to cover all key quality attributes of the element, thereby improving the accuracy of verification.

[0048] Please see Figure 2 This is a flowchart illustrating the construction of an incremental spatiotemporal index, provided as an exemplary embodiment of this application. Figure 2 As shown, for step S102, when constructing incremental spatiotemporal index data based on the incremental map data, steps S1021~S1025 are included: S1021: For each incremental map element, the snowflake algorithm is used to determine the element identifier based on the road segment prefix of the road segment to which the incremental map element belongs.

[0049] Among them, the Snowflake Algorithm is a pre-defined distributed ID generation algorithm that can generate globally unique identifiers.

[0050] The road segment prefix refers to the code identifier of the road segment where the incremental map element is located, such as G01 (Beijing-Harbin Expressway) and R_042 (road segment code of a certain city).

[0051] S1022: Perform curvature sampling on the shape of the incremental map element to obtain the vector point chain.

[0052] Here, curvature sampling, also known as "curvature adaptive sampling," is an algorithm that dynamically adjusts the sampling density based on the curvature of the geometry. High-density sampling is maintained in areas of high curvature (sharp curves) to accurately recreate the curve shape; while sampling density is reduced in areas of low curvature (straight lines or gentle curves) to decrease the amount of data.

[0053] A vector point chain refers to a sequence of coordinate points obtained by curvature sampling of the original geometry of incremental map elements, such as... ,in, The point chains are arranged in the geometric direction of the element, which fully describes the geometric shape of the element.

[0054] S1023: Extract the metadata from the attribute information of the incremental map element; the metadata includes at least one of the following: level, lane, function type, administrative division, map approval batch, and over-limit calculation status.

[0055] Attribute information refers to the complete attribute data carried by incremental map elements, including metadata (structured labels used for indexing and management) and other detailed attributes (such as specific label types, shapes, colors, etc.).

[0056] Metadata refers to the core tag fields extracted from attribute information and used for indexing and management. Metadata is equivalent to the "identity tag" of a feature.

[0057] In this implementation, the metadata includes: grade (the technical grade of the road, such as expressway / first-class highway), lanes (number of lanes, such as two-way four-lane), function type (the functional classification of the road, such as arterial road / ramp), administrative division (the geographical administrative region where the element is located, such as Beijing / Chaoyang District), review batch (the review batch number to which the element belongs, such as 2025-07-12-Batch-056), and over-limit calculation status (whether the element has passed the over-limit calculation verification, such as pending calculation / calculated but not over-limit / calculated and over-limit).

[0058] Of course, the specific content included in the metadata fields can be selected and expanded according to the actual application scenario and map data specifications.

[0059] S1024: Based on the element identifier, vector point chain, and metadata of each incremental map element, determine the spatial index data, construct the time index data based on the time information, and determine the index data based on the spatial index data and the time index data.

[0060] It is easy to understand that after generating element identifiers, vector point chains, and metadata, this information can be further assembled into spatial index data, while time index data can be constructed based on the time information of the features, and then the two can be merged into complete index data.

[0061] Spatial index data supports rapid location by spatial range (e.g., "find all incremental features within a rectangular area"), while temporal index data supports rapid location by time version (e.g., "find incremental features collected after a certain time point"). The combination of these two forms a spatiotemporal joint index—capable of both spatial retrieval and time interval filtering, supporting millisecond-level spatiotemporal joint queries.

[0062] S1025: Generate the incremental spatiotemporal index data based on the index data of each incremental map element.

[0063] Following the preprocessing steps S1021-S1024, each incremental map element has a structured index record (element identifier + vector point chain + metadata + time index). In S1025, these records are organized into a global index according to spatial range, and the detailed index data of each feature is organized into a local index, forming a two-layer index structure: the global index is responsible for quickly locating "which area has changed", and the local index is responsible for accurately locating "which feature has changed".

[0064] The incremental spatiotemporal index data generated in this implementation method has the following beneficial effects: Feature-level refined management: Each feature has a globally unique ID, supporting refined indexing and independent traceability; Millisecond-level retrieval: A two-layer index (global + local) and a spatiotemporal joint primary key enable millisecond-level spatiotemporal joint queries; Second-level version switching: The time index supports one-click switching to any historical version, providing version management and rollback capabilities for map review; Standardized input for model inference: Structured index data eliminates the need for additional cleaning and formatting of the model, allowing it to focus directly on core verification tasks.

[0065] In some implementations, when constructing incremental spatiotemporal index data based on the incremental map data, state arbitration processing can be performed on each incremental map element, and corresponding processing operations can be performed on the incremental map elements according to the state arbitration results. The incremental spatiotemporal index data can be constructed based on each incremental map element and its corresponding state arbitration results.

[0066] Status arbitration processing includes new arbitration, invalid arbitration, or review arbitration.

[0067] Here, the state arbitration process is actually the implementation of the confidence and arbitration mechanism. Before the four-dimensional verification, the incremental map elements are pre-screened. Through three processes, namely addition arbitration, failure arbitration, and review arbitration, some highly certain elements are pre-processed (direct addition or marking failure) and conflict situations that require manual intervention are identified to avoid wasting resources in subsequent verification.

[0068] (a) New Arbitration The incremental map element is subjected to a new arbitration process to determine whether the base map contains a map element that is the same as the incremental map element, and the new arbitration result is obtained. If the new arbitration result indicates that the base map does not contain a map element that is the same as the incremental map element, the incremental map element is added to the base map.

[0069] The newly added arbitration logic verifies the following rule: "If the nearest distance between the new feature and the base map is greater than..." , "If the standard deviation of the location is constant and there are no inconsistencies in the three vectorization levels, then insert directly." The core idea is: if the incremental feature has no "matching item" in the base map (the spatial location, feature type, and attributes do not correspond), it means that it is a truly new feature and can be directly inserted into the base map; conversely, if the same feature already exists in the base map, it does not need to be added again.

[0070] The determination of "same map elements" here is usually based on spatial location (distance > 1). The arbitration process involves a comprehensive determination of both semantic consistency (no inconsistencies across all three vectorization levels). This arbitration is performed before the four-dimensional verification and is suitable for new scenarios with high determinism.

[0071] For example, an incremental feature is a new lane line whose location is more than [distance] from all existing lane lines in the base map. If there are no conflicts in the hierarchical information, the new arbitration result will be "new", and the lane line will be directly inserted into the base map. This prevents the same element from being collected and inserted repeatedly, thus maintaining the uniqueness and cleanliness of the map data.

[0072] (ii) Invalid Arbitration The system determines whether the incremental map element has been collected within a preset continuous time period and whether the confidence level of the incremental map element is less than a confidence level threshold, in order to perform failure arbitration on the incremental map element and obtain a failure arbitration result.

[0073] If the failure arbitration result indicates that the incremental map element has not been collected within a preset continuous time period and the confidence level of the incremental map element is less than the confidence level threshold, the index data of the incremental map element is marked as failed data.

[0074] The preset continuous time period and confidence threshold can be set according to actual needs. For example, the preset continuous time period can be 7 days, 15 days, etc., and the confidence threshold can be 0.6 or 0.7.

[0075] The logic of invalidation arbitration follows these rules: "When there are no repeated observations for 7 consecutive days and the original confidence level is..." "<0.6, marked as 'soft deletion', retains geometry but sets to invalid, supports one-click rollback." The core idea is: if a feature has not been collected for a long time and its confidence level is already low, it means that the feature has likely disappeared from the actual road and should be marked as invalid in the base map.

[0076] The aforementioned "soft deletion" process preserves geometric information but marks it as invalid. This means the feature data remains in the database, simply marked as "invalid," and is not used for map publishing. This design supports one-click rollback, allowing for quick recovery if the feature is re-acquired in the future.

[0077] For example, a speed limit sign exists in the base map, but has not been detected by any vehicle for 7 consecutive days, and the confidence level of the sign is... Since the confidence level is 0.4, and the failure arbitration result is "failure", the flag is marked as soft deletion.

[0078] The aforementioned failure arbitration process can automatically identify low-confidence elements that have not been observed for a long time, promptly detect road changes, and further, the soft deletion retains data recovery capabilities, avoiding the permanent loss of important elements due to accidental failure to collect data.

[0079] (iii) Review and Arbitration The incremental map element is determined to exist in different incremental map data, and the confidence level of the incremental map element in different incremental map data is determined to be the same, so as to perform a review and arbitration on the incremental map element and obtain the review and arbitration result.

[0080] It is understandable that review arbitration refers to the arbitration process when there are conflicting confidence levels among incremental data of the same element from different sources. Specifically, if the same element appears in both high-confidence and low-confidence versions in different map incremental data, it indicates that different sources have given contradictory judgments. In this case, it is necessary to further calculate the degree of conflict and decide whether manual intervention is required.

[0081] If the review and arbitration results indicate that the incremental map element exists in different map incremental data, and the confidence level of the incremental map element is different in different map incremental data, the conflict index between different map incremental data is determined.

[0082] The conflict index is an indicator that quantifies the degree of confidence difference between the same element and data from different sources. It reflects the "degree of disagreement" among different sources regarding the existence of the same element; the greater the difference between the high-confidence version and the low-confidence version, the higher the conflict index. The conflict index can be calculated based on a three-dimensional confidence matrix of "sensor-map quotient-time," combining the confidence and confidence differences of observation data from multiple sources to comprehensively assess the degree of conflict.

[0083] If the conflict index is greater than the conflict threshold, a review prompt message is generated; the review prompt message is used to indicate that the incremental map element should be manually reviewed.

[0084] The review prompt information refers to the prompt information generated when manual review is triggered. It includes feature identifier, conflict details (high confidence version source and low confidence version source), conflict index, etc., and is used to guide manual reviewers in conducting the review.

[0085] Based on the above, the logic of the review arbitration conforms to the following rule: "If there are two versions of the same element with high confidence and low confidence, the 'evidence weight network' will be automatically fused. If the conflict index is >0.3, it will enter the manual review queue."

[0086] The review and arbitration system can automatically detect conflicting judgments on the same element from different sources, preventing low-quality data from polluting the map and avoiding the overwhelming of highly reliable data. Furthermore, when conflicts are severe, human map reviewers make the final decision, preventing the system from making incorrect judgments under uncertain circumstances.

[0087] It should be noted that for each incremental map element, if the arbitration result indicates that the element is newly added, then the element will undergo the subsequent prediction and multi-dimensional verification process. If the arbitration result indicates that the element is invalid data, then the element will not enter the subsequent prediction and multi-dimensional verification process. Similarly, if the arbitration result indicates that the element needs to be verified, then the element will not enter the subsequent prediction and multi-dimensional verification process.

[0088] Please see Figure 3 This is a flowchart of a multi-dimensional verification method provided as an exemplary embodiment of this application. For example... Figure 3 As shown, for step S104, when verifying the prediction result of each incremental map element from multiple verification dimensions to obtain the verification result of each incremental map element, the following steps S1041~S1045 are included: S1041: For each incremental map element, the predicted vector point chain of the incremental map element is compared with the vector point chain of the corresponding map element in the base map to verify the position from the spatial dimension and obtain the spatial position verification result.

[0089] As mentioned earlier, the predicted vector point chain can represent the spatial position of incremental map elements. Therefore, by comparing the predicted vector point chain of the incremental map element with the vector point chain of the corresponding map element in the base map, it can be determined whether the incremental map element has undergone an unacceptable positional "drift" relative to the base map.

[0090] Specifically, the predicted vector point chain includes multiple predicted vector points, and a corresponding map element in the base map corresponds to a vector point chain. The vector point chain includes multiple vector points. Therefore, for step S1041, the following steps (1) to (3) may be included: (1) Register multiple predicted vector points with multiple vector points to obtain multiple vector point pairs, and determine the positional differences of each vector point pair.

[0091] In this step, since the predicted vector point chain and the vector point chain may be in different coordinate systems or have acquisition posture deviations, it is necessary to transform the predicted point chain to the base map coordinate system through rotation matrix and translation vector so that the two can be spatially coincident. That is, registration means spatially aligning the predicted vector point chain and the vector point chain.

[0092] After registration, each predicted vector point has a nearest corresponding vector point in its corresponding vector point chain, forming a "vector point pair". Each vector point pair corresponds to a positional difference, which represents the deviation of the predicted vector point from its corresponding vector point in the base map.

[0093] (2) Based on the positional differences of each vector point pair, determine the confidence range, and based on the confidence range, determine the allowable fluctuation range of the true position of the incremental map element.

[0094] The confidence range refers to the statistical area determined based on the positional differences of all vector point pairs, used to characterize the range of uncertainty in the true position of incremental map elements.

[0095] Specifically, this can be achieved by determining the covariance matrix of location differences and generating a 95% confidence ellipse.

[0096] (3) Determine the distance between the boundary of the allowed fluctuation range and the boundary of the corresponding map element in the base map, and determine the spatial location verification result based on the distance.

[0097] The statistically obtained allowable fluctuation range is compared with the boundary of the base map to determine whether the position of the incremental element has undergone unacceptable drift.

[0098] Specifically, determine the shortest distance from the boundary of the confidence ellipse to the nearest lane boundary in the base map. .if big enough This indicates that even considering location uncertainty (95% confidence ellipse), the incremental element still significantly deviates from its intended position in the base map, and is thus classified as "drift." Within acceptable limits ( This indicates that the positional deviation is statistically reasonable, and the spatial dimension verification is passed.

[0099] In this implementation, by distance and The statistical comparison upgrades the determination of "whether it is drifting" from "subjective judgment by the naked eye" to "objective quantitative judgment", which is not affected by the subjective experience of the reviewers and the judgment standard is unified and reproducible.

[0100] S1042: Obtain the attribute trie and compare the predicted element attribute information of the incremental map element with the attribute trie according to the semantic hierarchy to verify from the attribute dimension and obtain the attribute verification result.

[0101] The attribute trie is a semantic hierarchy used to verify the integrity and validity of attributes of incremental map elements. It is divided into four levels according to semantic granularity, from coarse to fine: element level L1 (e.g., traffic signs), element subclass level L2 (e.g., prohibition signs), attribute field level L3 (e.g., sign color, sign shape), and candidate value range level L4 (e.g., red, circle). Each element type has a predefined set of standard attribute fields and a candidate value range for each field in the trie.

[0102] It is understandable that the attribute information of incremental map elements is key data describing the characteristics of the elements (such as the type, shape, color, and speed limit of traffic signs). If the predicted element's attribute information is missing (e.g., a speed limit sign has no speed limit), is illegal (e.g., the number of lanes = 7 is outside the normal range), or is conflicting (e.g., the same sign is identified as "speed limit 60" and "speed limit 80"), the element cannot be used correctly.

[0103] Automated comparison is achieved through an attribute trie: First, based on the element type (e.g., "prohibition sign"), a predefined set of standard attributes (e.g., "must have sign type, shape, color, position type, reflectivity level") is found in the trie. Then, incremental elements are checked field by field to see if they possess these attributes and whether their values ​​are within the allowed range. Any missing or conflicting fields are marked as an anomaly.

[0104] In this step, the automatic comparison of L1→L2→L3→L4 ensures that the attribute verification covers the entire semantic level from "element category" to "specific value", without missing any detections or making any misjudgments.

[0105] In addition, the attribute trie provides a unified attribute standard reference for all incremental elements, eliminating subjective differences from manual review.

[0106] Specifically, step S1042 includes the following steps (1) to (3): (1) Based on the prediction type of the incremental map element, traverse the attribute dictionary tree to determine the target element subclass corresponding to the prediction type, the set of attribute fields corresponding to the target element subclass, and the candidate value range corresponding to each attribute field.

[0107] The prediction type of an incremental map element refers to its element type, such as "highway" or "lane line," to determine which branch in the attribute trie to start the search from.

[0108] The target element subclass refers to the L2 node in the attribute trie that corresponds to the "element subclass" of the incremental map element. For example, "prohibition sign" corresponds to the "prohibition sign" node in L2, and "highway" corresponds to the "highway" node in L2. Each target element subclass contains all the attribute fields (L3) that the feature of that type should have, as well as the legal values ​​(L4) for each field.

[0109] For example, assuming the element type of the incremental map element is "prohibition sign": traverse the attribute trie, locate L1 "traffic sign" → L2 "prohibition sign"; obtain the standard attribute field set: {sign type, sign shape, sign color, location type, reflectivity level, speed limit}; candidate value range: sign type: {prohibition}, sign shape: {circle}, sign color: {red}, location type: {intersection, road segment}, reflectivity level: {Ⅰ,Ⅱ,Ⅲ,Ⅳ}; speed limit: {5,10,15,20,25,30,40,50,60,70,80,90,100,110,120}.

[0110] (2) For each predicted attribute field, the predicted attribute field of the incremental map element is compared with the attribute field set field by field, and the mask state of the predicted attribute field is determined according to the value of the predicted attribute field and the value range of the corresponding attribute field; the mask state includes valid state, missing state and conflict state.

[0111] Predicted attribute fields refer to the specific attribute items and values ​​output by the incremental validation model after inferring the attributes of incremental map elements. For example, the predicted attribute fields of a prohibition sign include: sign type = prohibition, sign shape = circle, sign color = red, and speed limit = 80 (km / h).

[0112] Furthermore, each predicted attribute field of the incremental element is compared with the standard attribute field set obtained in (1) one by one to check three issues: ① whether the field is in the standard set (whether it exists); ② whether the value of the field is within the candidate value range (whether it is legal); ③ whether there are multiple inconsistent values ​​(whether there is a conflict).

[0113] After the comparison, each predicted attribute field is assigned a status label to characterize its compliance. There are three types: valid (the field exists and its value is within the candidate value range), missing (the standard set requires the field to exist, but it is not provided in the incremental features), and conflicting (the field exists but its value is not within the candidate value range, or the same field has multiple inconsistent values). The mask status is the basis for calculating attribute consistency loss.

[0114] In some implementations, if the proportion of predicted attribute fields with missing states or conflicting states exceeds a preset threshold, the attribute verification result indicates failure.

[0115] (3) Determine the attribute verification result based on the mask status of each predicted attribute field.

[0116] Here, after obtaining the mask status of each predicted attribute field, the mask status (valid / missing / conflicting) of all predicted attribute fields is statistically analyzed to obtain the number or weighted sum of each status, which serves as the quantitative basis for attribute verification.

[0117] The attribute validation result refers to the final judgment of the attribute dimension validation, including whether the element passes the attribute validation, a list of abnormal fields (which fields are missing / conflicting), and the attribute consistency loss value. The specific judgment method is as follows: based on the mask status of each field and the preset weight of each field (different levels of importance), the attribute consistency loss is calculated. If the total loss is not greater than the preset threshold, the attribute validation passes; if the total loss is greater than the threshold, the attribute validation fails.

[0118] In this implementation, the attribute dimension verification process, through a three-step process of "locating standards, comparing field by field, and weighted judgment", realizes the automated and refined verification of map element attribute information. The verification result can accurately locate the specific abnormal field (such as "the color of the prohibition sign is green, it should be red"), rather than vaguely saying "the attribute is unqualified", which greatly reduces the cost of manual correction.

[0119] S1043: Compare the predicted type probability distribution of the incremental map elements with the preset type confidence threshold to verify from the type dimension and determine the type verification result.

[0120] The predicted type probability distribution is the probability distribution vector output by the incremental verification model after inferring the type of incremental map elements. Its dimension is equal to the total number of preset map element types (such as lane lines, traffic signs, guardrails, poles, etc.). Each element in the predicted type probability distribution represents the probability that the incremental map element belongs to the corresponding type, and the sum of all elements is 1.

[0121] The type confidence threshold is a preset probability threshold used to determine whether the type identification result is reliable, and can be set according to actual needs. For example, the type confidence threshold can be 0.8. When the probability of any element in the predicted type probability distribution is not less than 0.8, the type identification result is reliable; when any probability in the predicted type probability distribution is less than 0.8, it is determined to be a "low-confidence type", and the incremental map element is temporarily suspended from being added to the database.

[0122] In this step, if the confidence level of type recognition is insufficient, it will be automatically intercepted before incremental map elements are added to the database to avoid map semantic errors caused by misclassification.

[0123] S1044: Based on the predicted vector point chain corresponding to the incremental map element, different shape distortion detection methods are used to perform shape detection on the incremental map element in order to verify it from the feature dimension and obtain the shape verification result.

[0124] It should be noted that the vector point chain is a sequence of coordinate points obtained by "curvature adaptive sampling" of the original geometry of incremental map elements. That is, the sampling density is reduced on straight sections and full density is maintained at sharp bends to balance geometric accuracy and data volume.

[0125] Here, different shape distortion detection methods include, but are not limited to, the following three detection methods: self-intersection detection (scanning line + cross-test to check whether non-adjacent line segments intersect), back-turn point detection (accumulated detection of direction angle change to check whether the direction of the broken line reverses by more than 135°), and small tooth-like fluctuation detection (joint detection of curvature change frequency and amplitude to detect high-frequency small-amplitude sawtooth fluctuations).

[0126] It is understandable that the predicted vector point chains of incremental map elements may exhibit geometric distortions during the acquisition and processing process due to sensor noise, data processing errors, etc., such as self-intersection (point chains intersecting themselves), backtracking (sudden reversal of direction), and small jagged fluctuations (high-frequency sawtooth-like jitter). These distortions can prevent map elements from being used correctly (e.g., autonomous vehicles cannot correctly identify lane line shapes).

[0127] The shape verification result can be determined based on the detection results corresponding to different shape distortion detection methods. Specifically, the shape verification result is determined based on the self-intersection detection result, the back-point detection result, and the small tooth-shaped fluctuation detection result, as well as their respective weights.

[0128] Furthermore, if the shape verification result is greater than the preset threshold, it is marked as "geometric deformity"; if the self-intersection detection result indicates the existence of self-intersection, it is directly judged as "geometric deformity" (the degradation index is forcibly set to 1.0).

[0129] This step upgrades the "visual judgment of geometric deformities" to "automatic quantitative judgment by algorithm" by performing three types of shape distortion detection on the predicted vector point chain and calculating a weighted comprehensive degradation index. Furthermore, through self-intersection, backtracking, and small tooth-like undulation detection, it achieves refined geometric quality control and comprehensively covers all types of geometric deformities that may appear in the vector point chain. Once an incremental map element is determined to be geometrically deformed, it is intercepted to prevent erroneous geometric data from entering the map database.

[0130] S1045: Determine the verification result based on the spatial location verification result, the attribute verification result, the type verification result, and the shape verification result.

[0131] It is understandable that if the spatial location verification result, attribute verification result, type verification result, and shape verification result all indicate that the verification has passed, then the verification result is "verification passed"; if the verification of at least one dimension fails, then the verification result is "verification failed".

[0132] Optionally, in step S105, when updating the base map of the target area based on the incremental map elements that have passed verification according to the verification result, multi-source map incremental data corresponding to the incremental map elements can be obtained for the incremental map elements that have passed verification according to the verification result. Based on the multi-source map incremental data, the credibility of the incremental map elements can be determined. If the credibility is greater than the credibility threshold, the base map can be updated based on the incremental map elements that have passed verification according to the verification result.

[0133] The multi-source map incremental data refers to incremental map data collected at different times by different acquisition devices.

[0134] This is an additional verification step after the 4D verification passes but before the final update. Logically, this step is a trigger condition for the observation reliability constraint: that is, only after the 4D verification passes will the reliability judgment of "whether the element meets the update conditions" be triggered. In other words, the final update must meet two conditions simultaneously: the 4D verification passes and the observation reliability meets the standard. If the 4D verification fails, it is directly suspended and the observation reliability verification is not triggered again; if the 4D verification passes but the observation reliability is insufficient, the update is also suspended to exclude unreliable data from contaminating the map, such as data acquisition equipment failure, single abnormal observations, or temporary road changes (such as temporary road diversions for construction).

[0135] Specifically, when determining the credibility of the incremental map elements based on the multi-source map incremental data, steps (A) to (C) are included.

[0136] (A) Determine the credibility score of each map incremental data based on the source information of each map incremental data in the multi-source map incremental data.

[0137] The source information refers to the source attributes of each observation data, including: sensor type (LiDAR / camera / radar / GNSS / IMU), data source type (professional surveying vehicle / crowdsourced fleet / roadside equipment), and collection time (current hour / today / this week / this month).

[0138] The credibility score is a quantitative assessment of the credibility of a single observation. The value ranges from 0 to 1, with values ​​closer to 1 indicating a more credible observation. The credibility score is calculated using a "three-dimensional credibility matrix" (sensor type × data source type × acquisition time window): sensor accuracy coefficient × data source quality coefficient × time decay coefficient. For example, if an observation comes from "LiDAR (0.95) + professional surveying (0.95) + current hour (1.0)", the credibility score is 0.95 × 0.95 × 1.0 = 0.9025; if it comes from "camera (0.85) + crowdsourcing (0.80) + this week (0.85)", the score is 0.85 × 0.80 × 0.85 = 0.578.

[0139] (B) Based on the confidence scores corresponding to each map incremental data, determine the weight of each map incremental data, and perform Bayesian voting fusion on the multi-source map incremental data based on the weight of each map incremental data to obtain fusion estimation data and fusion confidence.

[0140] Here, weight refers to the importance coefficient of each observation in the fusion. The weight is positively correlated with the confidence score, that is, the higher the confidence score, the greater the weight assigned to the observation in the fusion.

[0141] Bayesian voting fusion treats each observation as an estimate with uncertainty (Gaussian distribution), calculates the optimal estimate by weighted averaging, and the fusion result comprehensively considers all observations and their credibility. Observations with higher credibility have a greater impact on the final result.

[0142] Fusion estimation data refers to the optimal estimates of incremental map elements obtained after Bayesian voting fusion, including fused geometric coordinates, attribute fields, etc. Fusion confidence is a quantitative measure of the reliability of fused estimation data, determined by the variance after fusion. Fusion confidence is negatively correlated with variance; the smaller the variance after fusion, the more consistent the observations and the more reliable the fusion result; conversely, the larger the variance, the greater the discrepancy among the observations and the higher the uncertainty of the fusion result.

[0143] For example, the positions of a certain element in three observations are (0,0), (0.05,0.05), and (-0.03,0.02), with weights of 0.8, 0.2, and 0.1, respectively. After fusion, the position is approximately (0.01,0.01), and the fusion confidence is high (small variances).

[0144] (C) Determine the consistency loss based on the difference between each map incremental data and the fusion estimated data; determine the confidence of the incremental map element based on the fusion confidence and the consistency loss.

[0145] In this step, for each observation, the deviation between it and the fused estimated data is calculated, i.e., "the degree to which a particular observation deviates from the consensus." If an observation is highly consistent with the fusion result, the difference is small; if an observation is an "outlier" (e.g., due to severe deviation caused by sensor malfunction), the difference will be large. The larger the difference, the less consistent the observation is with the consensus. It can be understood that the consistency loss is the total amount obtained after weighted statistical analysis of the deviations of all observations, reflecting the overall consistency among multi-source observation data. The smaller the consistency loss, the more consistent and mutually corroborative the observations are; the larger the consistency loss, the greater the divergence and the existence of contradictory information among the observations. In this step, fusion confidence and consistency loss reflect the quality of multi-source observations from two different perspectives: fusion confidence reflects the statistical reliability of the fusion result, and consistency loss reflects the degree of consensus among multi-source observations. Both dimensions must be satisfied simultaneously to be considered highly reliable. An element is considered reliable only when multi-source observations are both numerically consistent (small consistency loss) and statistically reliable (high fusion confidence). If the fusion confidence is low (large discrepancies among observations) or the consistency loss is high (significant outliers exist), the reliability is insufficient, and the element should be temporarily suspended from updating to prevent false positives from polluting the map. Finally, the reliability is compared with a preset threshold to determine whether the update conditions are met.

[0146] For example, if 10 observations of a certain element show high consistency, with a fusion confidence level of 0.95 and a consistency loss of 0.02, the confidence level is high, and the update is confirmed. If 10 observations of a certain element show significant divergence, with a fusion confidence level of 0.40 and a consistency loss of 0.35, the confidence level is insufficient, the update is suspended, and the case is transferred to manual review.

[0147] In this implementation, the dual verification of confidence level (statistical reliability) and consistency loss (degree of multi-source consensus) is more reliable than a single indicator, ensuring that only truly credible elements can be updated.

[0148] Please see Figure 4 This is a flowchart illustrating a map incremental data verification process, provided as an exemplary embodiment of this application. Figure 4As shown, firstly, incremental map data is acquired, and incremental spatiotemporal index data is constructed. During the construction process, state arbitration is performed (including new arbitration, failure arbitration, and review arbitration). The incremental spatiotemporal index data is input into the incremental verification model to obtain the prediction results. Then, the prediction results are verified from multiple verification dimensions (including spatial dimension, attribute dimension, type dimension, and feature dimension) to obtain the verification results. If the verification result indicates that the verification is passed, the base map is updated.

[0149] In this embodiment of the application, a cloud-edge-device collaborative architecture is adopted to complete the update and release of the basic map. Specifically, the process includes the following steps: 1. Data Upload: Vehicles collect road data and upload it to the cloud anonymously via 5G / V2X networks.

[0150] 2. Intelligent image review: The cloud combines multi-source data from the vehicle and the roadside, and uses trained models and rules to perform automated image review and verification.

[0151] 3. Result storage: The verification results of incremental data that have passed the drawing review are stored on the cloud using blockchain.

[0152] 4. Map Update: The map library model is iteratively upgraded in the cloud based on the incremental data.

[0153] 5. Rapid distribution: Updated map data is pushed to edge nodes for caching.

[0154] 6. Service delivery: Edge nodes distribute the latest map data to the vehicle with a latency of minutes, completing the entire update loop.

[0155] This "cloud-edge-device" collaborative architecture allows computationally intensive tasks (model training, global mapping) to be placed in the cloud, low-latency tasks (data distribution) to be moved to the edge, and data collection to be performed on the vehicle, achieving optimal resource allocation. The presence of edge nodes ensures the real-time nature of map data distribution, while the cloud guarantees the depth and breadth of data processing and model iteration. The blockchain-based evidence storage mechanism provides robust security and auditing guarantees for the entire map review process.

[0156] Please see Figure 5 The above is a flowchart illustrating the training process of an incremental verification model, provided as an exemplary embodiment of this application. Figure 5 As shown, steps S501 to S506 are included: S501: Obtain a training sample set; the training sample set includes historical map incremental data samples and corresponding labels; the labels refer to the compliance annotation information of the historical map incremental data samples.

[0157] Here, historical map incremental data samples refer to historical incremental data that has completed the map review process and has a clear conclusion. This data may include various scenarios (new features, invalid features, conflicting features, normal update features, etc.), covering various situations that the model may encounter in practical applications.

[0158] Labels (compliance labeling information) refer to information that, after confirmation by experts or existing reliable review processes, indicates whether a sample is compliant, which dimensions are qualified, and which dimensions are unqualified.

[0159] The specific content of the label includes: whether the element has passed the drawing review (pass / fail), and the compliance status of each verification dimension (such as whether the spatial location is compliant, whether the attributes are complete, whether the type is correct, whether the shape is qualified, etc.).

[0160] S502: Input the historical map incremental data sample into the initial incremental verification model to obtain the prediction output result of each incremental map element sample in the historical map incremental data sample.

[0161] Here, the predicted output refers to the multi-dimensional estimate output by the model after performing forward inference on the input incremental map elements, including the first predicted element attribute information, the first predicted engineering design parameters, and the first predicted correlation between the samples of each incremental map element.

[0162] S503: Determine the data-driven loss based on the difference between the predicted output of each incremental map element sample and the corresponding compliance labeling information.

[0163] Data-driven loss is a loss term used to measure the difference between the model's prediction and the label (compliance annotation information). The magnitude of the data-driven loss directly reflects the model's prediction accuracy; the smaller the difference, the closer the model's prediction is to the standard answer.

[0164] S504: Determine the rule-driven loss based on the degree of violation of the predicted output of each incremental map element sample relative to the preset map design rules.

[0165] Here, the preset map design rules refer to a set of predefined map element design specifications, including: ① topological constraints for constraining the relationships between elements (expert rules such as lane line head-to-tail connections and pole sign binding); ② engineering design specifications for constraining engineering design parameters (physical design specifications such as curvature-speed limit constraints, minimum radius of ramps, and lane width); ③ attribute semantic rules for constraining the semantic rationality of attribute field combinations (co-occurrence probability of attribute field combinations, judged by a pre-trained language model).

[0166] Based on the above, it can be seen that rule-driven loss is used to measure the degree to which the model's prediction results violate the preset map design rules.

[0167] Specifically, step S504 may include the following steps (1) to (4): (1) Using the topology graph constraints, the first predicted correlation between each incremental map element sample is verified. If there is a correlation that violates the topology graph constraints, the topology graph constraint loss is determined according to the degree of violation.

[0168] Topological constraints refer to the normative constraints on the spatial connectivity between map features. These are achieved by encoding expert rules such as "connecting lane lines 'head-to-tail,' binding signs 'pole-to-sign,' and ensuring guardrails are continuous and do not jump around" into differentiable graph constraints. Specifically, these include: lane line head-to-tail connections (the end of one lane line must connect to the start of another lane line), binding signs to poles (signs must be attached to their corresponding poles), and continuous guardrail segments (the direction of adjacent guardrail segments must be smooth and continuous, without abrupt changes).

[0169] For example, the first predicted association could be a prediction that the end point of lane line A is connected to the start point of lane line B, or a prediction that a sign is attached to a pole. These predicted associations are the model's "guess" about the topological structure between features.

[0170] The degree of violation refers to the extent to which the first predicted association deviates from the topology constraints, quantified by calculating the difference between the actual deviation metric and a preset geometric tolerance threshold. For example, if the Euclidean distance between the end point of lane line A and the start point of lane line B is 0.5m, the tolerance threshold is 0.1m, the excess is 0.4m, and the degree of violation is 0.4m. The greater the degree of violation, the higher the loss of topology constraints.

[0171] The topology graph constraint loss is the sum of the violation penalties for all first-predicted associations. Each association is penalized individually using a pre-defined penalty function (such as a squared distance or an exponential function) based on its degree of violation (the difference between the actual deviation metric and the geometric tolerance threshold). These penalties are then aggregated to form the topology graph constraint loss.

[0172] Specifically, determining the topology graph constraint loss includes the following steps (A) to (D): (A) For each first predicted association, determine the association type of the first predicted association; the association type includes the head-to-tail connection relationship between lane lines, the binding relationship between signs and poles, or the continuous direction relationship between guardrail segments.

[0173] It is understandable that different types of connections have completely different spatial measurement methods and tolerance thresholds (lane lines are measured by endpoint distance, signs by spatial distance, and guardrails by the rate of change of direction angle), so the association type should be determined first.

[0174] (B) Determine the corresponding spatial measurement method and preset deviation threshold according to the association type.

[0175] The three types of associations each correspond to different spatial measurement methods: ① Lane line head-to-tail connection - calculate the Euclidean distance (straight-line distance between two points) from the end of the first lane line to the beginning of the second lane line; ② Sign and pole binding - calculate the spatial distance (three-dimensional Euclidean distance) from the geometric center of the sign to the geometric center of the pole; ③ Continuous direction of guardrail segments - calculate the rate of change of the direction angle between adjacent guardrail segments (the change in angle per unit length).

[0176] The preset deviation threshold refers to the maximum allowable deviation value for various relationships, i.e., the geometric tolerance threshold. ① Lane line head-to-tail connection: tolerance threshold 0.1m (endpoint spacing shall not exceed 0.1m); ② Sign and pole binding: tolerance threshold 0.5m (center-to-center spacing shall not exceed 0.5m); ③ Continuous guardrail section: tolerance threshold 30° / m (direction angle change rate shall not exceed 30° / m).

[0177] (C) Using the spatial measurement method, determine the actual deviation metric value of the first predicted association, and based on the actual deviation metric value and the preset deviation threshold, determine the violation penalty item of the first predicted association.

[0178] Here, the actual deviation metric refers to the actual deviation value of the predicted correlation calculated using the corresponding spatial metric method. For example, the Euclidean distance between the end point of lane line A and the start point of lane line B is 0.5m (corresponding to a 0.1m threshold); the spatial distance between the center point of the sign and the pole is 0.8m (corresponding to a 0.5m threshold); and the rate of change of the orientation angle of adjacent guardrail segments is 45° / m (corresponding to a 30° / m threshold).

[0179] (D) Determine the topology graph constraint loss based on the violation penalty term of each first predicted association.

[0180] Here, the penalty terms for violations of all first-predicted associations are summed to obtain the topology graph constraint loss. The larger this loss value, the more severe the overall violation of topology rules the model predicts in the inter-feature connections.

[0181] In this implementation, the three types of association cover the main spatial connections between map features, ensuring the comprehensiveness of topology verification.

[0182] (2) Using the design specifications, the first predicted engineering design parameters of the incremental map element sample are verified. If the first predicted engineering design parameters exceed the preset specification range, the engineering design loss is determined according to the degree of exceedance.

[0183] Here, design specifications refer to the physical and normative requirements for the engineering design parameters of map elements. Essentially, design specifications transform road engineering design standards (such as the physical relationship between curvature and speed limits, and the relationship between the minimum radius of ramps and design speed) into calculable mathematical constraint expressions. Specifically, these include: curvature-speed limit constraints (the greater the curvature, the lower the speed limit), minimum ramp radius constraints (the higher the design speed, the larger the minimum radius requirement), and lane width constraints (different road grades correspond to different lane width standards).

[0184] The first predicted engineering design parameters refer to the predicted values ​​output by the model after inferring the engineering design parameters of incremental map elements in the training samples. Specifically, these include: predicted curvature, predicted speed limit, predicted minimum ramp radius, predicted lane width, and predicted road grade.

[0185] Preset specification ranges refer to the allowable value ranges of parameters determined by the analytical expressions of design specifications, including speed limit specification ranges, minimum radius specification lower limits for ramps, lane width specification ranges, etc. For example, curvature... The corresponding speed limit range is .

[0186] Engineering design loss refers to the sum of the values ​​of the exponential penalty function of all predicted engineering design parameters.

[0187] Specifically, when determining engineering design losses, the following steps (2.1) to (2.4) can be used: (2.1) Obtain the lane state of the lane, and use the constraint relationship between curvature and speed limit to determine the standard speed range based on the lane state and the predicted curvature, and determine the speed penalty value based on the predicted speed limit and the standard speed range.

[0188] Lane conditions refer to a set of parameters describing the physical conditions of a lane, including the lateral friction coefficient and the road cross slope. The lateral friction coefficient reflects the coefficient of adhesion between the road surface and the tires (0.15 for dry asphalt and 0.08 for wet surfaces); the road cross slope reflects the lateral inclination of the road surface (0.02 for general road sections, -0.03 for the inside of curves, and 0.03 for the outside). These parameters directly affect the physical constraint relationship between curvature and speed limits: the higher the friction coefficient, the greater the lateral force a vehicle can withstand in a curve, and the corresponding safe speed limit is also increased.

[0189] The constraint relationship between curvature and speed limit is shown in formula (1).

[0190] (1) in, This is the upper limit of the speed limit. The coefficient of lateral friction is... For the cross slope of the road, To predict curvature.

[0191] The standard speed range can be determined based on formulas (2) to (3): (2) (3) Furthermore, if the predicted speed limit is outside the standard speed range, the speed penalty value can be determined based on the difference between the predicted speed limit and the standard speed range.

[0192] (2.2) Determine the radius penalty value based on the predicted minimum radius of the ramp and the minimum radius threshold of the ramp.

[0193] The minimum radius threshold of the ramp satisfies formula (4): (4) in, The higher the design speed of a ramp, the larger the minimum radius required.

[0194] (2.3) Determine the lane width range threshold corresponding to the predicted road level of the lane, and determine the lane width penalty value based on the predicted lane width and the lane width range threshold.

[0195] In this implementation, roads are classified into five levels: expressway = 1, Class I highway = 2, Class II highway = 3, Class III highway = 4, and Class IV highway = 5. Different levels correspond to different lane width standards.

[0196] The lane width range thresholds can be: expressway [3.75 meters, 3.75 meters] (fixed value), Class I highway [3.75 meters, 3.75 meters], Class II highway [3.50 meters, 3.75 meters], Class III highway [3.25 meters, 3.50 meters], and Class IV highway [3.00 meters, 3.25 meters].

[0197] Furthermore, based on the difference between the predicted lane width and the lane width range threshold, a lane width penalty value is determined.

[0198] (2.4) Determine the engineering design loss based on the speed penalty value, the radius penalty value, and the lane width penalty value.

[0199] The engineering design loss is obtained by summing (or weighted summing) the speed penalty value, radius penalty value, and lane width penalty value. The larger the loss value, the more serious the violation of physical and design specifications the engineering design parameters predicted by the model are.

[0200] (3) Using the attribute semantic rules, the attribute information of the first predicted element of the incremental map element sample is verified. If the attribute information of the first predicted element violates the attribute semantic rules, the attribute semantic loss is determined according to the degree of violation.

[0201] Among them, attribute semantic rules refer to the constraints on the rationality of the combination of attribute fields of map elements. Attribute semantic rules analyze a large number of compliant attribute combinations that have passed map review to learn the reasonable co-occurrence patterns between attribute fields, and use this as a basis for judging whether the attribute combination is reasonable. For example, prohibition signs are usually paired with circles and red. If a combination of "prohibition sign + inverted triangle + green" appears, the co-occurrence probability of this combination is low, violating the attribute semantic rules.

[0202] Specifically, the predicted attribute field combination (e.g., "flag type = prohibition + shape = inverted triangle + color = green") is encoded and input into a pre-trained language model. The model outputs the probability distribution of this combination belonging to each canonical category, and the maximum probability value is extracted as the co-occurrence probability. If the co-occurrence probability is lower than the 1% threshold, the violation penalty term is calculated using a negative logarithmic function. For example, when the co-occurrence probability is 0.005, the violation penalty term is... The lower the co-occurrence probability, the greater the penalty.

[0203] For example, the predicted attribute combination for a certain prohibition sign is "type = prohibition + shape = inverted triangle + color = green". The pre-trained language model calculates its co-occurrence probability to be 0.3% (<1% threshold), which increases the semantic loss of the attribute. If the predicted attribute combination of the flag is "Type: Prohibition, Shape: Circle, Color: Red" (canonical combination), the co-occurrence probability is 85% (>1% threshold), and the attribute semantic loss is 0.

[0204] Specifically, determining the semantic loss of attributes includes the following steps (3.1) to (3.3): (3.1) Combine the plurality of first prediction attribute fields to obtain a prediction attribute field combination, and encode the prediction attribute field combination to obtain an encoding vector.

[0205] For example, the predictive attribute fields of a sign include: sign type = prohibition, sign shape = circle, sign color = red, location type = intersection, and reflectivity level = IV.

[0206] Combine the values ​​of all predicted attribute fields into field name-value pairs to form a complete attribute description of the feature. For example: {sign type: prohibition, sign shape: circle, sign color: red, location type: intersection, reflectivity level: IV}.

[0207] The predicted attribute fields are combined and converted into a fixed-dimensional numerical vector for input into the pre-trained language model.

[0208] The encoding method can be any method, such as one-hot encoding.

[0209] (3.2) Input the encoding vector into the preset large language model to obtain the probability distribution of the predicted attribute field combination belonging to multiple preset standard categories; the preset standard category refers to the category obtained by pre-clustering the attribute field combination that has passed the verification.

[0210] Here, the pre-trained language model can be a pre-trained language model (PLM), which learns in advance the semantic associations and co-occurrence rules between attribute fields.

[0211] The preset standard categories refer to multiple "standard combination pattern" categories automatically summarized through cluster analysis based on attribute field combinations that have passed verification in historical drawing review data. Each category corresponds to a set of attribute combination patterns that have appeared frequently in history and have been verified by experts as reasonable.

[0212] For example, when a certain combination of normal prohibition sign attributes is input into the model, the output probability distribution is "prohibition sign standard class: 0.85, instruction sign standard class: 0.10, warning sign standard class: 0.03, others: 0.02".

[0213] (3.3) The maximum probability value in the probability distribution is taken as the co-occurrence probability of the predicted attribute field combination, and the attribute semantic loss is determined according to the co-occurrence probability and the co-occurrence probability threshold.

[0214] Co-occurrence probability can be a quantitative indicator of the reasonableness of attribute combinations. It is the maximum value in the probability distribution output by the pre-trained language model, representing the similarity between the current attribute combination and the best-matching normative category. The higher the co-occurrence probability, the closer the attribute combination is to a certain type of historical compliance pattern, and the more reasonable it is; the lower the co-occurrence probability, the less the combination matches any known normative category, and the more likely it is to be abnormal.

[0215] The co-occurrence probability threshold can be set according to actual needs, and is not limited here.

[0216] For example, when a certain prohibition sign attribute combination "prohibition + inverted triangle + green" is input into the model, the output probability distribution is [0.005, 0.003, 0.002, ...], with the maximum probability value being... If the threshold is reached, the semantic loss of the attribute is... The maximum probability value of a normal combination "ban + circle + red". If the threshold is reached, the semantic loss of the attribute is 0.

[0217] (4) Determine the rule-driven loss based on the topology graph constraint loss, the engineering design loss and the attribute semantic loss.

[0218] Rule-driven loss can be a weighted sum of topology graph constraint loss, engineering design loss, and attribute semantic loss.

[0219] In this implementation, topology, design, and semantic rules constrain model prediction from different perspectives, achieving multi-dimensional rule coverage.

[0220] S505: Determine the multi-source fusion loss based on the degree of inconsistency between the first multi-source incremental map data corresponding to each incremental map element sample; the first multi-source incremental map data are historical incremental map data samples collected by different acquisition devices at different times.

[0221] In this step, the multi-source fusion loss is used to quantify the degree of inconsistency between multi-source observation data.

[0222] The multi-source fusion loss is used to train the model to recognize whether multi-source observations are consistent. When there are contradictions in observations of the same element from different sources, the multi-source fusion loss increases, and the model learns to be vigilant about such situations through training. If the training samples are single-source data (i.e., the same element is observed only once), then this loss term is 0.

[0223] For example, if a certain element is observed 10 times, and is considered to exist 9 times and not exist 1 time, this indicates a certain degree of inconsistency and a moderate multi-source fusion loss. If it is considered to exist 5 times and not exist 5 times, this indicates a high degree of inconsistency and a high multi-source fusion loss.

[0224] It should be noted that if the historical incremental map data sample is single-source data, the multi-source fusion loss is 0.

[0225] S506: Determine the total loss based on the data-driven loss, the rule-driven loss, and the multi-source fusion loss, and adjust the parameters of the initial incremental verification model based on the total loss to obtain the trained incremental verification model.

[0226] In this embodiment, the total loss can be determined based on the data-driven loss, the rule-driven loss, and the multi-source fusion loss, as well as their respective weights. Then, the gradient of the model parameters is calculated based on the total loss, and the model parameters are updated using optimization algorithms such as gradient descent, so that the model prediction is more accurate, more in line with the rules, and more sensitive to inconsistencies among multiple sources.

[0227] Corresponding to the aforementioned embodiments of the map incremental data verification method, this application also provides embodiments of a map incremental data verification device.

[0228] Please see Figure 6 This is a schematic diagram illustrating the structure of a map incremental data verification device according to an exemplary embodiment of this application. Figure 6 As shown, the map incremental data verification device 600 includes: Data acquisition module 610 is used to acquire incremental map data; the incremental map data includes multiple incremental map elements; The index building module 620 is used to build incremental spatiotemporal index data based on the incremental map data. The incremental spatiotemporal index data includes global index data and spatiotemporal index data. The global index data includes the index identifier of the target area covered by the incremental map data. The spatiotemporal index data includes time index data and spatial index data. The time index data is used to record the timestamp of each incremental map element. The spatial index data includes the metadata, element identifier, and vector point chain of each incremental map element. The vector point chain is used to represent the shape of the incremental map element. The model verification module 630 is used to input the incremental spatiotemporal index data into the trained incremental verification model to obtain the prediction result for each incremental map element. The multi-dimensional verification module 640 is used to verify the prediction result of each incremental map element from multiple verification dimensions to obtain the verification result of each incremental map element; the multiple verification dimensions include spatial dimension, attribute dimension, type dimension and feature dimension. The map update module 650 is used to update the base map of the target area based on the verification results indicating the incremental map elements that have passed the verification.

[0229] In some implementations, the prediction result includes predicted element attribute information, prediction type, and predicted vector point chain; the multidimensional verification module 640 is specifically used for: For each incremental map element, the predicted vector point chain of the incremental map element is compared with the vector point chain of the corresponding map element in the base map to verify the position from the spatial dimension and obtain the spatial position verification result. Obtain the attribute trie, and compare the predicted element attribute information of the incremental map element with the attribute trie according to the semantic hierarchy to verify from the attribute dimension and obtain the attribute verification result. The predicted type probability distribution of the incremental map elements is compared with the preset type confidence threshold to verify the type from the type dimension and determine the type verification result. Based on the predicted vector point chain corresponding to the incremental map element, different shape distortion detection methods are used to perform shape detection on the incremental map element in order to verify it from the feature dimension and obtain the shape verification result. The verification result is determined based on the spatial location verification result, the attribute verification result, the type verification result, and the shape verification result.

[0230] In some embodiments, the predicted vector point chain includes multiple predicted vector points, the map element corresponds to a vector point chain, and the vector point chain includes multiple vector points; the multidimensional verification module 640 is specifically used for: The multiple predicted vector points are registered with the multiple vector points to obtain multiple vector point pairs, and the positional difference of each vector point pair is determined. Based on the positional differences of each vector point pair, a confidence range is determined, and based on the confidence range, the allowable fluctuation range of the true position of the incremental map element is determined. Determine the distance between the boundary of the allowed fluctuation range and the boundary of the corresponding map element in the base map, and determine the spatial location verification result based on the distance.

[0231] In some implementations, the predicted element attribute information includes multiple predicted attribute fields; the attribute trie includes element-level, element subclass-level, attribute field-level, and candidate value range-level; the multidimensional verification module is specifically used for: Based on the prediction type of the incremental map element, the attribute trie is traversed to determine the target element subclass corresponding to the prediction type, the set of attribute fields corresponding to the target element subclass, and the candidate value range corresponding to each attribute field. For each predicted attribute field, the predicted attribute field of the incremental map element is compared with the attribute field set field by field, and the mask state of the predicted attribute field is determined according to the value of the predicted attribute field and the value range of the corresponding attribute field; the mask state includes valid state, missing state and conflict state. The attribute verification result is determined based on the mask status of each predicted attribute field.

[0232] In some implementations, the map update module 650 is specifically used for: For incremental map elements whose verification results indicate that the verification has passed, multi-source map incremental data corresponding to the incremental map elements is obtained; the multi-source map incremental data is incremental map data collected by different acquisition devices at different times; Based on the multi-source map incremental data, the credibility of the incremental map elements is determined, and if the credibility is greater than the credibility threshold, the base map is updated based on the verification results indicating the verified incremental map elements.

[0233] In some implementations, the map update module 650 is specifically used for: Based on the source information of each map incremental data in the multi-source map incremental data, a credibility score for each map incremental data is determined; Based on the confidence scores corresponding to each map incremental data, the weights of each map incremental data are determined, and Bayesian voting fusion is performed on the multi-source map incremental data based on the weights of each map incremental data to obtain fusion estimation data and fusion confidence. Based on the difference between each map incremental data and the fused estimated data, the consistency loss is determined; The confidence level of the incremental map element is determined based on the fusion confidence level and the consistency loss.

[0234] In some implementations, the map update module 650 is specifically used for: A three-dimensional credibility matrix is ​​established based on the source information of each map incremental data in the multi-source map incremental data. The three-dimensional credibility matrix includes sensor type dimension, data source type dimension and acquisition time window dimension, and each dimension corresponds to a credibility coefficient. Based on the three-dimensional credibility matrix, the credibility score of each map increment data is determined.

[0235] In some implementations, each incremental map element has shape information, attribute information, confidence level, and time information; the index building module 620 is specifically used for: For each incremental map element, the snowflake algorithm is used to determine the element identifier based on the road segment prefix of the road segment to which the incremental map element belongs; The shape of the incremental map element is sampled for curvature to obtain the vector point chain; The metadata is extracted from the attribute information of the incremental map elements; the metadata includes at least one of the following: level, lane, function type, administrative division, map approval batch, and over-limit calculation status. Based on the element identifier, vector point chain, and metadata of each incremental map element, the spatial index data is determined, and the temporal index data is constructed based on the temporal information. Based on the spatial index data and the temporal index data, the index data is determined. The incremental spatiotemporal index data is generated based on the index data of each incremental map element.

[0236] In some implementations, the index building module 620 is specifically used for: For each incremental map element, a state arbitration process is performed on the incremental map element, and corresponding processing operations are executed on the incremental map element according to the state arbitration result; the state arbitration process includes addition arbitration, invalidation arbitration, or review arbitration; The incremental spatiotemporal index data is constructed based on each incremental map element and its corresponding state arbitration result.

[0237] In some implementations, the state arbitration process includes adding arbitration, and the state arbitration result includes adding arbitration result; the index building module 620 is specifically used for: The incremental map element is subjected to a new arbitration process to determine whether the base map contains a map element that is the same as the incremental map element, and the new arbitration result is obtained. If the newly added arbitration result indicates that the base map does not contain a map element that is the same as the incremental map element, the incremental map element is added to the base map.

[0238] In some implementations, the state arbitration process includes failure arbitration, and the state arbitration result includes a failure arbitration result; the index building module 620 is specifically used for: Determine whether the incremental map element has been collected within a preset continuous time period, and determine whether the confidence level of the incremental map element is less than a confidence level threshold, so as to perform failure arbitration on the incremental map element and obtain the failure arbitration result. If the failure arbitration result indicates that the incremental map element has not been collected within a preset continuous time period and the confidence level of the incremental map element is less than the confidence level threshold, the index data of the incremental map element is marked as failed data.

[0239] In some implementations, the state arbitration process includes review arbitration, and the state arbitration result includes a review arbitration result; the index building module 620 is specifically used for: Determine whether the incremental map element exists in different map incremental data, and determine whether the confidence level of the incremental map element in different map incremental data is the same, so as to perform a review and arbitration on the incremental map element and obtain the review and arbitration result; If the review and arbitration results indicate that the incremental map element exists in different map incremental data, and the confidence level of the incremental map element is different in different map incremental data, the conflict index between different map incremental data is determined. If the conflict index is greater than the conflict threshold, a review prompt message is generated; the review prompt message is used to indicate that the incremental map element should be manually reviewed.

[0240] Please see Figure 7 This is a schematic diagram illustrating the structure of another map incremental data verification device according to an exemplary embodiment of this application. Figure 7 As shown, the map incremental data verification device 600 further includes a model training module 660; the model training module 660 is used for: Obtain a training sample set; the training sample set includes historical map incremental data samples and corresponding labels; the labels refer to the compliance annotation information of the historical map incremental data samples; The historical map incremental data samples are input into the initial incremental verification model to obtain the prediction output result of each incremental map element sample in the historical map incremental data samples. The data-driven loss is determined based on the difference between the predicted output of each incremental map element sample and the corresponding compliance labeling information. The rule-driven loss is determined based on the degree of violation of the predicted output of each incremental map element sample relative to the preset map design rules. The multi-source fusion loss is determined based on the degree of inconsistency between the first multi-source incremental map data corresponding to each incremental map element sample; the first multi-source incremental map data are historical incremental map data samples collected by different acquisition devices at different times; Based on the data-driven loss, the rule-driven loss, and the multi-source fusion loss, the total loss is determined, and the parameters of the initial incremental verification model are adjusted according to the total loss to obtain the trained incremental verification model.

[0241] In some implementations, the prediction output of each incremental map element sample includes first predicted element attribute information, first predicted engineering design parameters, and first predicted correlation relationships between the incremental map element samples; the preset map design rules include topological constraints for constraining element correlation relationships, design specifications for constraining engineering design parameters, and attribute semantic rules for constraining the semantic rationality of attribute field combinations; the model training module 660 is specifically used for: Using the topology graph constraints, the first predicted correlation between each incremental map element sample is verified. If there is a correlation that violates the topology graph constraints, the topology graph constraint loss is determined according to the degree of violation. Using the design specifications, the first predicted engineering design parameters of the incremental map element samples are verified. If the first predicted engineering design parameters exceed the preset specification range, the engineering design loss is determined according to the degree of exceedance. Using the attribute semantic rules, the attribute information of the first predicted element of the incremental map element sample is verified. If the attribute information of the first predicted element violates the attribute semantic rules, the attribute semantic loss is determined according to the degree of violation. The rule-driven loss is determined based on the topology graph constraint loss, the engineering design loss, and the attribute semantic loss.

[0242] In some implementations, the model training module 660 is specifically used for: For each first predicted association, the association type of the first predicted association is determined; the association type includes the head-to-tail connection relationship between lane lines, the binding relationship between signs and poles, or the continuous direction relationship between guardrail segments; Based on the association type, determine the corresponding spatial measurement method and preset deviation threshold; Using the spatial measurement method, the actual deviation metric value of the first predicted association is determined, and based on the actual deviation metric value and the preset deviation threshold, the violation penalty item of the first predicted association is determined; The topology graph constraint loss is determined based on the violation penalty term of each first predicted association.

[0243] In some implementations, the incremental map element samples include lanes, and the predicted engineering design parameters include predicted curvature, predicted speed limit, predicted minimum ramp radius, predicted lane width, and predicted road grade; the design specifications include the constraint relationship between curvature and speed limit, the minimum ramp radius threshold, and the lane width range threshold; the model training module 660 is specifically used for: The lane state of the lane is obtained, and the standard speed range is determined based on the lane state and the predicted curvature using the constraint relationship between curvature and speed limit. The speed penalty value is determined based on the predicted speed limit and the standard speed range. Based on the predicted minimum radius of the ramp and the minimum radius threshold of the ramp, a radius penalty value is determined; Determine the lane width range threshold corresponding to the predicted road level of the lane, and determine the lane width penalty value based on the predicted lane width and the lane width range threshold; The engineering design loss is determined based on the speed penalty value, the radius penalty value, and the lane width penalty value.

[0244] In some implementations, the first predicted element attribute information includes multiple first predicted attribute fields; the model training module 660 is specifically used for: The plurality of first prediction attribute fields are combined to obtain a prediction attribute field combination, and the prediction attribute field combination is encoded to obtain an encoding vector; The encoded vector is input into a preset large language model to obtain the probability distribution of the predicted attribute field combination belonging to multiple preset standard categories; the preset standard category refers to the category obtained by pre-clustering the attribute field combination that has passed the verification. The maximum probability value in the probability distribution is used as the co-occurrence probability of the predicted attribute field combination, and the attribute semantic loss is determined based on the co-occurrence probability and the co-occurrence probability threshold.

[0245] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0246] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. 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 the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0247] Corresponding to the above-described method for verifying incremental map data, this disclosure also provides a computer device, such as... Figure 8 The diagram shown is a structural schematic of a computer device provided in an embodiment of this disclosure. Figure 8 As shown, the computer device 800 includes a processor 810, an internal bus 820, memory 830, a network interface 840, and non-volatile memory 850, and may also include other hardware required for its functions. One or more embodiments of this specification can be implemented in software, for example, the processor 810 reads the corresponding computer program from the non-volatile memory 850 into the memory 830 and then runs it. Of course, besides software implementation, one or more embodiments of this specification do not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution entity of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.

[0248] The memory 830, also known as internal memory, is used to temporarily store the computational data in the processor 810, as well as the data exchanged with non-volatile memory 850 such as hard disk. The processor 810 exchanges data with the non-volatile memory 850 through the memory 830.

[0249] In this embodiment, memory 830 is specifically used to store application code that executes the solution of this application, and its execution is controlled by processor 810. That is, when the computer device is running, processor 810 communicates with network interface 840, memory 830 and non-volatile memory 850 through internal bus 820, so that processor 810 executes the application code stored in memory 830 and non-volatile memory 850, and then executes the map incremental data verification method described in the above method embodiment.

[0250] Processor 810 may be an integrated circuit chip with signal processing capabilities. The aforementioned processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware microservices. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0251] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the computer device 800. In other embodiments of this application, the computer device 800 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0252] This disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the map incremental data verification method described in the above method embodiments. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0253] This disclosure also provides a computer program product carrying program code. The program code includes instructions that can be used to execute the steps of the map incremental data verification method in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.

[0254] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0255] The embodiments of the subject matter and functional operation described in this specification can be implemented in the following ways: digital electronic circuits, tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or combinations thereof. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier for execution by a data processing apparatus or for controlling the operation of a data processing apparatus. Alternatively or additionally, the program instructions may be encoded on artificially generated propagation signals, such as machine-generated electrical, optical, or electromagnetic signals, which are generated to encode information and transmit it to a suitable receiving device for execution by the data processing apparatus. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or combinations thereof.

[0256] The processing and logic flow described in this specification can be executed by one or more programmable computers that execute one or more computer programs to perform corresponding functions by operating on input data and generating output. The processing and logic flow can also be executed by dedicated logic circuitry—such as FPGAs (Field-Programmable Gate Arrays) or ASICs (Application-Specific Integrated Circuits), and the device can also be implemented as dedicated logic circuitry.

[0257] Computers suitable for executing computer programs include, for example, general-purpose and / or special-purpose microprocessors, or any other type of central processing unit. Typically, the central processing unit receives instructions and data from read-only memory and / or random access memory. Basic computer microservices include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, a computer will also include one or more mass storage devices for storing data, such as disks, magneto-optical disks, or optical disks, or the computer will be operatively coupled to such mass storage devices to receive data from or transfer data to them, or both. However, a computer is not required to have such devices. Furthermore, a computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name a few.

[0258] Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, such as semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD-ROM and DVD-ROM disks. Processors and memory may be supplemented by or incorporated into dedicated logic circuitry.

[0259] While this specification contains numerous specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather are primarily intended to describe features of specific embodiments of a particular invention. Certain features described in the various embodiments herein may also be implemented in combination in a single embodiment. Conversely, various features described in a single embodiment may also be implemented separately in various embodiments or in any suitable sub-combination. Furthermore, while features may function in certain combinations as described above and even initially claimed in this way, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may refer to a sub-combination or a variation thereof.

[0260] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and microservices in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program microservices and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0261] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings are not necessarily shown in a specific order or sequence to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

[0262] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for verifying incremental map data, characterized in that, include: Obtain incremental map data; The incremental map data includes multiple incremental map elements; Based on the incremental map data, incremental spatiotemporal index data is constructed. The incremental spatiotemporal index data includes global index data and spatiotemporal index data. The global index data includes the index identifier of the target area covered by the incremental map data. The spatiotemporal index data includes time index data and spatial index data. The time index data is used to record the timestamp of each incremental map element. The spatial index data includes the metadata, element identifier, and vector point chain of each incremental map element. The vector point chain is used to represent the shape of the incremental map element. The incremental spatiotemporal index data is input into the trained incremental verification model to obtain the prediction result for each incremental map element. The prediction result for each incremental map element is verified from multiple verification dimensions to obtain the verification result for each incremental map element; the multiple verification dimensions include spatial dimension, attribute dimension, type dimension and feature dimension. Based on the verification results indicating the incremental map elements that have passed verification, the base map of the target area is updated.

2. The method according to claim 1, characterized in that, The prediction results include predicted element attribute information, prediction type, and predicted vector point chain; The prediction result for each incremental map element is verified from multiple verification dimensions to obtain the verification result for each incremental map element, including: For each incremental map element, the predicted vector point chain of the incremental map element is compared with the vector point chain of the corresponding map element in the base map to verify the position from the spatial dimension and obtain the spatial position verification result. Obtain the attribute trie, and compare the predicted element attribute information of the incremental map element with the attribute trie according to the semantic hierarchy to verify from the attribute dimension and obtain the attribute verification result. The predicted type probability distribution of the incremental map elements is compared with the preset type confidence threshold to verify the type from the type dimension and determine the type verification result. Based on the predicted vector point chain corresponding to the incremental map element, different shape distortion detection methods are used to perform shape detection on the incremental map element in order to verify it from the feature dimension and obtain the shape verification result. The verification result is determined based on the spatial location verification result, the attribute verification result, the type verification result, and the shape verification result.

3. The method according to claim 2, characterized in that, The predicted vector point chain includes multiple predicted vector points, and each map element corresponds to a vector point chain, which includes multiple vector points. The step of comparing the predicted vector point chain of the incremental map element with the vector point chain of the corresponding map element in the base map to perform spatial verification and obtain a spatial location verification result includes: The multiple predicted vector points are registered with the multiple vector points to obtain multiple vector point pairs, and the positional difference of each vector point pair is determined. Based on the positional differences of each vector point pair, a confidence range is determined, and based on the confidence range, the allowable fluctuation range of the true position of the incremental map element is determined. Determine the distance between the boundary of the allowed fluctuation range and the boundary of the corresponding map element in the base map, and determine the spatial location verification result based on the distance.

4. The method according to claim 2, characterized in that, The predicted element attribute information includes multiple predicted attribute fields; the attribute trie includes element-level, element subclass-level, attribute field-level, and candidate value range-level; the step of comparing the predicted element attribute information of the incremental map element with the attribute trie according to semantic hierarchy to verify from the attribute dimension and obtain attribute verification results includes: Based on the prediction type of the incremental map element, the attribute trie is traversed to determine the target element subclass corresponding to the prediction type, the set of attribute fields corresponding to the target element subclass, and the candidate value range corresponding to each attribute field. For each predicted attribute field, the predicted attribute field of the incremental map element is compared with the attribute field set field by field, and the mask state of the predicted attribute field is determined according to the value of the predicted attribute field and the value range of the corresponding attribute field; the mask state includes valid state, missing state and conflict state. The attribute verification result is determined based on the mask status of each predicted attribute field.

5. The method according to claim 1, characterized in that, The process of updating the base map based on incremental map elements that have passed verification, according to the verification results, includes: For incremental map elements whose verification results indicate that the verification has passed, multi-source map incremental data corresponding to the incremental map elements is obtained; the multi-source map incremental data is incremental map data collected by different acquisition devices at different times; Based on the multi-source map incremental data, the credibility of the incremental map elements is determined, and if the credibility is greater than the credibility threshold, the base map is updated based on the verification results indicating the verified incremental map elements.

6. The method according to claim 5, characterized in that, The process of determining the credibility of incremental map elements based on the multi-source map incremental data includes: Based on the source information of each map incremental data in the multi-source map incremental data, a credibility score for each map incremental data is determined; Based on the confidence scores corresponding to each map incremental data, the weights of each map incremental data are determined, and Bayesian voting fusion is performed on the multi-source map incremental data based on the weights of each map incremental data to obtain fusion estimation data and fusion confidence. Based on the difference between each map incremental data and the fused estimated data, the consistency loss is determined; The confidence level of the incremental map element is determined based on the fusion confidence level and the consistency loss.

7. The method according to claim 6, characterized in that, The step of determining the credibility score of each map incremental data based on the source information of each map incremental data in the multi-source map incremental data includes: A three-dimensional credibility matrix is ​​established based on the source information of each map incremental data in the multi-source map incremental data. The three-dimensional credibility matrix includes sensor type dimension, data source type dimension and acquisition time window dimension, and each dimension corresponds to a credibility coefficient. Based on the three-dimensional credibility matrix, the credibility score of each map increment data is determined.

8. The method according to claim 1, characterized in that, Each incremental map element has shape information, attribute information, confidence level, and time information; the construction of incremental spatiotemporal index data based on the incremental map data includes: For each incremental map element, the snowflake algorithm is used to determine the element identifier based on the road segment prefix of the road segment to which the incremental map element belongs; The shape of the incremental map element is sampled for curvature to obtain the vector point chain; The metadata is extracted from the attribute information of the incremental map elements; the metadata includes at least one of the following: level, lane, function type, administrative division, map approval batch, and over-limit calculation status. Based on the element identifier, vector point chain, and metadata of each incremental map element, the spatial index data is determined, and the temporal index data is constructed based on the temporal information. Based on the spatial index data and the temporal index data, the index data is determined. The incremental spatiotemporal index data is generated based on the index data of each incremental map element.

9. The method according to claim 1, characterized in that, The construction of incremental spatiotemporal index data based on the incremental map data includes: For each incremental map element, a state arbitration process is performed on the incremental map element, and corresponding processing operations are executed on the incremental map element according to the state arbitration result; the state arbitration process includes addition arbitration, invalidation arbitration, or review arbitration; The incremental spatiotemporal index data is constructed based on each incremental map element and its corresponding state arbitration result.

10. The method according to claim 9, characterized in that, The state arbitration process includes adding arbitration, and the state arbitration result includes adding arbitration result; the process of performing state arbitration on the incremental map element and performing corresponding processing operations on the incremental map element according to the state arbitration result includes: The incremental map element is subjected to a new arbitration process to determine whether the base map contains a map element that is the same as the incremental map element, and the new arbitration result is obtained. If the newly added arbitration result indicates that the base map does not contain a map element that is the same as the incremental map element, the incremental map element is added to the base map.

11. The method according to claim 9, characterized in that, The status arbitration process includes invalidation arbitration, and the status arbitration result includes the invalidation arbitration result; The step of performing state arbitration on the incremental map element and executing corresponding processing operations on the incremental map element according to the state arbitration result includes: Determine whether the incremental map element has been collected within a preset continuous time period, and determine whether the confidence level of the incremental map element is less than a confidence level threshold, so as to perform failure arbitration on the incremental map element and obtain the failure arbitration result. If the failure arbitration result indicates that the incremental map element has not been collected within a preset continuous time period and the confidence level of the incremental map element is less than the confidence level threshold, the index data of the incremental map element is marked as failed data.

12. The method according to claim 9, characterized in that, The state arbitration process includes a review arbitration, and the state arbitration result includes a review arbitration result; the process of performing state arbitration on the incremental map element and executing corresponding processing operations on the incremental map element according to the state arbitration result includes: Determine whether the incremental map element exists in different map incremental data, and determine whether the confidence level of the incremental map element in different map incremental data is the same, so as to perform a review and arbitration on the incremental map element and obtain the review and arbitration result; If the review and arbitration results indicate that the incremental map element exists in different map incremental data, and the confidence level of the incremental map element in different map incremental data is different, the conflict index between different map incremental data is determined. If the conflict index is greater than the conflict threshold, a review prompt message is generated; the review prompt message is used to indicate that the incremental map element should be manually reviewed.

13. The method according to any one of claims 1-12, characterized in that, The trained incremental verification model is obtained through the following steps: Obtain a training sample set; the training sample set includes historical map incremental data samples and corresponding labels; the labels refer to the compliance annotation information of the historical map incremental data samples; The historical map incremental data samples are input into the initial incremental verification model to obtain the prediction output result of each incremental map element sample in the historical map incremental data samples. The data-driven loss is determined based on the difference between the predicted output of each incremental map element sample and the corresponding compliance labeling information. The rule-driven loss is determined based on the degree of violation of the predicted output of each incremental map element sample relative to the preset map design rules. The multi-source fusion loss is determined based on the degree of inconsistency between the first multi-source incremental map data corresponding to each incremental map element sample; the first multi-source incremental map data are historical incremental map data samples collected by different acquisition devices at different times; Based on the data-driven loss, the rule-driven loss, and the multi-source fusion loss, the total loss is determined, and the parameters of the initial incremental verification model are adjusted according to the total loss to obtain the trained incremental verification model.

14. The method according to claim 13, characterized in that, The prediction output for each incremental map element sample includes first predicted element attribute information, first predicted engineering design parameters, and first predicted correlation relationships among the incremental map element samples; the preset map design rules include topological constraints for constraining element correlation relationships, design specifications for constraining engineering design parameters, and attribute semantic rules for constraining the semantic rationality of attribute field combinations; determining the rule-driven loss based on the degree of violation of the preset map design rules by the prediction result of each incremental map element sample includes: Using the topology graph constraints, the first predicted correlation between each incremental map element sample is verified. If there is a correlation that violates the topology graph constraints, the topology graph constraint loss is determined according to the degree of violation. Using the design specifications, the first predicted engineering design parameters of the incremental map element samples are verified. If the first predicted engineering design parameters exceed the preset specification range, the engineering design loss is determined according to the degree of exceedance. Using the attribute semantic rules, the attribute information of the first predicted element of the incremental map element sample is verified. If the attribute information of the first predicted element violates the attribute semantic rules, the attribute semantic loss is determined according to the degree of violation. The rule-driven loss is determined based on the topology graph constraint loss, the engineering design loss, and the attribute semantic loss.

15. The method according to claim 14, characterized in that, The first predicted correlation between the incremental map element samples is verified using the topology graph constraints. If a correlation violates the topology graph constraints, the topology graph constraint loss is determined based on the degree of violation, including: For each first predicted association, the association type of the first predicted association is determined; the association type includes the head-to-tail connection relationship between lane lines, the binding relationship between signs and poles, or the continuous direction relationship between guardrail segments; Based on the association type, determine the corresponding spatial measurement method and preset deviation threshold; Using the spatial measurement method, the actual deviation metric value of the first predicted association is determined, and based on the actual deviation metric value and the preset deviation threshold, the violation penalty item of the first predicted association is determined; The topology graph constraint loss is determined based on the violation penalty term of each first predicted association.

16. The method according to claim 14, characterized in that, The incremental map element samples include lanes; the predicted engineering design parameters include predicted curvature, predicted speed limit, predicted minimum ramp radius, predicted lane width, and predicted road grade; the design specifications include the constraint relationship between curvature and speed limit, the minimum ramp radius threshold, and the lane width range threshold; the first predicted engineering design parameter of the incremental map element samples is verified using the design specifications; if the first predicted engineering design parameter exceeds the preset specification range, the engineering design loss is determined according to the degree of exceedance, including: The lane state of the lane is obtained, and the standard speed range is determined based on the lane state and the predicted curvature using the constraint relationship between curvature and speed limit. The speed penalty value is determined based on the predicted speed limit and the standard speed range. Based on the predicted minimum radius of the ramp and the minimum radius threshold of the ramp, a radius penalty value is determined; Determine the lane width range threshold corresponding to the predicted road level of the lane, and determine the lane width penalty value based on the predicted lane width and the lane width range threshold; The engineering design loss is determined based on the speed penalty value, the radius penalty value, and the lane width penalty value.

17. The method according to claim 14, characterized in that, The first predicted element attribute information includes multiple first predicted attribute fields; the step of using the attribute semantic rules to verify the first predicted element attribute information of the incremental map element sample, and if the first predicted element attribute information violates the preset attribute semantic rules, determining the attribute semantic loss according to the degree of violation, includes: The plurality of first prediction attribute fields are combined to obtain a prediction attribute field combination, and the prediction attribute field combination is encoded to obtain an encoding vector; The encoded vector is input into a preset large language model to obtain the probability distribution of the predicted attribute field combination belonging to multiple preset standard categories; the preset standard category refers to the category obtained by pre-clustering the attribute field combination that has passed the verification. The maximum probability value in the probability distribution is used as the co-occurrence probability of the predicted attribute field combination, and the attribute semantic loss is determined based on the co-occurrence probability and the co-occurrence probability threshold.

18. A device for verifying incremental map data, characterized in that, include: The data acquisition module is used to acquire incremental map data; The incremental map data includes multiple incremental map elements; An index building module is used to build incremental spatiotemporal index data based on the incremental map data. The incremental spatiotemporal index data includes global index data and spatiotemporal index data. The global index data includes the index identifier of the target area covered by the incremental map data. The spatiotemporal index data includes time index data and spatial index data. The time index data is used to record the timestamp of each incremental map element. The spatial index data includes the metadata, element identifier, and vector point chain of each incremental map element. The vector point chain is used to represent the shape of the incremental map element. The model validation module is used to input the incremental spatiotemporal index data into the trained incremental validation model to obtain the prediction result for each incremental map element. The multi-dimensional verification module is used to verify the prediction result of each incremental map element from multiple verification dimensions to obtain the verification result of each incremental map element; the multiple verification dimensions include spatial dimension, attribute dimension, type dimension and feature dimension. The map update module is used to update the base map of the target area based on the verification results indicating the incremental map elements that have passed verification.

19. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the map incremental data verification method according to any one of claims 1-17.

20. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the map incremental data verification method according to any one of claims 1-17.

Images