High-precision map data processing method and device

Abstract

The application discloses a high-precision map data processing method and device, comprising: a self-adaptive standardization engine based on space-time semantic perception is used to standardize geographic information data to obtain standardized geographic information data; the standardized geographic information data is classified horizontally and graded vertically to obtain a hierarchical data set; the hierarchical data set is subjected to data encryption and desensitization to obtain an encrypted and desensitized data set; the encrypted and desensitized data set is used to perform spatial range expansion auditing, relative height requirement auditing and auxiliary positioning element compliance expression auditing on a high-precision map to obtain an auditing result; in the case that the auditing result is passed, the encrypted and desensitized data set is used to optimize the high-precision map data, and the optimized high-precision map data is published to a map service platform. The application can improve the auditing efficiency and accuracy, and effectively guarantee the integrity and security of the high-precision map data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-precision map auditing, in particular to a high-precision map data processing method and device. BACKGROUND

[0002] The high-precision map is mainly applied to high-level automatic driving of L4 and above, and the behavior subject is an automatic driving algorithm rather than a human driver. The algorithm needs safety redundancy, so compared with the traditional navigation electronic map and the high-level auxiliary driving map, the algorithm puts forward higher level safety redundancy requirements for the spatial expression range of the high-precision map, the relative height information in multiple forms, and the auxiliary positioning elements. Therefore, these high-level safety redundancy requirements need to be considered when auditing the map. The traditional map auditing method often focuses on the auditing of the geometric precision, road network connectivity and other basic information of the map. However, for the high-precision map, these information cannot meet the high requirements of the automatic driving algorithm for safety redundancy, thereby resulting in low efficiency and poor accuracy of the map auditing, and it is difficult to ensure the integrity and safety of the high-precision map data. SUMMARY

[0003] The main purpose of the present application is to provide a high-precision map data processing method and device, which aims to solve the technical problem of low efficiency and poor accuracy of the traditional map auditing method, and difficulty in ensuring the integrity and safety of the high-precision map data.

[0004] To achieve the above purpose, the present application provides a high-precision map data processing method, which comprises:

[0005] obtaining geographical information data of a high-precision map;

[0006] standardizing the geographical information data based on a spatio-temporal semantic perception adaptive standardization engine to obtain standardized geographical information data;

[0007] horizontally classifying the standardized geographical information data based on an automatic clustering algorithm of multi-modal feature extraction and vertically classifying based on a knowledge graph reasoning algorithm of influence degree evaluation to obtain a hierarchical data set, wherein the hierarchical data set comprises a plurality of data sets of different categories, each category of data set comprises a plurality of data of different levels, the data sets of different categories comprise an image data set, a point cloud data set, a trajectory data set, an inertial navigation data set and a graph construction data set, and the data of different levels comprise core data, important data and general data;

[0008] performing data encryption and desensitization on the hierarchical data set to obtain an encrypted and desensitized data set;

[0009] According to the encrypted and desensitized data set, spatial range expansion auditing, relative height requirement auditing, and auxiliary positioning element compliance expression auditing are performed on the high-precision map, and an auditing result of the high-precision map is obtained.

[0010] In a case where the auditing result is that the high-precision map passes the auditing, high-precision map data is optimized according to the encrypted and desensitized data set, and the optimized high-precision map data is published to a map service platform.

[0011] In an embodiment, the adaptive standardization engine based on spatio-temporal semantic perception standardizes the geographic information data to obtain standardized geographic information data, including:

[0012] A unified spatio-temporal reference framework is constructed according to a global dynamic reference frame, an international terrestrial reference frame, and a regional crustal deformation model.

[0013] The geographic information data is aligned in spatio-temporal semantics according to the unified spatio-temporal reference framework to obtain spatio-temporally aligned data.

[0014] The spatio-temporally aligned data is converted in coordinate system by using a coordinate system conversion algorithm to obtain coordinate system converted data.

[0015] A geographic feature ontology knowledge base is constructed, and a standard attribute structure of an entity is determined based on the geographic feature ontology knowledge base.

[0016] The coordinate system converted data is subjected to attribute semantic normalization processing according to the standard attribute structure of the entity to obtain attribute semantically normalized data.

[0017] The attribute semantically normalized data is encoded in a preset hybrid encoding format to obtain standardized geographic information data.

[0018] In an embodiment, the standardized geographic information data is subjected to horizontal classification based on an automatic clustering algorithm extracted based on multi-modal features and vertical classification based on a knowledge graph reasoning algorithm evaluated based on an influence degree to obtain a hierarchical data set, including:

[0019] The standardized geographic information data is subjected to feature extraction based on a multi-modal embedding model to obtain feature vectors of various types of data.

[0020] The feature vectors of the various types of data are mapped to a unified semantic embedding space to obtain unified semantic embedding vectors.

[0021] The unified semantic embedding vectors are clustered based on an HDBSCAN clustering algorithm to obtain data sets of different categories.

[0022] construct a high-precision map risk knowledge graph, wherein the nodes in the high-precision map risk knowledge graph include geographic elements, functional impacts, and AD level;

[0023] Based on the high-precision map risk knowledge graph, the influence degree of the data in the different categories of data sets is evaluated using an influence degree evaluation algorithm to obtain a decision influence score;

[0024] According to the decision influence score, a knowledge graph reasoning path is constructed, and the different categories of data sets are vertically classified using the knowledge graph reasoning path to obtain a hierarchical data set.

[0025] In an embodiment, the hierarchical data set is subjected to data encryption and desensitization to obtain an encrypted and desensitized data set, comprising:

[0026] The hierarchical data set is desensitized using a classification desensitization strategy to obtain desensitized data sets of different categories, wherein the classification desensitization strategy includes a local blur desensitization strategy guided by semantic segmentation for image data sets, an attribute differential privacy desensitization strategy for point cloud data sets, a spatiotemporal randomization desensitization strategy for trajectory data sets, a noise addition desensitization strategy for inertial navigation data sets, and a hash confusion desensitization strategy for composition data sets;

[0027] The different levels of data in the desensitized data sets of different categories are encrypted using a hierarchical encryption strategy to obtain a hierarchically encrypted data set.

[0028] In an embodiment, the high-precision map is subjected to spatial range expansion audit, relative height requirement audit, and auxiliary positioning element compliance expression audit according to the encrypted and desensitized data set to obtain an audit result of the high-precision map, comprising:

[0029] According to the encrypted and desensitized data set, image data, point cloud data, trajectory data, inertial navigation data, and composition data are determined;

[0030] The spatial range expansion audit is performed according to the image data, the point cloud data, and the inertial navigation data to obtain an expanded range of the road periphery;

[0031] The relative height requirement audit is performed according to the point cloud data, the trajectory data, and the composition data to obtain relative height grading information of the map elements;

[0032] The auxiliary positioning element compliance expression audit is performed according to the composition data to obtain relative height grading information of the auxiliary positioning elements;

[0033] According to the road periphery expansion range, the relative height grading information of the map element, and the relative height grading information of the auxiliary positioning element, comprehensive evaluation is performed to obtain an audit result of the high-precision map.

[0034] In an embodiment, the road periphery expansion range is obtained by performing spatial range expansion auditing according to the image data, the point cloud data, and the inertial navigation data.

[0035] The road image is extracted based on the image data, and image perspective inverse transformation based on vanishing points is performed on the road image to obtain a corrected road image;

[0036] The point cloud data is preprocessed by a radius filtering algorithm, and road surface point cloud data is extracted based on the preprocessed point cloud data;

[0037] Road traffic markings are extracted based on the corrected road image and the road surface point cloud data;

[0038] Vehicle speed data is extracted based on the inertial navigation data, and a regression model of speed-reaction time-environment is constructed based on the vehicle speed data and corresponding safety distance requirements;

[0039] The reference distance matrix of road periphery expansion is determined based on the regression model of speed-reaction time-environment;

[0040] The road periphery expansion range is determined based on the road traffic markings and the reference distance matrix of road periphery expansion.

[0041] In an embodiment, the relative height grading information of the map element is obtained by performing relative height requirement auditing according to the point cloud data, the trajectory data, and the composition data, including:

[0042] A three-dimensional scene model is constructed based on the point cloud data, and height slicing processing is performed on the three-dimensional scene model to obtain slice layers of different heights;

[0043] The moving trajectory of a moving object is determined according to the trajectory data, and the moving trajectory is marked in the three-dimensional scene model;

[0044] The position and shape of a map element are determined based on the composition data, and the map element is marked in the three-dimensional scene model according to the position and shape;

[0045] The relative height information of the map element and the moving trajectory is obtained by analyzing the vertical spatial relationship between the map element and the moving trajectory in the slice layers;

[0046] The discrete object model and the continuous field model are respectively established according to the continuity of the relative height information;

[0047] performing a grading error analysis on the relative height information of the discrete object model to obtain grading fineness information;

[0048] performing slope curvature decomposition on the relative height information of the continuous field model to obtain spatial grid density information;

[0049] determining relative height grading information of the map element according to the grading fineness information and the spatial grid density information.

[0050] In an embodiment, the auxiliary positioning element compliance expression auditing according to the composition data includes:

[0051] extracting road feature points, intersection feature points and points of interest based on the composition data;

[0052] constructing an auxiliary positioning element network based on the position information of the road feature points, the intersection feature points and the points of interest;

[0053] performing relative height analysis based on the auxiliary positioning element network to obtain relative height information of the auxiliary positioning element;

[0054] performing grading processing on the relative height information of the auxiliary positioning element to obtain relative height grading information of the auxiliary positioning element.

[0055] In an embodiment, the comprehensive evaluation according to the road surrounding expansion range, the relative height grading information of the map element and the relative height grading information of the auxiliary positioning element includes:

[0056] determining a road expansion range deviation rate based on a preset road expansion range compliance standard;

[0057] performing obstacle recognition simulation and prediction based on the relative height grading information of the map element to obtain an obstacle recognition accuracy rate;

[0058] performing positioning accuracy simulation and prediction according to the relative height grading information of the auxiliary positioning element to obtain a positioning accuracy rate;

[0059] performing comprehensive evaluation according to the road expansion range deviation rate, the obstacle recognition accuracy rate and the positioning accuracy rate to obtain a comprehensive evaluation score of the high-precision map;

[0060] comparing the comprehensive evaluation score with a preset map review qualified score line to obtain a map review result of the high-precision map.

[0061] In addition, to achieve the above object, the application further provides a high-precision map data processing device, which comprises:

[0062] an acquisition module configured to acquire geographic information data of a high-precision map;

[0063] a standardization module configured to standardize the geographic information data based on a spatiotemporal semantic perception adaptive standardization engine to obtain standardized geographic information data;

[0064] a classification and grading module configured to perform horizontal classification on the standardized geographic information data based on an automatic clustering algorithm extracted based on multi-modal features and vertical grading based on a knowledge graph reasoning algorithm evaluated based on an influence degree to obtain a hierarchical data set, wherein the hierarchical data set comprises a plurality of data sets of different categories, each category of data set comprises a plurality of data of different levels, the data sets of different categories comprise an image data set, a point cloud data set, a trajectory data set, an inertial navigation data set and a layout data set, and the data of different levels comprise core data, important data and general data;

[0065] an encryption and desensitization module configured to perform data encryption and desensitization on the hierarchical data set to obtain an encrypted and desensitized data set;

[0066] an auditing module configured to perform spatial range expansion auditing, relative height requirement auditing and auxiliary positioning element compliance expression auditing on the high-precision map based on the encrypted and desensitized data set to obtain a map auditing result of the high-precision map;

[0067] a publishing module configured to, in the case that the map auditing result is passed, optimize the high-precision map data based on the encrypted and desensitized data set, and publish the optimized high-precision map data to a map service platform.

[0068] One or more technical solutions proposed in the present application acquire geographic information data of a high-precision map; a self-adaptive standardization engine based on spatiotemporal semantic perception is used to standardize the geographic information data, to obtain standardized geographic information data; a horizontal classification is performed on the standardized geographic information data based on an automatic clustering algorithm of multi-modal feature extraction, and a vertical classification is performed based on a knowledge graph reasoning algorithm of influence degree evaluation, to obtain a hierarchical data set, wherein the hierarchical data set includes a plurality of data sets of different categories, each category of data set includes a plurality of data of different levels, different categories of data sets include image data sets, point cloud data sets, trajectory data sets, inertial navigation data sets, and composition data sets, and different levels of data include core data, important data, and general data; data encryption and desensitization is performed on the hierarchical data set, to obtain an encrypted and desensitized data set; according to the encrypted and desensitized data set, spatial range expansion review, relative height requirement review, and auxiliary positioning element compliance expression review are performed on the high-precision map, to obtain a review result of the high-precision map; in the case that the review result of the high-precision map is passed, the high-precision map data is optimized according to the encrypted and desensitized data set, and the optimized high-precision map data is published to a map service platform. Through the above-mentioned manner, by introducing a self-adaptive processing mechanism of classification and grading and a data encryption and desensitization technology, efficient utilization of high-precision map data is realized, and then through a plurality of reviews on the high-precision map, the review efficiency and accuracy are improved, and the integrity and security of the high-precision map data are effectively ensured. BRIEF DESCRIPTION OF DRAWINGS

[0069] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and serve to explain the principles of the present application together with the specification.

[0070] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0071] Figure 1 A flowchart is provided for the high-precision map data processing method embodiment one of the present application;

[0072] Figure 2 A flowchart is provided for the high-precision map data processing method embodiment two of the present application;

[0073] Figure 3 A module structure diagram of the high-precision map data processing device of the present application embodiment is provided.

[0074] The object, functional features and advantages of the present application will be further illustrated in combination with embodiments and with reference to the drawings. DETAILED DESCRIPTION

[0075] It should be understood that the specific embodiments described herein are merely intended to explain the technical solutions of the present application, and are not intended to limit the present application.

[0076] In order to better understand the technical solutions of the present application, the following will be described in detail in combination with the drawings and specific embodiments.

[0077] The main solution of the embodiment of the present application is: obtaining geographic information data of a high-precision map; performing standardization on the geographic information data based on a spatio-temporal semantic perception adaptive standardization engine to obtain standardized geographic information data; performing horizontal classification on the standardized geographic information data based on an automatic clustering algorithm of multi-modal feature extraction and performing vertical classification based on a knowledge graph reasoning algorithm of influence degree evaluation to obtain a hierarchical data set, wherein the hierarchical data set includes a plurality of data sets of different categories, each category of data set includes a plurality of data of different levels, different categories of data sets include image data set, point cloud data set, trajectory data set, inertial navigation data set and composition data set, and different levels of data include core data, important data and general data; performing data encryption and desensitization on the hierarchical data set to obtain an encrypted and desensitized data set; performing spatial range expansion audit, relative height requirement audit and auxiliary positioning element compliance expression audit on the high-precision map according to the encrypted and desensitized data set to obtain an audit result of the high-precision map; in the case that the audit result of the high-precision map is passed, optimizing the high-precision map data according to the encrypted and desensitized data set, and publishing the optimized high-precision map data to a map service platform.

[0078] The traditional map audit method often focuses on the audit of basic information such as geometric precision and road network connectivity of the map. However, for high-precision maps, these information cannot meet the high requirements of safety redundancy of automatic driving algorithms, resulting in low efficiency and poor accuracy of map audit, and difficulty in ensuring the integrity and security of high-precision map data.

[0079] The present application provides a solution by introducing a classification and grading adaptive processing mechanism and data encryption and desensitization technology, which realizes efficient use of high-precision map data, and further improves the efficiency and accuracy of map audit by performing multiple audits on high-precision maps, while effectively ensuring the integrity and security of high-precision map data.

[0080] It should be noted that the execution subject of the embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, or an electronic device capable of realizing the above functions, a high-precision map data processing device, etc. The following will take the high-precision map data processing device as an example to describe the embodiment and the following embodiments.

[0081] Based on this, the embodiment of the application provides a high-precision map data processing method, which refers to Figure 1 , Figure 1 The flowchart of the first embodiment of the high-precision map data processing method of the application is shown in the figure.

[0082] In the embodiment, the high-precision map data processing method comprises steps S10-S60:

[0083] Step S10: Obtain geographic information data of a high-precision map.

[0084] It should be noted that the high-precision map is also called an autonomous driving map, which is a high-precision positioning and environment perception map for autonomous driving. The geographic information data of the high-precision map is obtained by collecting and integrating the accessed data through a multi-source heterogeneous data access layer. The multi-source heterogeneous data access layer includes a unified data access middleware, supports access to vehicle-mounted LiDAR, unmanned aerial vehicle aerial image, RTK-GNSS track, IMU inertial navigation data, crowdsourcing and crowdsourcing data, road sign OCR recognition results, etc. The access protocol supports MQTT, HTTP / 3, ROS2, etc. and is suitable for real-time uploading of autonomous driving vehicles, thereby realizing real-time updating and comprehensive coverage of high-precision map data. These data cover multiple categories such as images, point clouds, tracks, inertial navigation and composition, providing rich basic materials for subsequent data processing.

[0085] Step S20: The self-adaptive standardization engine based on spatio-temporal semantic perception standardizes the geographic information data to obtain standardized geographic information data.

[0086] It should be noted that the standardization processing is to ensure that data of different sources and different formats can be uniformly processed and analyzed. The self-adaptive standardization engine based on spatio-temporal semantic perception adopted in the embodiment can perform spatio-temporal semantic alignment, coordinate system intelligent conversion and attribute semantic normalization on the geographic information data, and finally output a standardized data format, thereby solving the problem of data heterogeneity and providing a unified data format and standard for subsequent data processing.

[0087] In an implementable embodiment, step S20 can include: constructing a unified spatiotemporal reference framework according to the global dynamic reference frame, the international terrestrial reference frame, and the regional crustal deformation model; performing spatiotemporal semantic alignment on the geographic information data according to the unified spatiotemporal reference framework to obtain spatiotemporally aligned data; performing coordinate system conversion on the spatiotemporally aligned data by using a coordinate system conversion algorithm to obtain coordinate system converted data; constructing a geographic feature ontology knowledge base, determining an entity standard attribute structure based on the geographic feature ontology knowledge base; performing attribute semantic normalization processing on the coordinate system converted data according to the entity standard attribute structure to obtain attribute semantically normalized data; and encoding the attribute semantically normalized data in a preset hybrid encoding format to obtain standardized geographic information data.

[0088] It should be noted that the global dynamic reference frame, i.e., GDBF, is a high-precision geospatial reference framework for providing dynamically updated reference data to ensure the accuracy and reliability of global positioning and measurement. The international terrestrial reference frame, i.e., ITRF, is a global geodetic reference system for providing an accurate global terrestrial reference frame. The regional crustal deformation model is a mathematical model describing the deformation of the crust in a specific region over time.

[0089] It can be understood that by introducing the global dynamic reference frame and combining the international terrestrial reference frame and the regional crustal deformation model, a unified spatiotemporal reference framework is constructed, which can achieve centimeter-level coordinate dynamic correction. By performing spatiotemporal semantic alignment on geographic information data through the unified spatiotemporal reference framework, the spatiotemporal accuracy and consistency of geographic information data are effectively improved.

[0090] It can be understood that the Transformer-based GeoAlignNet model is used to automatically identify the original coordinate system, timestamp format, and unit system in the spatiotemporally aligned data and intelligently match them to the target standard, such as WGS84+EGM2008 elevation+UTC time, to achieve intelligent conversion between different coordinate systems, thereby ensuring the accuracy and consistency of geographic information.

[0091] Attribute semantic normalization processing is based on the geographic feature ontology knowledge base to uniformly describe and standardize the attribute information of data, solving the problem of inconsistent attribute information. By encoding the data in a hybrid encoding format, not only is the standardized storage of data achieved, but also quick retrieval and efficient utilization of data are facilitated.

[0092] It is worth noting that the geographic feature ontology knowledge base is a database for storing and managing geographic feature ontology knowledge. According to the geographic feature ontology knowledge base, the standard attribute structure of entities such as "lane line", "traffic light" and "roadside guardrail" can be defined. Then, the standard attribute structure of the entity is parsed and normalized to ensure that the attribute information of each geographic feature conforms to the unified standard and specification.

[0093] The preset hybrid encoding format can be eoJSON-Lite+Protobuf hybrid encoding format, which takes into account the readability and transmission efficiency, and the fields strictly follow the ISO 19157 geographic information quality standard, thereby ensuring the efficient transmission and accuracy of the data. According to the eoJSON-Lite+Protobuf hybrid encoding format, the attribute semantic normalized data is efficiently compressed and stored, which not only reduces the storage space occupation, but also speeds up the data reading speed, and improves the overall efficiency of data processing.

[0094] Step S30: The standardized geographic information data is classified horizontally based on the automatic clustering algorithm of multi-modal feature extraction and vertically classified based on the knowledge graph reasoning algorithm of influence degree evaluation, to obtain a hierarchical data set, wherein the hierarchical data set includes a plurality of data sets of different categories, each category of data set includes a plurality of data of different levels, different categories of data sets include image data set, point cloud data set, track data set, inertial navigation data set and composition data set, and different levels of data include core data, important data and general data.

[0095] It should be noted that the classification and grading processing of the standardized geographic information data is to improve the utilization efficiency of the data. The automatic clustering algorithm of multi-modal feature extraction used in this embodiment can automatically identify the features of the data and perform clustering, so as to classify the data into different categories such as image, point cloud, track, inertial navigation and composition. This automatic clustering method not only improves the efficiency of data processing, but also ensures the accuracy and objectivity of classification.

[0096] At the same time, the knowledge graph reasoning algorithm based on influence degree evaluation can vertically classify the data. This algorithm analyzes the relevance and influence degree between data and classifies data into different levels such as core data, important data and general data. This hierarchical processing method helps to adopt different processing strategies and priorities for data of different levels in subsequent data processing and auditing process, thereby improving the overall processing efficiency and accuracy.

[0097] Through the classification and grading processing, a hierarchical data set is obtained. The set not only contains a plurality of different categories of data sets, such as image data sets, point cloud data sets, trajectory data sets, inertial navigation data sets and composition data sets, but also contains different levels of data in each category, such as core data, important data and general data.

[0098] It is worth noting that the image data is image or video data collected by a space-time data sensor or an unmanned aerial vehicle; the point cloud data includes three-dimensional point cloud data collected by a LiDAR sensor; the trajectory data is vehicle driving trajectory data collected by an RTK-GNSS, an IMU and other sensors, and is data that can determine the absolute coordinates of the vehicle on the earth; the inertial navigation data is vehicle attitude and motion state data collected by an IMU inertial navigation system, such as vehicle attitude angle (or angular rate), acceleration and derived data; and the composition data is vector data containing absolute coordinates generated based on the above data.

[0099] It is worth noting that the data level is determined according to the following rules: 1. The intelligent automobile basic map data meeting any of the following conditions shall be determined as core data: important data covering the national range and having a large scale; meeting the core data level determination rules specified in GB / T 43697. 2. The intelligent automobile basic map data meeting any of the following conditions shall be determined as important data: reflecting the geographic information of important and sensitive areas such as military management areas, national defense science and technology units and party and government organs above the county level, national strategic reserve capacity, related attributes of important infrastructure such as transportation and energy, image data, point cloud data, trajectory data, inertial navigation data, composition data, map product data and other data related to important targets and areas reaching a certain scale; meeting the important data level determination rules specified in GB / T 43697. 3. Other data not identified as core data or important data is determined as general data.

[0100] In an implementable embodiment, step S30 can include: performing feature extraction on the standardized geographic information data based on a multi-modal embedding model to obtain feature vectors of various types of data; mapping the feature vectors of the various types of data to a unified semantic embedding space to obtain unified semantic embedding vectors; clustering the unified semantic embedding vectors based on an HDBSCAN clustering algorithm to obtain data sets of different categories; constructing a high-definition map risk knowledge graph, wherein nodes in the high-definition map risk knowledge graph include geographic elements, functional impacts, and safety levels; performing impact degree evaluation on data in the data sets of different categories based on the high-definition map risk knowledge graph using an impact degree evaluation algorithm to obtain decision influence scores; constructing a knowledge graph reasoning path according to the decision influence scores, and performing vertical classification on the data sets of different categories using the knowledge graph reasoning path to obtain hierarchical data sets.

[0101] It should be noted that the multi-modal embedding model MM-Embedder is used to extract feature vectors of various types of data. For example, for image data, CNN is used to extract texture, color, and object distribution features; for point cloud data, PointNet++ is used to extract geometric density, curvature, and reflectivity intensity distribution; for trajectory data, LSTM is used to extract motion patterns and acceleration distribution; for inertial navigation data, wavelet analysis is used to extract vibration spectrum features; and for composition data, graph neural network is used to extract topological connectivity.

[0102] Mapping the feature vectors of various types of data to a unified semantic embedding space can eliminate the semantic gap between different modal data, making the data comparable and consistent at the semantic level, thereby facilitating clustering.

[0103] Clustering the unified semantic embedding vectors based on the HDBSCAN clustering algorithm can automatically identify the internal structure and distribution law of the data, and divide the data into different categories, such as image, point cloud, trajectory, inertial navigation, and composition.

[0104] In a specific implementation, since the embedding vectors are high-dimensional vectors in a semantic space, cosine distance is used to measure similarity, as follows:

[0105]

[0106] wherein, is the cosine distance, z i is the unified semantic embedding vector of the i-th geographic information data, z j represents the semantic embedding vector of the j-th geographic information data.

[0107] HDBSCAN defines clusters based on local density, the core distance is the core distance, assuming a cluster contains at least k points, for each point z i , the core distance is: k (z i )=kth smallest distance to other points, that is, take the kth smallest value in all dist(z i ,z j ), the core distance is used to measure the density of the region where the point is located, the higher the density, the smaller the core distance.

[0108] If both points are dense, that is, the core distance is small, then the mutual reach distance is introduced, the mutual reach distance takes the maximum value of the core distance core_distance i (z k ), the core distance core_distance i (z j ) of z k and z j and the pre-distance between z i and z j . In this case, the mutual reach distance is close to the original distance. If a point is sparse, the mutual reach distance is enlarged, which suppresses it from becoming a cluster center.

[0109] Further, a complete graph is constructed using the mutual reach distance matrix, wherein the nodes in the complete graph are all N embedding vectors, and the edge weight is the mutual reach distance; then Kruskal algorithm is run based on the complete graph to generate a minimum spanning tree; the minimum spanning tree is pruned according to the edge weight from large to small, simulating the clustering state under different density thresholds,

[0110] generate a cluster hierarchy, that is, multiple candidate clusters; for each candidate cluster, calculate its existence time under different density thresholds, that is, cluster survival time cst, as follows:

[0111]

[0112] wherein, is the cluster survival time of data point i, is the cluster C containing point i, λ=1 / distance, indicating the density intensity, is the birth intensity parameter of the cluster, is the death intensity parameter of the cluster.

[0113] Select the clustering division that maximizes the total cst to obtain K effective clusters and the cluster label of each point, and map the cluster label to a semantic category such as an image category. Specifically, for each cluster C k , count the original data sources of its members as follows:

[0114]

[0115] wherein, is the probability that a given cluster C k is the probability that a member of C k is the kth cluster, is the counting symbol, denoting the number of elements in a set, is the source function of vector z i .

[0116] Set threshold automatic labeling, if , it is labeled as image class; , it is labeled as point cloud class; if , it is labeled as trajectory / INS class; , it is labeled as composition class.

[0117] After completing the data classification, a high-precision map risk knowledge graph is constructed, wherein the nodes in the high-precision map risk knowledge graph represent key information such as geographic elements, functional impacts, and safety levels, and the edges represent the association relationships between the nodes. For each geographic element e, all propagation paths thereof in the knowledge graph are found, and the weight of each path is:

[0118]

[0119] wherein, is the weight of the propagation path p, is a normalization coefficient, is the confidence of the edge, 0~1, from expert labeling or historical data statistics, is a decay factor, usually set to 0.9, is the functional impact severity score, is the safety level mapping value.

[0120] According to the weight of the path, the element impact degree score of the element can be further determined:

[0121]

[0122] wherein, is the element impact degree score of the geographic element e, is the weight of the propagation path p, and P(e) is all paths starting from e.

[0123] The geographic elements are classified according to the data sources, and then the decision impact score of each type of data is determined according to the element impact degree score. Based on the decision impact degree score, a reasoning path is constructed to realize the vertical hierarchical management of data, i.e., core>important>general.

[0124] Step S40: data encryption and desensitization is performed on the hierarchical data set to obtain a data set after encryption and desensitization.

[0125] It should be noted that by encrypting and desensitizing the data, data leakage and misuse can be effectively prevented, and data security can be ensured. In this embodiment, the hierarchical data set is classified and desensitized and encrypted.

[0126] Classification desensitization refers to using different desensitization strategies to process data according to its category, for example, for image data, pixelization, blurring or adding noise can be used for desensitization, for point cloud data, attribute differential privacy processing, random offset or removing part of the feature can be used for desensitization; for trajectory data and inertial navigation data, noise addition, time blurring or spatial generalization can be used for processing; for composition data, desensitization can be achieved by adjusting the topological structure, changing the node attribute or adding false nodes, and the present embodiment does not make specific limitations.

[0127] Hierarchical encryption refers to using different encryption algorithms and key lengths to encrypt data according to its sensitivity level, to ensure the security of data during transmission and storage, for example, for core data, high-strength encryption algorithm and long key can be used for encryption to ensure high security of data; for important data, medium-strength encryption algorithm and key length can be used; for general data, lower-strength encryption algorithm can be used for protection. Through such classification desensitization and hierarchical encryption, the security of the data can be effectively improved, and data leakage and misuse can be prevented.

[0128] In one possible implementation, step S40 can include: using a classification desensitization strategy to desensitize the hierarchical data set to obtain a desensitized data set of different categories, wherein the classification desensitization strategy includes using a semantic segmentation guided local blur desensitization strategy for image data set, using an attribute differential privacy desensitization strategy for point cloud data set, using a spatio-temporal randomization desensitization strategy for trajectory data set, using a noise addition desensitization strategy for inertial navigation data set, and using a hash confusion desensitization strategy for composition data set; using a hierarchical encryption strategy to encrypt different levels of data in the desensitized data set of different categories to obtain a hierarchically encrypted data set.

[0129] It should be noted that in the present embodiment, different categories of data sets are subjected to corresponding desensitization strategies. For image data sets, a semantic segmentation guided local blur desensitization strategy is adopted, for example, only sensitive areas such as human faces, license plates, and door numbers are replaced by GAN generated replacement, and the road structure clarity is preserved. For point cloud data sets, a perturbation and attribute differential privacy desensitization strategy is adopted, for example, points higher than 1.5 meters above the ground are subjected to Z-axis perturbation with a perturbation height of ±0.3m, and the ground geometry is kept unchanged; attribute differential privacy is used to add Laplace noise to the reflectivity; for trajectory data sets, a spatio-temporal randomization desensitization strategy is adopted, that is, through spatio-temporal k-anonymity + continuous privacy protection, it is ensured that at least k trajectories in any time period are indistinguishable; for inertial navigation data sets, a noise addition desensitization strategy is adopted, for example, Fourier transform is performed on the acceleration and angular velocity sequence, noise is added in the frequency domain, and then inverse transform is performed to restore the motion trend; for composition data sets, a hash confusion desensitization strategy is adopted, for example, the node ID in the graph is hashed and confused, and the edge relationship is preserved but the attribute is desensitized.

[0130] It can be understood that the different levels of data in the desensitized different categories of data sets are hierarchically encrypted according to the data level, for example, the core data is subjected to end-to-end encryption using an anti-quantum encryption algorithm such as CRYSTALS-Kyber, and the key is managed by a trusted execution environment TEE; important data is encrypted using AES-256-GCM, with additional integrity verification; general data is encrypted using lightweight ChaCha20, which is suitable for fast processing by edge devices.

[0131] In specific implementation, the specific implementation of the classification desensitization and hierarchical encryption strategy can be adjusted and optimized according to actual needs. For example, in the classification desensitization stage, a deep learning model can be used for automatic desensitization processing, and the most suitable desensitization strategy is automatically selected according to the characteristics of the data; in the hierarchical encryption stage, the encryption algorithm and key length can be dynamically adjusted according to the access rights and use scenarios of the data, to realize more flexible and fine-grained data protection. In addition, in order to ensure the efficiency and accuracy of data processing, parallel processing and distributed computing techniques can also be used to efficiently process large-scale data sets.

[0132] Step S50: According to the encrypted and desensitized data set, the high-precision map is subjected to spatial range expansion audit, relative height requirement audit, and auxiliary positioning element compliance expression audit, to obtain the audit result of the high-precision map.

[0133] It should be noted that the spatial range expansion is due to the complexity of some scenes, blind area or weak signal area of traffic section, in order to supplement the lack of timely perception of vehicles, effectively deal with the emergency braking situation such as "ghost probe", the advanced automatic driving algorithm needs to know the traffic environment in a certain buffer area on both sides of the road through high-precision map "a priori", to exchange space for braking time, to ensure the driving safety under a certain driving speed. However, the expanded area may involve rich geographical elements around the road, may contain sensitive geographical places, high-voltage line substation and other important public service facilities, and potential geographic information security risks, so it is necessary to study the distance of the expandable buffer area

[0134] It can be understood that the relative height requirement is due to the need for advanced automatic driving algorithm to understand some important relative height, height level, near ground surface fluctuation (relative height difference) and other information, such as the relative height of traffic signal lamp, which is used to distinguish traffic signal lamp and tail light of car in front, the relative height level relationship of viaduct, which is used to ensure that the road level can be quickly and clearly determined when starting on the viaduct, and the need to understand the relative height fluctuation of near ground surface, which is used to distinguish the near ground surface road surface protrusion and the low objects temporarily appearing on the roadside, such as the person squatting on the roadside or the small animals suddenly rushing out, etc. However, the continuous relative height information relates to the security of geographic information, and the existing high auxiliary map does not allow direct expression of height information. Therefore, the audit of relative height needs to use indirect and fuzzy way to express relative height information, in order to ensure the safety and compliance of map use.

[0135] The auxiliary positioning element requirement is due to the fact that the spatial position accuracy of high-level automatic driving is generally within 20 cm. In order to make up for the inaccuracy of positioning caused by weak positioning signal, it is necessary to use high-precision map to provide prior spatial feature positioning information, that is, to select spatial feature positioning elements such as point, line and surface geometric elements in road and surrounding scene, to construct the description of spatial scene, and then to realize spatial accurate positioning by matching the spatial scene with the timely perception of vehicle.

[0136] Therefore, through the comprehensive audit of spatial range expansion, relative height requirement, auxiliary positioning element compliance expression and other audits, the safety, compliance and accuracy of high-precision map in automatic driving application can be ensured. The spatial range expansion audit mainly focuses on whether the expanded area involves sensitive geographical places and whether the expanded distance is reasonable, in order to avoid leaking important geographical information. The relative height requirement audit focuses on checking whether the expression method of relative height information is indirect and fuzzy, in order to prevent the leakage of continuous relative height information. The auxiliary positioning element compliance expression audit focuses on whether the selected spatial feature positioning elements are accurate and compliant, and whether they can effectively support the spatial accurate positioning of automatic driving.

[0137] Step S60: In the case that the review result of the high-precision map is passed, the high-precision map data is optimized according to the encrypted and desensitized data set, and the optimized high-precision map data is published to the map service platform.

[0138] It should be noted that through the optimization of high-precision map data, it can ensure that the high-precision map data can be efficiently and accurately utilized by the autonomous driving system on the map service platform. The map service platform can be a cloud server or a distributed storage system for storing, managing and providing high-precision map data services. Before publishing the optimized high-precision map data to the map service platform, the data can be further format-converted, compressed and processed, etc. to adapt to the needs of different autonomous driving systems and devices.

[0139] In specific implementation, the optimization step can include but is not limited to precision improvement of map data, data redundancy removal, path planning algorithm optimization. Precision improvement can be achieved by further fine processing of the encrypted and desensitized data set, such as more detailed depiction of road boundaries, traffic signs, obstacles, etc. to improve the recognition ability of autonomous driving vehicles to road environment. Data redundancy removal is to reduce the size of map data and improve the efficiency of data transmission and loading, which can be achieved by compressing and merging similar elements of map data. Path planning algorithm optimization is to analyze the map data intelligently according to the driving needs of autonomous driving vehicles, optimize the path planning algorithm, and improve the efficiency and safety of autonomous driving.

[0140] The embodiment provides a high-precision map data processing method, geographical information data of a high-precision map is acquired, the geographical information data is standardized based on a spatiotemporal semantic perception adaptive standardization engine, and standardized geographical information data is obtained; the standardized geographical information data is subjected to horizontal classification based on an automatic clustering algorithm of multi-modal feature extraction and vertical classification based on a knowledge graph reasoning algorithm of influence degree evaluation, and a hierarchical data set is obtained, wherein the hierarchical data set comprises a plurality of data sets of different categories, each category of data set comprises a plurality of data of different levels, the data sets of different categories comprise an image type data set, a point cloud type data set, a trajectory type data set, an inertial navigation type data set and a mapping type data set, and the data of different levels comprises core data, important data and general data; the hierarchical data set is subjected to data encryption and desensitization, and an encrypted and desensitized data set is obtained; the high-precision map is subjected to spatial range expansion audit, relative height requirement audit and auxiliary positioning element compliance expression audit according to the encrypted and desensitized data set, and an audit result of the high-precision map is obtained; in the case that the audit result of the high-precision map is passed, the high-precision map data is optimized according to the encrypted and desensitized data set, and the optimized high-precision map data is published to a map service platform. In the above manner, by introducing the adaptive processing mechanism of classification and grading and the data encryption and desensitization technology, efficient utilization of high-precision map data is realized, and then the high-precision map is subjected to a plurality of audits, the audit efficiency and accuracy are improved, and the integrity and security of the high-precision map data are effectively ensured.

[0141] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as the above embodiment one can refer to the above introduction, and the following will not be described in detail. On this basis, please refer to Figure 2 , step S10 includes steps S501-S504:

[0142] Step S501: Determine the image type data, point cloud type data, trajectory type data, inertial navigation type data and mapping type data according to the encrypted and desensitized data set.

[0143] It should be noted that the security and privacy protection of the data in the encrypted and desensitized data set are enhanced. The image type data is usually derived from satellite remote sensing, unmanned aerial vehicle aerial photography or vehicle-mounted camera, and provides rich road and surrounding environment information; the point cloud type data is obtained through laser radar and other devices, and can accurately depict the three-dimensional shape of an object; the trajectory type data records the motion path of a vehicle or other moving object; the inertial navigation type data contains acceleration, angular velocity and other inertial navigation information, which helps to determine the position and attitude of the vehicle; and the mapping type data is vector data containing absolute coordinates obtained by integrating the above-mentioned various types of data.

[0144] Step S502: performing spatial range expansion auditing according to the image data, the point cloud data and the inertial navigation data, to obtain a road surrounding expansion range.

[0145] It should be noted that in the process of spatial range expansion auditing, the macroscopic road and surrounding environment overview provided by the image data, the object three-dimensional shape information accurately depicted by the point cloud data, and the vehicle position and attitude information determined by the inertial navigation data are comprehensively considered, which jointly act on the determination of the expansion range of the road surrounding, to ensure that the expansion area can cover all areas where the autonomous vehicle can drive, and does not involve too sensitive geographical sites. Through algorithmic analysis of these data, a reasonable expansion distance can be obtained, which meets the safety requirements of autonomous driving and avoids leaking key geographical information.

[0146] In a feasible implementation, step S502 can include: extracting a road image based on the image data, and performing image perspective inverse transformation based on vanishing points on the road image to obtain a corrected road image; pre-processing the point cloud data through a radius filtering algorithm, and extracting road surface point cloud data based on the pre-processed point cloud data; extracting road traffic markings based on the corrected road image and the road surface point cloud data; extracting vehicle speed data based on the inertial navigation data, and constructing a speed-reaction time-environment regression model based on the vehicle speed data and corresponding safety distance requirements; determining a reference distance matrix of road surrounding outward expansion based on the speed-reaction time-environment regression model; and determining a road surrounding expansion range based on the road traffic markings and the reference distance matrix of road surrounding outward expansion.

[0147] It should be noted that the image perspective inverse transformation based on vanishing points restores the spatial information in the image by determining the vanishing points in the image, i.e., the points where all parallel lines converge in the distance, so that the road image that has been distorted due to perspective effects can be corrected. The radius filtering algorithm is a pre-processing method for point cloud data, which retains points within a certain range around each point by setting an appropriate radius, removes noise points and outliers, and improves the accuracy and reliability of the point cloud data. Extracting road surface point cloud data based on pre-processed point cloud data is to analyze the spatial distribution characteristics of the point cloud data through algorithm, and identify the point cloud data belonging to the road surface.

[0148] It can be understood that by combining the corrected road image and the road surface point cloud data to extract the road traffic markings, the complementary advantages of image and point cloud data can be fully utilized, and the accuracy and robustness of road traffic marking extraction can be improved. Road traffic markings are an important basis for path planning and driving decision-making of autonomous vehicles, and accurate road traffic marking information helps autonomous vehicles correctly recognize road rules and improve driving safety and efficiency.

[0149] In a specific implementation, different vehicle speeds are extracted according to inertial navigation type data, and a speed-reaction time-environment regression model is established in combination with the safety distance requirements extended under different braking reaction time, different weather, road conditions and the like. The model shows that the buffer distance is positively correlated with the vehicle speed, positively correlated with the reaction braking time, and strongly correlated with the weather, road flatness and the like. The speed-reaction time-environment regression model is as follows:

[0150]

[0151] wherein, is the required safety buffer distance, that is, the safety distance requirement, v is the vehicle speed, is the braking reaction time, is the comprehensive friction coefficient, which is affected by the weather w and the road condition r, and g is the gravitational acceleration.

[0152] According to the speed-reaction time-environment regression model, a reference distance matrix of the road periphery expansion under different conditions is formed through statistical analysis. A plurality of road test auxiliary positioning elements are selected as sample objects according to road traffic markings, the distance distribution from the road edge is counted, and the road periphery expansion reference range is finally determined. Taking the roads in Guangzhou as an example, 36 road test auxiliary positioning elements are selected as sample objects in combination with the construction conditions of the municipal roads in Guangzhou, the spatial relationship of the surrounding traffic facilities and geographical sites, such as the Xingdao North Ring Road point (a simple urban road), the University City Outer Ring West Road point (a complex urban road) and the Nansha Port Expressway point (a high-speed expressway), the distance distribution from the road edge is counted, and 15 meters is finally selected as the road periphery expansion reference range in Guangzhou. Among them, the construction condition selection rule is that the road should not be under construction; the surrounding facility selection rule is that there are temporary parking locations and roadside objects such as support poles, tree trunks and lamp poles around the road, and the expansion range should include the roadside objects.

[0153] As shown in Table 1, Table 1 is a data table of roadside buffer reference distance, which includes the roadside buffer reference distance under different speed limits and different upper limits of vehicle sensor reaction time, as well as the corresponding reference recommended buffer distance. The recommended reference is 0.5 seconds as the upper limit of the standard reaction time. For example, when the speed limit is 80 km / h and the upper limit of the vehicle sensor reaction time is 0.5 s, the roadside buffer reference distance is 15 meters, and the reference recommended buffer distance is 16 meters. The buffer interval of 15 meters can include most auxiliary positioning element characteristic objects, therefore, 15 meters is selected as the road periphery expansion range in Guangzhou.

[0154] Table 1

[0155]

[0156] Step S503: According to the point cloud type data, the trajectory type data and the composition type data, the relative height demand audit is carried out, and the relative height grading information of the map element is obtained.

[0157] It should be noted that in the process of relative height demand audit, the object three-dimensional shape information provided by the point cloud type data, the vehicle motion path information recorded by the trajectory type data, and the absolute coordinate information integrated by the composition type data are comprehensively considered. These information jointly act on the determination of the relative height of the map element. Through the relative height information, the safe driving of the autonomous vehicle under different road conditions can be guaranteed, and the problems of data redundancy and privacy leakage caused by too detailed height information can be avoided. The map elements are divided into different grades according to their relative heights, so as to determine the obstacle recognition accuracy, so as to realize the relative height demand audit.

[0158] In a feasible implementation, step S503 can include: constructing a three-dimensional scene model based on the point cloud type data, and performing height slicing processing on the three-dimensional scene model to obtain slice layers of different heights; determining the moving track of the moving object according to the trajectory type data, and marking the moving track in the three-dimensional scene model; determining the position and shape of the map element based on the composition type data, and marking the map element in the three-dimensional scene model according to the position and shape; analyzing the vertical spatial relationship between the map element and the moving track in the slice layer to obtain the relative height information of the map element and the moving track; respectively establishing a discrete object model and a continuous field model according to the continuity of the relative height information; performing grading error analysis on the relative height information of the discrete object model to obtain grading fineness information; performing slope curvature decomposition on the relative height information of the continuous field model to obtain spatial grid density information; determining the relative height grading information of the map element according to the grading fineness information and the spatial grid density information.

[0159] It should be noted that the three-dimensional scene model constructed based on the point cloud type data can truly reflect the spatial characteristics of the road and its surrounding environment. Height slicing processing is to cut the three-dimensional scene model according to different heights to obtain a series of two-dimensional slice layers. These layers show the spatial information at different heights. The marking of the moving track recorded by the trajectory type data in the three-dimensional scene model helps to understand the motion path of the moving object, such as the autonomous vehicle, in space. The position and shape information of the map element provided by the composition type data marks the key elements such as roads, buildings and trees in the three-dimensional scene model. By analyzing the vertical spatial relationship between the map element and the moving track in the slice layer, the relative height information between them can be obtained, which is an important basis for the autonomous vehicle to make path planning and obstacle avoidance decisions.

[0160] The discrete object model is suitable for representing those map elements which are relatively independent in space and have highly discontinuous information, such as traffic signal lights, traffic signs, etc., which are represented by points, lines and surfaces at specific positions. By performing error analysis on the relative height information of such elements, the appropriate grading fineness, i.e., the difference between adjacent height grades, can be determined to ensure that the autonomous vehicle can accurately identify and avoid these obstacles. The continuous field model is suitable for representing those map elements which are continuously distributed in space and have relatively smooth height information, such as roads, bridges, etc. By performing slope curvature decomposition on the relative height information of such elements, the spatial grid density information, i.e., the parameter describing the speed of change of height information in space, can be obtained, which helps the autonomous vehicle to maintain safe driving under different road conditions.

[0161] In a specific implementation, according to the continuity of the relative height, discrete object models and continuous field models are established, respectively. The relative height of the discrete object model is expressed by grading, and the main review content is the grading fineness. The continuous field model mainly uses the way of slope and curvature decomposition of height difference, and the main review content is the fineness of two-dimensional plane grid unit and decomposition based on slope and curvature.

[0162] For traffic signal lights, road signs and other elements belonging to the discrete object model, 0.2m, 0.3m, 0.4m, 0.5m, 0.6m and 1.0m error analysis is performed, and the sample capacity is 288. Through field test verification, the correct recognition rate is determined, and the grading fineness is 0.2m and 0.5m, respectively.

[0163] As shown in Table 2, Table 2 is a relative height grading table of traffic signal lights, which includes the model of traffic signal lights, the number of samples, and the number of misjudgments and the misjudgment ratio under different grading. The sample number of 400 type red and green light is 126, and the sample number of 300 type red and green light is 90. The number of misjudgments and the misjudgment ratio under 0.2m grading are both 0.

[0164] Table 2

[0165]

[0166] As shown in Table 3, Table 3 is a relative height grading table of traffic signs, which includes the shape of traffic signs, the number of samples, and the number of misjudgments and the misjudgment ratio under different grading. The sample number of circular traffic signs is 104, and the sample number of square traffic signs is 112. The number of misjudgments and the misjudgment ratio under 0.5m grading are both 0.

[0167] Table 3

[0168]

[0169] For the continuous field model relative height information decomposition expression, the slope curvature is used to decompose the relative height. A spatial grid is used to express the height information, and the grid density that meets both compliance and autonomous driving requirements is demonstrated. Two kinds of fineness, 0.5m x 0.5m and 1m x 1m, are tested and verified. The road shoulder and the test points within 10 meters of the road shoulder are selected for case-by-case verification. It is found that the 0.5m x 0.5m density grid map can accurately restore the relative relief around the vehicle and meet the safety requirements of autonomous driving. The 1m x 1m density grid map is not accurate enough and affects the obstacle recognition of autonomous driving, which further affects the safety of the vehicle and the safety of the road traffic. Therefore, 0.5m x 0.5m unit grid is used.

[0170] As shown in Table 4, Table 4 is the error verification data after grid sparsification processing. The table includes the test range, the total number of sampling points, and the number and proportion of points in different error ranges. For example, the total number of sampling points in the road shoulder is 58461238, the number of points with an error exceeding 20cm is 89072, and the proportion is 0.15%. The number of points with an error exceeding 30cm is 69107, and the proportion is 0.12%.

[0171] Table 4

[0172]

[0173] Step S504: performing auxiliary positioning element compliance expression auditing according to the composition data to obtain relative height grading information of the auxiliary positioning element.

[0174] It should be noted that the auxiliary positioning element compliance expression auditing refers to grading the relative height of the auxiliary positioning element to ensure its accurate expression and compliance in the autonomous driving map. The positioning accuracy can be determined according to the relative height grading information of the auxiliary positioning element, and the compliance expression auditing of the auxiliary positioning element is realized.

[0175] In a feasible implementation, step S504 can include: extracting road feature points, intersection feature points and interest points based on the composition data; constructing an auxiliary positioning element network based on the position information of the road feature points, intersection feature points and interest points; performing relative height analysis based on the auxiliary positioning element network to obtain relative height information of the auxiliary positioning element; and performing grading processing according to the relative height information of the auxiliary positioning element to obtain relative height grading information of the auxiliary positioning element.

[0176] It should be noted that the composition data usually contains high-precision road geometry information, intersection layout, point of interest (POI) location and other key information. Road feature points mainly reflect the direction and curvature of the road, which is the basis for building a road network. Intersection feature points describe the layout of road intersections in detail, including intersection angles, lane allocation, etc., which is crucial for autonomous vehicles to navigate at intersections. POIs cover important landmarks and service facilities along the way, providing rich navigation information for autonomous vehicles.

[0177] The construction of the auxiliary positioning element network aims to organically integrate these discrete road feature points, intersection feature points and POIs to form a coherent and accurate positioning reference framework. This network not only improves the accuracy and reliability of positioning information, but also provides a more detailed description of the road environment for autonomous vehicles, helping to improve the vehicle's path planning and decision-making capabilities. Relative height analysis is based on this network to further extract the relative height information of each auxiliary positioning element.

[0178] Finally, according to the relative height information of the auxiliary positioning elements, the grading processing is carried out, which is similar to the height grading of the discrete correspondence model, and this embodiment will not be repeated here. According to the test statistical results, 0.2 meters is also selected as the grading interval, and the auxiliary positioning elements only express the collective form information, and do not represent semantic content.

[0179] Step S505: Comprehensive evaluation according to the road surrounding expansion range, the relative height grading information of the map elements and the relative height grading information of the auxiliary positioning elements to obtain the review result of the high-precision map.

[0180] It should be noted that the road surrounding expansion range can provide sufficient safety buffer area for autonomous vehicles to ensure that the vehicle can identify and respond to obstacles in the road surrounding area in a timely manner during driving. The relative height grading information of the map elements is closely related to the accuracy of obstacle recognition, and reasonable grading can improve the ability of autonomous vehicles to recognize obstacles. The relative height grading information of the auxiliary positioning elements helps to ensure that autonomous vehicles can accurately position in complex road environments. By integrating these information, a comprehensive evaluation of the high-precision map can be made to ensure that it meets the safety and accuracy requirements of autonomous driving. During the evaluation process, whether the road surrounding expansion range is sufficient, whether the relative height grading of the map elements is reasonable, and whether the relative height grading of the auxiliary positioning elements is accurate will be considered. Only when these elements meet the safety and accuracy requirements of autonomous driving, the high-precision map can be determined to be qualified.

[0181] In an implementable embodiment, step S505 can include determining a road expansion range deviation rate based on the preset road expansion range compliance standard for the road perimeter expansion range; performing obstacle identification simulation and prediction based on the relative height grading information of the map elements to obtain an obstacle identification accuracy rate; performing positioning accuracy simulation and prediction according to the relative height grading information of the auxiliary positioning elements to obtain a positioning accuracy rate; performing comprehensive evaluation according to the road expansion range deviation rate, the obstacle identification accuracy rate, and the positioning accuracy rate to obtain a comprehensive evaluation score of the high-precision map; and comparing the comprehensive evaluation score with a preset map review pass score line to obtain a map review result of the high-precision map.

[0182] It should be noted that the preset road expansion range compliance standard is generally determined based on traffic regulations, road design guidelines, and safety requirements of autonomous vehicles. The road expansion range deviation rate is used to measure the difference between the actual road expansion range and the preset standard, which helps to evaluate the accuracy of the high-precision map in expressing the road safety buffer area.

[0183] Obstacle identification simulation and prediction is performed by simulating various obstacle scenarios that an autonomous vehicle may encounter during driving, and using the relative height grading information of the map elements to evaluate the vehicle's ability to identify these obstacles. This process can reveal the rationality of the map element grading and its impact on autonomous driving safety.

[0184] Positioning accuracy simulation and prediction is based on the relative height grading information of the auxiliary positioning elements to simulate the positioning process of the autonomous vehicle in complex road environments, in order to evaluate the positioning accuracy of the vehicle. This step helps to ensure that the auxiliary positioning elements in the high-precision map can provide reliable positioning reference in actual application.

[0185] The comprehensive evaluation score is the result of comprehensive consideration of the road expansion range deviation rate, the obstacle identification accuracy rate, and the positioning accuracy rate, which reflects the overall performance of the high-precision map in terms of safety, accuracy, and compliance. Comparing the comprehensive evaluation score with the preset map review pass score line can directly determine whether the high-precision map meets the application requirements of autonomous driving, thereby obtaining the map review result.

[0186] In this embodiment, by comprehensively auditing the spatial range expansion, the relative height requirement, and the compliance expression of auxiliary positioning elements, and then comparing the comprehensive evaluation score with the preset map review pass score line, the quality and compliance of the high-precision map can be comprehensively and accurately evaluated, effectively improving the efficiency and accuracy of map review.

[0187] It should be noted that the above examples are only for understanding the present application and do not constitute a limitation on the high-precision map data processing method of the present application. More forms of simple transformation based on this technical concept are within the protection scope of the present application.

[0188] The present application also provides a high-precision map data processing device, please refer to Figure 3 The high-precision map data processing device comprises:

[0189] The acquisition module 10 is configured to acquire geographic information data of a high-precision map.

[0190] The standardization module 20 is configured to standardize the geographic information data based on a spatiotemporal semantic perception adaptive standardization engine to obtain standardized geographic information data.

[0191] The classification and grading module 30 is configured to perform horizontal classification on the standardized geographic information data based on a multi-modal feature extraction automatic clustering algorithm and vertical grading based on an influence degree evaluation knowledge graph reasoning algorithm to obtain a hierarchical data set, wherein the hierarchical data set comprises a plurality of data sets of different categories, each category of data set comprises a plurality of data of different levels, different categories of data sets comprise image data sets, point cloud data sets, trajectory data sets, inertial navigation data sets and composition data sets, and different levels of data include core data, important data and general data.

[0192] The encryption and desensitization module 40 is configured to perform data encryption and desensitization on the hierarchical data set to obtain an encrypted and desensitized data set.

[0193] The review module 50 is configured to perform spatial range expansion review, relative height requirement review and auxiliary positioning element compliance expression review on the high-precision map according to the encrypted and desensitized data set to obtain a map review result of the high-precision map.

[0194] The publishing module 60 is configured to optimize the high-precision map data according to the encrypted and desensitized data set if the map review result is a high-precision map review pass, and publish the optimized high-precision map data to a map service platform.

[0195] The high-precision map data processing device provided by the application adopts the high-precision map data processing method in the above embodiment, and can solve the technical problems of low efficiency, poor accuracy and difficulty in ensuring the integrity and security of high-precision map data of the traditional map review method. Compared with the prior art, the beneficial effects of the high-precision map data processing device provided by the application are the same as those of the high-precision map data processing method provided by the above embodiment, and the other technical features of the high-precision map data processing device are the same as those disclosed in the above embodiment method, and will not be repeated here.

[0196] The above is only part of the embodiments of the application, and does not limit the patent scope of the application. Any equivalent structural transformation, direct / indirect application in other related technical fields, or direct / indirect application in other related technical fields under the technical concept of the application, including in the patent protection scope of the application.

Claims

1. A high-precision map data processing method, characterized in that, The method includes: Acquire geographic information data for high-precision maps; An adaptive standardization engine based on spatiotemporal semantic awareness standardizes the geographic information data to obtain standardized geographic information data. The standardized geographic information data is horizontally classified based on an automatic clustering algorithm using multimodal feature extraction and vertically graded based on a knowledge graph reasoning algorithm using influence assessment to obtain a hierarchical data set. The hierarchical data set includes multiple data sets of different categories, and each category of data set includes multiple data sets of different levels. The different categories of data sets include image data sets, point cloud data sets, trajectory data sets, inertial navigation data sets, and mapping data sets. The different levels of data include core data, important data, and general data. The hierarchical data set is encrypted and desensitized to obtain the encrypted and desensitized data set. Based on the encrypted and desensitized data set, the high-precision map is reviewed for spatial range expansion, relative height requirements, and compliance of auxiliary positioning elements, resulting in the map review results. If the high-precision map is approved in the review result, the high-precision map data is optimized based on the encrypted and desensitized data set, and the optimized high-precision map data is published to the map service platform. The spatial scope expansion review specifically includes: Road images are extracted from image data, and the road images are subjected to inverse perspective transformation based on vanishing points to obtain corrected road images. Point cloud data is preprocessed using a radius filtering algorithm, and road surface point cloud data is extracted based on the preprocessed point cloud data. Road traffic markings are extracted based on the corrected road image and the road surface point cloud data; Vehicle speed data is extracted based on inertial navigation data, and a speed-reaction time-environment regression model is constructed based on the vehicle speed data and the corresponding safety distance requirements. The reference distance matrix for the outward expansion of the road perimeter is determined based on the speed-reaction time-environment regression model. The extent of the road perimeter is determined based on the road traffic markings and the reference distance matrix extending outward from the road perimeter.

2. The method as described in claim 1, characterized in that, The adaptive standardization engine based on spatiotemporal semantic awareness standardizes the geographic information data to obtain standardized geographic information data, including: A unified spatiotemporal reference framework is constructed based on the global dynamic reference frame, the international Earth reference frame, and regional crustal deformation models. Based on the unified spatiotemporal reference framework, the geographic information data is spatiotemporally and semantically aligned to obtain spatiotemporally aligned data. The spatiotemporally aligned data is transformed using a coordinate system transformation algorithm to obtain the coordinate-transformed data. Construct a geographic element ontology knowledge base, and determine the entity standard attribute structure based on the geographic element ontology knowledge base; The data after coordinate system transformation is normalized according to the entity standard attribute structure to obtain the data after attribute semantic normalization. The data after the attribute semantics are normalized is encoded according to a preset hybrid encoding format to obtain standardized geographic information data.

3. The method as described in claim 1, characterized in that, The standardized geographic information data is horizontally classified using an automatic clustering algorithm based on multimodal feature extraction and vertically hierarchically classified using a knowledge graph reasoning algorithm based on influence assessment, resulting in a hierarchical data set, including: Feature extraction is performed on the standardized geographic information data based on a multimodal embedding model to obtain feature vectors for various types of data; The feature vectors of the various types of data are mapped to a unified semantic embedding space to obtain a unified semantic embedding vector; The unified semantic embedding vectors are clustered based on the HDBSCAN clustering algorithm to obtain data sets of different categories. Construct a high-precision map risk knowledge graph, wherein the nodes in the high-precision map risk knowledge graph include geographic elements, functional impacts, and security levels; Based on the high-precision map risk knowledge graph, the impact assessment algorithm is used to assess the impact of data in the different categories of datasets to obtain decision influence scores. Based on the decision influence score, a knowledge graph reasoning path is constructed, and the knowledge graph reasoning path is used to vertically classify the different categories of data sets to obtain a hierarchical data set.

4. The method as described in claim 1, characterized in that, The step of encrypting and de-identifying the hierarchical data set to obtain the encrypted and de-identified data set includes: The hierarchical data set is desensitized using a classification desensitization strategy to obtain different categories of data sets after desensitization. The classification desensitization strategies include a semantic segmentation-guided local fuzzy desensitization strategy for image data sets, an attribute differential privacy desensitization strategy for point cloud data sets, a spatiotemporal randomization desensitization strategy for trajectory data sets, a noise addition desensitization strategy for inertial navigation data sets, and a hash obfuscation desensitization strategy for mapping data sets. The data at different levels in the different categories of the de-identified data sets are encrypted using a hierarchical encryption strategy to obtain hierarchically encrypted data sets.

5. The method as described in claim 1, characterized in that, The process of reviewing the high-precision map based on the encrypted and desensitized data set, including spatial extent expansion review, relative height requirement review, and compliance review of auxiliary positioning element expression, yields the high-precision map review result, including: Based on the encrypted and desensitized data set, image data, point cloud data, trajectory data, inertial navigation data, and composition data are determined; Based on the image data, point cloud data, and inertial navigation data, a spatial range expansion review is performed to obtain the extended range around the road. Based on the point cloud data, trajectory data, and composition data, the relative height requirement is reviewed to obtain the relative height classification information of map elements; Based on the aforementioned compositional data, the compliance review of auxiliary positioning elements is conducted to obtain the relative height classification information of the auxiliary positioning elements; The high-precision map review result is obtained by comprehensively evaluating the surrounding extension range of the road, the relative height classification information of map elements, and the relative height classification information of auxiliary positioning elements.

6. The method as described in claim 5, characterized in that, The step of reviewing relative height requirements based on the point cloud data, trajectory data, and composition data to obtain relative height classification information for map elements includes: A 3D scene model is constructed based on point cloud data, and the 3D scene model is then sliced ​​at different heights to obtain slice layers at different heights. The movement trajectory of the moving object is determined based on the trajectory data, and the movement trajectory is marked in the three-dimensional scene model; The location and shape of map elements are determined based on the composition data, and the map elements are marked in the three-dimensional scene model according to the location and shape. Analyze the vertical spatial relationship between map features and movement trajectories in the slice layer to obtain the relative height information between map features and movement trajectories; Based on the degree of continuity of the relative height information, a discrete object model and a continuous field model are established respectively; The relative height information of the discrete object model is subjected to a grading error analysis to obtain grading detail information; The relative height information of the continuous field model is decomposed by slope curvature to obtain spatial grid density information; The relative height classification information of map features is determined based on the classification detail information and spatial grid density information.

7. The method as described in claim 5, characterized in that, The step of reviewing the compliance of auxiliary positioning elements based on the composition data to obtain the relative height classification information of the auxiliary positioning elements includes: Extracting road feature points, intersection feature points, and points of interest from graph-based data; Based on the location information of road feature points, intersection feature points, and points of interest, an auxiliary positioning element network is constructed; Relative height analysis is performed based on the aforementioned auxiliary positioning element network to obtain the relative height information of the auxiliary positioning elements; The relative height information of the auxiliary positioning elements is processed to obtain the relative height classification information of the auxiliary positioning elements.

8. The method as described in claim 5, characterized in that, The process of comprehensively evaluating the road's surrounding extension range, the relative height classification information of map features, and the relative height classification information of auxiliary positioning features to obtain the high-precision map review result includes: The deviation rate of the road extension range is determined based on the preset road extension range compliance standards; Obstacle recognition simulation and prediction are performed based on the relative height classification information of the map elements to obtain the obstacle recognition accuracy. The positioning accuracy is simulated and predicted based on the relative height classification information of the auxiliary positioning elements to obtain the positioning accuracy rate; A comprehensive evaluation score for the high-precision map is obtained by comprehensively evaluating the road extension range deviation rate, the obstacle recognition accuracy rate, and the positioning accuracy rate. The comprehensive evaluation score is compared with the preset map review passing score to obtain the map review result of the high-precision map.

9. A high-precision map data processing device, characterized in that, The high-precision map data processing device includes: The acquisition module is used to acquire geographic information data from high-precision maps; The standardization module is used to standardize the geographic information data based on the spatiotemporal semantic awareness adaptive standardization engine to obtain standardized geographic information data. The classification and grading module is used to perform horizontal classification of the standardized geographic information data based on an automatic clustering algorithm using multimodal feature extraction and vertical grading based on a knowledge graph inference algorithm using influence assessment, resulting in a hierarchical data set. The hierarchical data set includes multiple data sets of different categories, and each category of data set includes multiple data sets of different levels. The different categories of data sets include image data sets, point cloud data sets, trajectory data sets, inertial navigation data sets, and mapping data sets. The different levels of data include core data, important data, and general data. An encryption and desensitization module is used to encrypt and desensitize the hierarchical data set to obtain an encrypted and desensitized data set. The review module is used to review the spatial range expansion, relative height requirements, and compliance of auxiliary positioning elements of the high-precision map based on the encrypted and desensitized data set, and obtain the review result of the high-precision map. The publishing module is used to optimize the high-precision map data based on the encrypted and desensitized data set when the map review result is that the high-precision map has passed the review, and then publish the optimized high-precision map data to the map service platform. The spatial extent expansion review specifically includes: extracting road images from image-based data and performing an inverse perspective transformation based on vanishing points to obtain a corrected road image; preprocessing point cloud data using a radius filtering algorithm and extracting road surface point cloud data from the preprocessed point cloud data; extracting road traffic markings based on the corrected road image and the road surface point cloud data; extracting vehicle speed data from inertial navigation data and constructing a speed-reaction time-environment regression model based on the vehicle speed data and the corresponding safety distance requirements; determining a reference distance matrix for the road perimeter expansion based on the speed-reaction time-environment regression model; and determining the road perimeter expansion range based on the road traffic markings and the reference distance matrix for the road perimeter expansion.

Images