Vector map processing method, system and electronic device

Abstract

This application provides a method, system, and electronic device for processing vector maps. The method includes: projecting at least one vector feature from a vector map of a vehicle's environment onto a plane containing image data of the environment to obtain a projected feature corresponding to the vector feature. The vector feature represents the geometric structure of an environmental object, and the image content of the image data includes the environmental object. The method also includes performing an offset operation on the projected feature and generating at least one candidate checkpoint based on the offset projected feature. The candidate checkpoint characterizes the permissible offset degree of the vector feature when projected onto the plane. The method further includes checking the candidate checkpoint using its category attributes and the category attributes of the vector feature to obtain a check result. Finally, the method generates label information for the environmental object based on the check result. This application solves the technical problem of poor processing performance in vector maps.

Description

Technical Field

[0001] This application relates to the field of autonomous driving perception technology, and more specifically, to a vector map processing method, system, and electronic device. Background Technology

[0002] Currently, when training large-scale perception models for vehicles, it is necessary to handle the occlusion consistency between vector maps (also known as vector ground truth maps) and image data. Related technologies primarily rely on manual annotation or geometric projection-based methods to address occlusion consistency. However, manual annotation is inefficient and costly, making it difficult to meet large-scale demands; geometric projection-based methods can only identify occlusion between map features, failing to recognize occlusion caused by non-map features, resulting in poor robustness and low recall. Therefore, the technical problem of poor vector map processing performance remains.

[0003] There is currently no good solution to the above problems. Summary of the Invention

[0004] This application provides a method, system, and electronic device for processing vector maps, to at least solve the technical problem of poor processing results for vector maps.

[0005] According to one aspect of the embodiments of this application, a method for processing a vector map is provided. The method may include: projecting at least one vector feature from a vector map of the environment in which a vehicle is located onto a plane containing image data of the environment to obtain a projected feature corresponding to the vector feature, wherein the vector feature represents the geometric structure of an environmental object, and the image content of the image data includes the environmental object; performing an offset operation on the projected feature, and generating at least one candidate checkpoint corresponding to the offset projected feature based on the offset projected feature, wherein the candidate checkpoint characterizes the degree of offset allowed for the vector feature when projected onto the plane; checking the candidate checkpoint using the category attributes of the candidate checkpoint and the category attributes of the vector feature to obtain a check result, wherein the check result represents the visibility of the candidate checkpoint; and generating label information for the environmental object based on the check result, wherein the label information represents the degree to which the environmental object is occluded.

[0006] Optionally, at least one vector feature in the vector map of the vehicle's environment is projected onto the plane containing the image data of the environment to obtain the projection feature corresponding to the vector feature, including: obtaining the position information corresponding to the multiple vector feature points to be constituted as vector features; projecting the multiple vector feature points onto the plane according to the position information to obtain multiple projection points, wherein the multiple projection points correspond one-to-one with the multiple vector feature points; and constructing the projection feature from the multiple projection points.

[0007] Optionally, the location information includes first location information, second location information, and third location information. The first location information indicates the position of the vector feature point in the world coordinate system, the second location information indicates the position of the vector feature point in the camera coordinate system of the vehicle, and the third location information indicates the position of the projection point on the plane. Based on the location information, multiple feature points are projected onto the plane to obtain multiple projection points, including: transforming multiple vector feature points from the world coordinate system to the camera coordinate system based on the first and second location information; and projecting multiple vector feature points from the camera coordinate system onto the plane based on the second and third location information to obtain multiple projection points.

[0008] Optionally, obtaining projection elements composed of multiple projection points includes: using the image view of the image data to detect the multiple projection points respectively, and obtaining multiple detection results corresponding to the multiple projection points; based on the multiple detection results corresponding to the multiple projection points, determining multiple target projection points from the multiple projection points, wherein the detection results corresponding to the target projection points are used to indicate that the target projection points are in the image view; and constructing projection elements from the multiple target projection points.

[0009] Optionally, the projection feature consists of multiple projection points. The offset operation on the projection feature includes: determining the offset direction of the offset operation to be performed on the projection feature; and performing the offset operation on the multiple projection points according to the offset direction to obtain the offset projection points.

[0010] Optionally, the vector feature consists of multiple vector feature points. Obtaining the offset direction for the offset operation to be performed on the projected feature includes: determining the line segment direction of the line segment formed by two adjacent vector feature points in the vector feature; determining the offset direction as the horizontal axis direction in response to the line segment direction being vertical; and determining the offset direction as the vertical axis direction in response to the line segment direction being horizontal.

[0011] Optionally, according to the offset direction, offset operations are performed on multiple projection points to obtain offset projection points, including: determining the allowable offset amount for each of the multiple projection points to be constituted as projection features based on the category attribute of the vector features; performing offset operations on the projection points according to the offset direction and offset amount to generate offset projection points; and / or, based on the offset projection features, generating at least one candidate checkpoint corresponding to the offset projection features, including: determining the offset projection point as a candidate checkpoint.

[0012] Optionally, the candidate checkpoints are checked using the category attributes of the candidate checkpoints and the category attributes of the vector features to obtain the check results, including: in response to the fact that the category attributes of the candidate checkpoints are the same as the category attributes of the vector features, the check result is determined to indicate that the visibility of the candidate checkpoints is greater than the visibility threshold; in response to the fact that the category attributes of the candidate checkpoints are different from the category attributes of the vector features, the check result is determined to indicate that the visibility of the candidate checkpoints is less than or equal to the visibility threshold.

[0013] Optionally, the method further includes: determining a set of category attributes for vector features based on the type information of the vector features, wherein the set of category attributes includes the category attributes of at least one vector feature.

[0014] Optionally, based on the inspection results, generating label information for environmental objects includes: determining the visibility ratio of vector features based on the inspection results, wherein the visibility ratio represents the proportion of the visible portion of a vector feature to the total portion of the vector feature; determining label information as first label information in response to a visibility ratio greater than a first visibility ratio threshold; determining label information as second label information in response to a visibility ratio less than or equal to the first visibility ratio threshold and greater than a second visibility ratio threshold, wherein the occlusion degree of the environmental object indicated by the second label information is greater than the occlusion degree of the environmental object indicated by the first label information, and the second visibility ratio threshold is less than the first visibility ratio threshold; and determining label information as third label information in response to a visibility ratio less than or equal to the second visibility ratio threshold, wherein the occlusion degree of the environmental object indicated by the third label information is greater than the occlusion degree of the environmental object indicated by the second label information.

[0015] Optionally, based on the inspection results, the visibility ratio of vector features is determined, including: based on the inspection results, determining the number of visible points corresponding to at least one candidate inspection point and the total number of points, wherein the number of visible points is used to represent the number of candidate inspection points whose visibility is greater than the visibility threshold, and the total number of points is used to represent the number of candidate inspection points; the ratio of the number of visible points to the total number of points is determined as the visibility ratio.

[0016] According to another aspect of the embodiments of this application, a vector map processing system is also provided. This system may include: a data input module for acquiring a vector map of the environment in which a vehicle is located, and image data of the environment; a projection calculation module for projecting at least one vector feature from the vector map of the environment onto a plane containing the image data of the environment to obtain a projection feature corresponding to the vector feature, wherein the vector feature represents the geometric structure of an environmental object, and the image content of the image data includes the environmental object; an error tolerance module for performing an offset operation on the projection feature and generating at least one candidate checkpoint corresponding to the offset projection feature based on the offset projection feature, wherein the candidate checkpoint characterizes the degree of offset allowed for the vector feature when projected onto the plane; a consistency verification module for checking the candidate checkpoint using the category attributes of the candidate checkpoint and the category attributes of the vector feature to obtain a check result, wherein the check result represents the visibility of the candidate checkpoint; a visibility determination module for determining the visibility level of the vector feature based on the check result; and a label generation module for generating label information of the environmental object based on the visibility level, wherein the label information represents the degree to which the environmental object is occluded.

[0017] According to another aspect of the embodiments of this application, a vector map processing apparatus is also provided. The apparatus may include: a projection module, configured to project at least one vector feature from a vector map of the environment in which a vehicle is located onto a plane containing image data of the environment, to obtain a projection feature corresponding to the vector feature, wherein the vector feature represents the geometric structure of an environmental object in the environment, and the image content of the image data includes the environmental object; an offset module, configured to perform an offset operation on the projection feature and generate at least one candidate checkpoint corresponding to the offset projection feature based on the offset projection feature, wherein the candidate checkpoint characterizes the degree of offset of the vector feature when it is projected onto the plane; an inspection module, configured to inspect the candidate checkpoint using the category attributes of the candidate checkpoint and the category attributes of the vector feature, to obtain an inspection result, wherein the inspection result represents the visibility of the candidate checkpoint; and a generation module, configured to generate label information of the environmental object based on the inspection result, wherein the label information represents the degree of occlusion of the environmental object.

[0018] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0019] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0020] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0021] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the methods in various embodiments of this application.

[0022] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0023] In this embodiment, at least one vector feature from the vector map of the vehicle's environment can be projected onto the plane containing the image data, resulting in a projected feature corresponding to the vector feature. An offset operation can be performed on the projected feature, and at least one candidate checkpoint can be generated based on the offset projected feature. The visibility of the candidate checkpoint can be checked using the category attributes of the candidate checkpoint and the category attributes of the vector feature, yielding a check result. Based on the check result, label information indicating whether an environmental object is occluded can be generated. In other words, in this embodiment, by projecting vector features onto the image plane and generating candidate checkpoints based on a projection error tolerance mechanism, the category attributes of the candidate points are queried using image segmentation results. By comparing the consistency between the category attributes of the candidate checkpoints and the category attributes of the vector features, the visibility is quantified, and label information is generated. This method replaces inefficient manual annotation and geometric projection methods that can only identify occlusion between map features, achieving efficient and robust automatic visibility recognition of vector maps. This solves the technical problem of poor vector map processing performance and improves the overall processing effect of vector maps. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of an application scenario for vector map processing according to an embodiment of this application; Figure 2This is a flowchart of a vector map processing method according to an embodiment of this application; Figure 3 This is a flowchart of a vector ground map occlusion recognition method based on image segmentation according to an embodiment of this application; Figure 4 This is a schematic diagram of a vector truth map occlusion recognition system based on image segmentation according to an embodiment of this application; Figure 5 This is a schematic diagram of a vector map processing system according to an embodiment of this application; Figure 6 This is a schematic diagram of a vector map processing apparatus according to an embodiment of this application; Figure 7 This is a schematic diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] Figure 1 This is a schematic diagram illustrating an application scenario of vector map processing according to an embodiment of this application, such as... Figure 1As shown, the scenario described above may include a terminal device 10, a network 20, and a server 30. The terminal device 10 can be used to acquire vector map and image data input by the user indicating whether processing of the vector map is required. The terminal device can be an interactive interface for interacting with the server. The vector map and image data can be sent to the server 30 via the network 20. At this point, the server 30 needs to execute steps S102 to S108 to implement the vector map processing procedure.

[0028] The following steps can be performed by server 30: Step S102, project at least one vector feature in the vector map of the vehicle's environment onto the plane where the image data of the environment is located, to obtain the projection feature corresponding to the vector feature; Step S104, perform an offset operation on the projection feature, and generate at least one candidate checkpoint corresponding to the offset projection feature based on the offset projection feature; Step S106, use the category attributes of the candidate checkpoint and the category attributes of the vector feature to check the candidate checkpoint and obtain the check result; Step S108, generate the label information of the environment object based on the check result.

[0029] In this embodiment, through steps S102 to S108, vector features are projected onto the image plane, and candidate checkpoints are generated based on a projection error tolerance mechanism. The category attributes of the candidate points are then queried using the image segmentation results. By comparing the consistency between the category attributes of the candidate checkpoints and the category attributes of the vector features, the visibility is quantified and label information is generated. This method replaces inefficient manual annotation and geometric projection methods that can only identify occlusion between map features, achieving efficient and robust automatic visibility recognition of vector maps. This solves the technical problem of poor vector map processing performance and improves the overall processing effect of vector maps.

[0030] According to an embodiment of this application, a method for processing vector maps is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] This embodiment provides a method for processing vector maps. Figure 2 This is a flowchart of a vector map processing method according to an embodiment of this application, such as... Figure 2 As shown, the method may include the following steps.

[0032] Step S202: Project at least one vector feature from the vector map of the vehicle's environment onto the plane containing the image data of the environment to obtain the projection feature corresponding to the vector feature.

[0033] In the technical solution provided by step S202 of the embodiments of this application, vector elements can be used to represent the geometric structure of environmental objects in the environment. The image content of the image data includes environmental objects. The vector map mentioned above can refer to map data containing precise location information of road geometric elements in the environment where the vehicle is located, and can also be called a vector ground truth map or vector ground truth map data. The coordinate system of the vector map can be the world coordinate system. The vector elements mentioned above can refer to the basic data units that constitute the vector map, used to represent the geometric structure of environmental objects in the environment, and can also be called map elements, which may include, but are not limited to: lane line elements, curb elements, and ground marking elements. The environmental objects mentioned above can refer to physical entities or road components in the environment where the vehicle is located that are geometrically represented by vector elements, such as lane lines, curbs, ground markings, and other road infrastructure. The image data mentioned above can refer to the environmental visual information collected by the vehicle's onboard camera. The image data mentioned above can include the original red, green, and blue (RGB) images (also called the original images) and the corresponding pixel-level semantic segmentation results, wherein the original images can be used to record the visual appearance of environmental objects. The semantic segmentation results can be used to provide semantic classification labels for each pixel in the original images.

[0034] Optionally, the aforementioned projection element can refer to the two-dimensional geometric representation formed by mapping the vector elements in the vector map to the plane (also known as the two-dimensional plane) where the image data is located through camera intrinsic and extrinsic parameter transformation. In other words, it is the corresponding projection form of the vector elements in the image coordinate system of the image data.

[0035] In this embodiment, at least one vector feature in the vector map can be projected onto the plane containing the image data of the environment to obtain the projected feature corresponding to the vector feature. This embodiment proposes an implementation process for projecting at least one vector feature from the vector map of the vehicle's environment onto the plane containing the image data of the environment to obtain the projected feature corresponding to the vector feature. The above process may include three steps: data preparation, vector feature extraction, and coordinate projection, aiming to establish a geometric mapping relationship between three-dimensional space and two-dimensional space.

[0036] Optionally, during the acquisition of image data of the vehicle's environment, the data can be collected using the vehicle's onboard camera, including the original RGB image and the corresponding pixel-level semantic segmentation results. Simultaneously, the camera's intrinsic and extrinsic parameters can be acquired. The intrinsic parameters may include focal length and principal point coordinates, while the extrinsic parameters may include the rotation matrix and translation vector from the world coordinate system to the camera coordinate system.

[0037] Optionally, in the process of identifying and extracting vector features from a vector map, since the vector map is composed of multiple vector features, each representing the geometric structure of environmental objects, the identification process can include: traversing the vector map and parsing the sequence of three-dimensional (3D) coordinate points that constitute the vector features from the traversed vector map. These vector features may include, but are not limited to, lane line features, curb features, and ground marking features. For example, lane line features may include lane center lines, lane boundary lines, etc., composed of a series of vector feature points (also called 3D coordinate points); curb features may include curb stones, guardrails, etc., represented by a sequence of 3D coordinate points; and ground marking features may include pedestrian crossings, ground arrows, stop lines, etc., represented by polygons or point sequences. For each vector feature to be processed, the included 3D coordinate points are extracted to form a point set for subsequent projection calculations.

[0038] Optionally, in the process of projecting vector elements onto the plane containing image data to obtain the corresponding projected elements, for each 3D coordinate point in the vector element, the 3D coordinate point is transformed from the world coordinate system to the camera coordinate system of the vehicle camera using camera extrinsic parameters to obtain intermediate coordinates in the camera coordinate system. The intermediate coordinates in the camera coordinate system are then projected onto the image plane using camera intrinsic parameters, and the two-dimensional image coordinates of the intermediate coordinate points on the plane containing image data are calculated using perspective projection formulas. This transformation and projection process is repeated for each 3D coordinate point in the vector element, and the projected two-dimensional image coordinates are connected or combined according to the connection relationships between the original 3D coordinate points in the vector element to obtain the corresponding projected elements. The positional information of the vector element points can be the 3D coordinates of the location of the 3D coordinate point.

[0039] In this embodiment, the above method accurately maps a vector map in three-dimensional space onto a plane of two-dimensional image data, establishing a spatial correspondence between vector elements and projected pixels. This geometric projection method provides a foundation for subsequently verifying the visibility of vector elements using image semantic information.

[0040] Step S204: Perform an offset operation on the projected features, and generate at least one candidate checkpoint corresponding to the offset projected features based on the offset projected features.

[0041] In the technical solution provided by step S204 of this application embodiment, candidate checkpoints are used to characterize the degree of offset of vector features when they are projected onto a plane. These candidate checkpoints can refer to discrete spatial location points generated based on offset operations and used to verify the visibility of vector features. Candidate checkpoints can characterize the set of locations within the maximum allowable offset range for vector features when they are projected onto the plane containing the image data. Each candidate checkpoint corresponds to an offset-corrected 3D coordinate and is mapped to the image coordinate system through projection transformation. This is used to query a specific semantic category from the semantic segmentation results, thereby determining whether the location belongs to the expected visible semantic type.

[0042] Optionally, the aforementioned offset operation can refer to the displacement transformation of the coordinate position of the projected elements to compensate for projection deviations caused by calibration errors, positioning errors, and time synchronization errors of the vehicle-mounted camera. This offset operation can generate a series of displaced coordinate points in a specific direction (e.g., vertical line segments in the X direction, horizontal line segments in the Y direction) based on the directional characteristics of the line segments in the projected elements. By introducing a tolerance range, the offset operation allows projected points that were originally deviated from their actual visual position due to errors to cover the real observation area, thereby correcting the alignment deviation between the geometric projection and the actual content of the image.

[0043] In this embodiment, after projecting vector features onto the plane containing the image data to obtain corresponding projected features, an offset operation can be performed on the projected features to obtain offset projected features. At least one candidate checkpoint can be generated based on the offset projected features. This embodiment proposes an implementation process for performing an offset operation on projected features and generating candidate checkpoints. The above process aims to solve the problem of inconsistency between the projected position of vector features and the actual image observation position caused by camera calibration errors, positioning errors, and time synchronization errors. By introducing an offset tolerance mechanism, a series of candidate checkpoints are generated around the projected features to cover the possible true location range of the vector features in the image, providing robust sampling support for subsequent visibility verification based on semantic segmentation results.

[0044] Optionally, during the process of performing an offset operation on the projected features to correct geometric projection deviations, the direction of line segments within the projected features is determined. If the absolute value of the coordinate difference between two adjacent points in the Y-axis direction is greater than the absolute value of the coordinate difference in the X-axis direction, the line segment is determined to be vertical, and an offset operation is performed in the X-axis direction. If the absolute value of the coordinate difference in the Y-axis direction is less than or equal to the absolute value of the coordinate difference in the X-axis direction, the line segment is determined to be horizontal, and an offset operation is performed in the Y-axis direction. The maximum offset is determined based on the type of vector feature, with a maximum offset of 3 for ordinary lane line features, a maximum offset of 30 for curb-type features, and a maximum offset of 10 for other features. Within the determined maximum offset range, the original coordinates of the projected features are gradually shifted along the determined offset direction in increments of a preset offset step size (preferably 0.1 meters), generating a series of intermediate coordinate points after displacement correction, thereby completing the offset operation on the projected features.

[0045] Optionally, during the process of generating at least one candidate checkpoint based on the offset projected features, the intermediate coordinate points generated by the offset operation can be used as references to perform the coordinate projection process again. These intermediate coordinate points are then transformed from the world coordinate system to the image coordinate system using camera intrinsic and extrinsic parameters to obtain the corresponding two-dimensional image coordinates. For each intermediate coordinate point generated by the offset, its position in the image coordinate system is established as a candidate checkpoint. These candidate checkpoints are distributed within a rectangular or linear region centered on the original projected point. The size of this rectangular or linear region can be determined by the maximum offset and the offset step size. Each candidate checkpoint can represent a potential location of a vector feature on the plane containing the image data, used for subsequent querying of the semantic classification at that location from the semantic segmentation results to verify whether the location belongs to the expected semantic type of the vector feature.

[0046] In this embodiment, the method described above, by performing an offset operation and generating candidate checkpoints, effectively compensates for spatial alignment errors during multi-source data fusion, significantly improving the robustness of the occlusion recognition algorithm to calibration bias and positioning noise. The introduction of candidate checkpoints is no longer limited to judging a single projection point, but rather involves multi-sampling point verification within a tolerable range, greatly increasing the probability of accurately identifying the visibility of vector elements in complex scenes and avoiding misjudgments caused by minor errors.

[0047] Step S206: Use the category attributes of the candidate checkpoints and the category attributes of the vector features to check the candidate checkpoints and obtain the check results.

[0048] In the technical solution provided by step S206 of the embodiments of this application, the inspection result is used to represent the visibility of the candidate checkpoint. The inspection result can refer to the judgment conclusion regarding the visibility of the candidate checkpoint by comparing its category attributes (actual observed semantics / semantic classification / image observed semantics) with the category attributes (expected semantic type set / geometric semantics) of the vector element. The inspection result can be used to characterize whether the actual observed semantics at the candidate checkpoint falls within the expected semantic type set of the vector element. If the category attribute of the candidate checkpoint belongs to the expected semantic type set included in the category attributes of the vector element, the inspection result is visible, indicating that the candidate checkpoint is not occluded and has consistent semantics; if the category attribute of the candidate checkpoint does not belong to the expected semantic type set included in the category attributes of the vector element, the inspection result is invisible or occluded, indicating that the candidate checkpoint is occluded by non-map elements or has a semantic mismatch.

[0049] The category attribute of the aforementioned candidate checkpoints can refer to the semantic classification label obtained after semantic analysis of the image region corresponding to the candidate checkpoint using an image segmentation model. This semantic classification label characterizes the visual semantic content of the candidate checkpoint on the plane of the image data, reflecting the type of environmental object actually observed at that location. For example, the category attribute of a candidate checkpoint can include semantic types such as road, lane line, curb, ground markings, vehicle, pedestrian, or vegetation.

[0050] The category attributes of the aforementioned vector features can refer to a predefined set of expected semantic types based on the geometric structure type of the vector features in the vector map. These category attributes can be used to characterize the legal semantic range that a vector feature should present in image data under ideal, unobstructed conditions. For example, for ordinary lane line features, the category attributes of the vector features can be defined as a set including roads, lane lines, and ground markings; for curb-type features, the category attributes of the vector features can be defined as a set including roads, lane lines, ground markings, and curbs.

[0051] In this embodiment, after generating candidate checkpoints corresponding to the offset projected features, the category attributes of the candidate checkpoints and the category attributes of the vector features can be used to check the candidate checkpoints and obtain the check results. This embodiment proposes a process of using the category attributes of candidate checkpoints and the category attributes of vector features to perform semantic consistency verification, thereby determining the visibility of candidate checkpoints. The core of this embodiment lies in establishing a mapping relationship between image observation semantics and the geometric semantics of the vector map. By comparing the actual semantic classification of each candidate checkpoint with the expected semantic type set of the vector features, a clear check result is output.

[0052] Optionally, during the process of obtaining the category attributes of candidate checkpoints and vector features, for each generated candidate checkpoint, an image segmentation model is invoked to perform semantic analysis on the image region corresponding to the candidate checkpoint. The pixel-level semantic classification of that location is extracted from the image segmentation result, and this classification result serves as the category attribute of the candidate checkpoint. Its values ​​include specific semantic types such as road, lane line, curb, ground markings, vehicle, pedestrian, or vegetation. Simultaneously, based on the geometric structure type of the vector feature to be processed in the vector map, predefined mapping rules are retrieved to determine the expected semantic type set corresponding to the vector feature. This set serves as the category attribute of the vector feature. For example, if the vector feature is a regular lane line, the category attribute of the vector feature includes road, lane line, and ground markings; if the vector feature is a curb, the category attribute of the vector feature includes road, lane line, ground markings, and curb.

[0053] Optionally, during the inspection of candidate checkpoints and the generation of inspection results, the category attribute of the candidate checkpoint is compared one by one with each expected semantic type contained in the category attribute of the vector feature. If the category attribute of the candidate checkpoint exists in the set of expected semantic types defined by the category attribute of the vector feature, semantic consistency is determined to be valid, and the generated inspection result indicates that the candidate checkpoint is visible, indicating that the observed environmental object at this location matches the theoretical expectation of the vector feature and is not occluded by non-map features. If the category attribute of the candidate checkpoint does not exist in the set of expected semantic types defined by the category attribute of the vector feature, semantic consistency is determined to be invalid, and the generated inspection result indicates that the candidate checkpoint is invisible or occluded, indicating that the observed environmental object at this location does not match the type of the vector feature or is occluded by other objects. By traversing each candidate checkpoint and generating a corresponding inspection result for each candidate checkpoint, a judgment sequence representing the degree of local visibility is formed.

[0054] In this embodiment, the method described above performs semantic consistency checks by utilizing the category attributes of candidate checkpoints and vector features, achieving a leap from geometric projection to semantic verification. This effectively solves the technical problem that relying solely on geometric projection cannot identify occlusion caused by non-map features such as pedestrians, vehicles, and vegetation. It can accurately distinguish between the truly visible and occluded parts of vector features, and the generated check results have clear semantic basis, significantly improving the accuracy and robustness of occlusion identification.

[0055] Step S208: Based on the inspection results, generate label information for the environment objects.

[0056] In the technical solution provided by step S208 of this application embodiment, the tag information is used to indicate the degree of occlusion of the environmental object. The aforementioned occlusion tag can refer to a structured data record generated based on the inspection results, used to characterize the visibility status of vector elements in image data. This tag information is used to indicate the degree of occlusion and visibility attributes of the vector element relative to the environmental object. The occlusion tag can include key parameters such as visibility level, visible point count, occlusion ratio, and visible length. Specifically, the visibility level is divided into three levels based on the visible point ratio: fully visible, partially visible, and invisible; the visible point count is used to quantify the number of unoccluded geometric units in the vector element; the occlusion ratio is used to reflect the proportion of the occluded part in the overall vector element; and the visible length is used to calculate the actual observable geometric length of the vector element in the image.

[0057] In this embodiment, after inspecting candidate checkpoints using the category attributes of both candidate checkpoints and vector features, and obtaining the inspection results, label information for environmental objects can be generated based on the secondary inspection results. This embodiment proposes an implementation process for generating label information for environmental objects based on inspection results, i.e., occlusion labeling. The above process aims to aggregate discrete, pixel-level or point-level semantic consistency verification results (also known as inspection results) into occlusion state descriptions with clear physical meaning at the vector feature level. By statistically analyzing the visible point ratio and calculating geometric length, structured data containing visibility levels and detailed statistical information is constructed, thereby providing high-quality, automated supervision signals for downstream perceptual model training.

[0058] Optionally, the inspection results of each candidate checkpoint in the vector feature are statistically analyzed, and the proportion of visible points is calculated. Each candidate checkpoint corresponding to the vector feature is traversed, and the number of candidate checkpoints whose inspection results are determined to be "visible," as well as the total number of candidate checkpoints corresponding to that vector feature, are counted.

[0059] Optionally, the visibility level of a vector feature is determined based on the occlusion visibility ratio. The calculated occlusion visibility ratio is compared with a preset threshold to determine the visibility level of the vector feature. If the occlusion visibility ratio is greater than 0.8, the visibility level of the vector feature is determined to be fully visible; if the occlusion visibility ratio is greater than 0.4 and less than or equal to 0.8, the visibility level of the vector feature is determined to be partially visible; if the occlusion visibility ratio is less than or equal to 0.4, the visibility level of the vector feature is determined to be invisible. For curb-type features, the proportion of the curb being occluded by other curbs needs to be calculated additionally, and the initially determined visibility level is corrected and updated accordingly to ensure the accuracy of the determination result.

[0060] Optionally, occlusion labels containing the aforementioned statistical information are encapsulated and generated. The determined visibility level, the total number of points obtained statistically, the number of visible points, the occlusion visibility ratio, and the visible length are combined to form complete label information for the environmental object, i.e., the occlusion label. This occlusion label, as structured data output, is directly associated with the corresponding vector element, clearly representing the true visibility state and geometric coverage of the vector element in the image, and can be automatically generated without manual intervention.

[0061] In this embodiment, the method described above generates label information for environmental objects based on the inspection results, achieving an automated conversion from low-level pixel semantic verification to high-level vector feature state description, replacing the traditional manual annotation process. The generated occlusion labels not only provide qualitative visibility level judgments but also contain quantitative proportion and length information, greatly enriching the semantic dimension of the training data. This high-precision automatic annotation mechanism significantly improves data annotation efficiency, reduces manual costs, and effectively improves the training effect and final model accuracy of weakly supervised or unsupervised perceptual large models by eliminating inconsistencies between vector ground truth maps and visual observations.

[0062] In steps S202 to S208 of this embodiment, at least one vector feature in the vector map of the vehicle's environment can be projected onto the plane where the image data is located. The corresponding projected feature is obtained. An offset operation can be performed on the projected feature, and at least one candidate checkpoint can be generated based on the offset projected feature. The visibility of the candidate checkpoint can be checked using the category attributes of the candidate checkpoint and the category attributes of the vector feature, yielding a check result. Based on the check result, label information indicating whether an environmental object is occluded can be generated. In other words, in this embodiment, by projecting vector features onto the image plane and generating candidate checkpoints based on a projection error tolerance mechanism, the category attributes of the candidate points are queried using image segmentation results. By comparing the consistency between the category attributes of the candidate checkpoints and the category attributes of the vector features, the visibility is quantitatively determined and label information is generated. This method replaces inefficient manual annotation and geometric projection methods that can only identify occlusion between map features, achieving efficient and robust automatic recognition of vector map visibility. This solves the technical problem of poor vector map processing performance and improves the overall processing effect of vector maps.

[0063] The embodiments of this application will be described in detail below with reference to the steps described above.

[0064] As an optional implementation, step S202 involves projecting at least one vector feature from the vector map of the vehicle's environment onto the plane containing the image data of the environment to obtain the projection feature corresponding to the vector feature. This includes: obtaining the position information corresponding to the multiple vector feature points to be formed into a vector feature; projecting the multiple vector feature points onto the plane according to the position information to obtain multiple projection points, wherein the multiple projection points correspond one-to-one with the multiple vector feature points; and forming a projection feature from the multiple projection points.

[0065] In this embodiment, during the process of projecting vector elements onto projection elements, the position information corresponding to multiple vector element points constituting the vector elements can be obtained. Based on the position information, these multiple vector element points can be projected onto a plane to obtain multiple projection points. These vector element points refer to the basic spatial data units that constitute various road geometric elements in the vector map, representing the discrete spatial position of environmental objects in a three-dimensional world coordinate system. The vector element points are stored in three-dimensional coordinate form, containing values ​​in the X, Y, and Z dimensions, used to accurately describe the geometric shape and spatial distribution of vector elements such as lane lines, curbs, and ground markings.

[0066] Optionally, the aforementioned location information can refer to the three-dimensional coordinate data of the vector feature point in the world coordinate system, used to determine the absolute position of the point in the environmental space where the vehicle is located. Specifically, the location information includes the spatial coordinates of the point in the X-axis direction, the spatial coordinates in the Y-axis direction, and the elevation coordinates in the Z-axis direction. The aforementioned projection point can refer to the two-dimensional spatial position of the vector feature point on the plane of the environmental image data after coordinate transformation and perspective projection processing. The projection point consists of the abscissa and ordinate in the image coordinate system, representing the theoretical observation position of the vector feature point in the image. A one-to-one correspondence is maintained between the projection point and the original vector feature point.

[0067] This implementation proposes a specific process for mapping vector features from a vector map of the vehicle's environment to the plane of the image data. This process involves extracting discrete spatial point information from the vector map data structure, performing coordinate transformation and projection calculations using a camera geometry model, and finally recombining the discrete two-dimensional projection points into projection features corresponding to the geometric structure of the original vector features. This process achieves the transformation from a three-dimensional world coordinate system to a two-dimensional image coordinate system.

[0068] Optionally, in the process of obtaining the location information corresponding to the multiple vector feature points to be constructed, the geometric definition of the target vector feature can be read from the vector map data. This vector feature consists of a series of sequentially connected vector feature points. For each vector feature point, its three-dimensional spatial coordinate data in the world coordinate system is extracted. This data is the location information, containing values ​​for the X, Y, and Z dimensions. The location information accurately describes the absolute spatial position of each vector feature point in the vehicle's environment, including horizontal position and elevation information, and serves as the basic data source for subsequent projection transformation.

[0069] Optionally, during the process of projecting multiple vector feature points onto a plane based on their position information to obtain multiple projection points, a projection transformation can be performed on each vector feature point using pre-calibrated camera intrinsic and extrinsic parameters. The position information of the vector feature point in the world coordinate system is converted into three-dimensional coordinates in the camera coordinate system using camera extrinsic parameters (e.g., rotation matrix and translation vector). The three-dimensional coordinates in the camera coordinate system are then mapped to a two-dimensional image plane using perspective projection formulas using camera intrinsic parameters (e.g., focal length and principal point coordinates), calculating the two-dimensional coordinates of the point on the plane containing the image data. These two-dimensional coordinates are the projection points. Each vector feature point generates a projection point after the above calculations, thus establishing a one-to-one correspondence between multiple projection points and multiple vector feature points.

[0070] Optionally, in the process of constructing a projection feature from multiple projection points, the multiple projection points obtained in the previous steps can be connected or combined sequentially according to the connection order and topological structure of the vector feature points in the original vector feature. If the original vector feature is a line segment, then adjacent projection points are connected by straight lines to form a line segment-shaped projection feature; if the original vector feature is a polygon, then the first and last projection points are connected to form a polygonal outline projection feature. The resulting projection feature retains the geometric shape and topological relationship of the original vector feature on the two-dimensional image plane, serving as the object for offset operations and semantic queries in subsequent steps.

[0071] In this embodiment, the method described above aligns the geometric information of the vector map with the observation space of the visual image by projecting vector features onto the plane containing the image data. This projection method, based on precise camera parameters, ensures the consistency between the theoretical and actual visual positions of the vector features in the image. Simultaneously, transforming discrete vector feature points into continuous projected features preserves the geometric characteristics of the road features, enabling subsequent visibility determination to be based on the overall shape of the features rather than isolated points.

[0072] As an optional implementation, the location information includes first location information, second location information, and third location information. The first location information represents the position of the vector feature point in the world coordinate system, the second location information represents the position of the vector feature point in the camera coordinate system of the vehicle, and the third location information represents the position of the projection point on the plane. Based on the location information, multiple feature points are projected onto the plane to obtain multiple projection points, including: transforming multiple vector feature points from the world coordinate system to the camera coordinate system based on the first and second location information; and projecting multiple vector feature points from the camera coordinate system onto the plane based on the second and third location information to obtain multiple projection points.

[0073] In this embodiment, during the process of projecting multiple feature points onto a plane to obtain multiple projection points, the multiple vector feature points can be transformed from the world coordinate system to the camera coordinate system based on the first position information and the second position information. The multiple vector feature points can also be projected from the camera coordinate system onto the plane based on the second and third position information to obtain multiple projection points. The first position information refers to the three-dimensional spatial coordinate data of the vector feature points in the world coordinate system, used to determine the physical position of the vector feature points in the vehicle's environment. The first position information can be represented as a 3D coordinate point sequence P={p1,p2,...,p...} n}, where each point p i =(x i ,y i ,z i It contains values ​​in three dimensions: X, Y, and Z, which respectively represent the precise position of the vector feature point in the horizontal, vertical, and elevation directions.

[0074] Optionally, the aforementioned second position information may refer to the three-dimensional coordinate data of the vector feature point in the vehicle's camera coordinate system, used to describe the spatial relative position of the vector feature point with respect to the camera's optical center. The second position information can be expressed as p_camera = (x_c, y_c, z_c), where x_c, y_c, and z_c are the projection values ​​of the vector feature point on the X-axis, Y-axis, and Z-axis of the camera coordinate system, respectively.

[0075] Optionally, the aforementioned third positional information can refer to the two-dimensional coordinate data of the projection point on the plane where the image data is located, used to characterize the theoretical observation position of the vector feature point on the imaging plane of the image sensor. The third positional information can be represented as image coordinates (u,v), where u and v represent the pixel index or physical coordinate value of the point in the horizontal and vertical directions of the image, respectively. The aforementioned third positional information can be the final result calculated by performing perspective projection mapping between the second positional information and camera intrinsic parameters (including focal lengths fx and fy and principal point coordinates cx and cy), directly corresponding to the pixel position in the image segmentation result.

[0076] This embodiment proposes a specific implementation process for gradually mapping vector feature points from the world coordinate system to the image plane based on first, second, and third location information. The process consists of two stages: first, transforming the vector feature points from the global world coordinate system to a local camera coordinate system centered on the camera using spatial transformation parameters; and second, projecting the three-dimensional points in the camera coordinate system onto the two-dimensional image plane using an optical projection model, thereby obtaining projection points for subsequent analysis. This process achieves a precise geometric mapping from physical world coordinates to digital image pixel coordinates.

[0077] Optionally, during the transformation of multiple vector feature points from the world coordinate system to the camera coordinate system based on the first and second position information, the first position information, i.e., the three-dimensional coordinates of the vector feature point in the world coordinate system, is extracted for each vector feature point. Combined with pre-calibrated camera extrinsic parameters, including the rotation matrix and translation vector from the world coordinate system to the camera coordinate system, a spatial rigid body transformation operation is performed. The first position information is rotated using the rotation matrix, and then translated using the translation vector to calculate the three-dimensional coordinates of the vector feature point in the camera coordinate system. This calculation result is the second position information. Through the above process, each vector feature point is transformed into a local coordinate system with the camera's optical center as the origin, eliminating the influence of differences in vehicle attitude and camera mounting position.

[0078] Optionally, during the process of projecting multiple vector feature points from the camera coordinate system onto the plane based on the second and third position information to obtain multiple projection points, the second position information in the camera coordinate system can be extracted for each vector feature point. Combined with camera intrinsic parameters, including focal length (fx, fy) and principal point coordinates (cx, cy), a perspective projection transformation is performed. Using the pinhole camera model formula, the three-dimensional coordinates in the camera coordinate system are divided by the depth component (Z-axis coordinate), multiplied by the focal length, and the principal point offset is added to calculate the two-dimensional coordinates of the vector feature point on the image plane. These two-dimensional coordinates are the third position information, specifically represented as image coordinates (u, v). Through the above method, each vector feature point obtains its corresponding projection point on the plane where the image data is located, completing the geometric projection process from three-dimensional space to a two-dimensional image.

[0079] In this embodiment, the method described above decomposes the projection process of vector feature points into two stages: a transformation from the world coordinate system to the camera coordinate system and a projection from the camera coordinate system to the image plane, thus achieving high-precision spatial position mapping. This step-by-step calculation method not only conforms to the physical geometry principles of camera imaging but also accurately handles spatial distortions and viewpoint deviations caused by camera intrinsic and extrinsic parameters. The resulting multiple projection points accurately reflect the theoretical observation positions of vector features in the image, providing a reliable spatial basis for subsequent semantic consistency verification based on image segmentation results.

[0080] As an optional implementation, obtaining a projection element composed of multiple projection points includes: using the image viewport of the image data to detect the multiple projection points respectively, and obtaining multiple detection results corresponding to the multiple projection points; based on the multiple detection results corresponding to the multiple projection points, determining multiple target projection points from the multiple projection points, wherein the detection result corresponding to the target projection point is used to indicate that the target projection point is in the image viewport; and constructing a projection element from the multiple target projection points.

[0081] In this embodiment, during the acquisition of projection elements composed of multiple projection points, the image view area of ​​the image data can be used to detect the multiple projection points separately, obtaining multiple detection results corresponding to the multiple projection points. Based on the multiple detection results corresponding to the multiple projection points, a target projection point can be determined from the multiple projection points. Multiple target projection points can be used to construct projection elements. Here, the image view area refers to the effective area of ​​the image data that the camera can actually capture and record environmental information; it can also be called the effective image range. The aforementioned image view area can be defined by the physical boundaries of the image data and can be represented as a two-dimensional rectangular space defined by the image width W and the image height H. The image view area is used to determine whether a projection point is within the camera's field of view and can be the primary geometric constraint condition for judging the validity of a projection point.

[0082] The aforementioned target projection points refer to projection points that, after image visibility validity detection, are determined to be located within the valid range of the image data. The target projection points are a subset of the initially generated multiple projection points, and their image coordinates satisfy the boundary constraints of the image visibility, i.e., the horizontal coordinate is within the range defined by the image width and the vertical coordinate is within the range defined by the image height.

[0083] This embodiment proposes a specific implementation process for filtering valid points from multiple initially generated projection points and constructing the final projection features. This process aims to eliminate invalid projection points located outside the image's field of view, retain valid projection points (i.e., target projection points) within the image's visible area, and reconstruct geometric features on the image plane by connecting these target projection points. This filtering and reconstruction mechanism ensures that the projection features can include the portion actually observable by the camera, providing accurate spatial definition for subsequent semantic queries.

[0084] Optionally, in the process of detecting multiple projection points using the image's visible field of view and obtaining multiple detection results for each projection point, the two-dimensional coordinates in the image coordinate system can be extracted for each initial projection point generated by projecting vector feature points. This coordinate position is then compared with the boundary parameters of the image's visible field of view, which is defined by the image width W and image height H, and its effective range covers the area from the horizontal coordinate 0 to W and the vertical coordinate 0 to H. If the horizontal coordinate of a projection point is greater than or equal to 0 and less than or equal to W, and its vertical coordinate is greater than or equal to 0 and less than or equal to H, then the projection point is determined to be within the image's visible field of view, and the generated detection result indicates that the projection point is valid. If either the horizontal or vertical coordinate of a projection point exceeds the above range, then the projection point is determined to be outside the image's visible field of view, and the generated detection result indicates that the projection point is invalid. By traversing each initial projection point, a clear detection result is generated for each projection point, indicating whether it is within the visible field of view.

[0085] Optionally, in the process of determining multiple target projection points from multiple projection points based on multiple detection results corresponding to multiple projection points, projection points indicated by the detection results can be filtered out and marked as target projection points. Target projection points constitute the visible geometric segments of the original vector features within the image's visual field. They maintain the topological order of the original vector features but remove portions located in blind spots. Through this filtering process, a set of points containing only valid observation information is obtained, ensuring that the subsequently constructed projection features are entirely within the valid area of ​​the image data, avoiding geometric errors or semantic query failures caused by including points outside the field of view.

[0086] Optionally, in the process of constructing projection features from multiple target projection points, the selected target projection points can be geometrically connected according to the connection order of the projection points in the original vector features. If the original vector features are linear features, adjacent target projection points are connected sequentially by straight line segments to form linear projection features; if the original vector features are polygonal features, closed polygonal contours are formed by connecting the first and last target projection points. The resulting projection features accurately reflect the actual visible part of the original vector features on the image plane, and their geometric shape and topological relationship remain consistent with the original vector features, with only the length or area potentially reduced due to field-of-view truncation.

[0087] In this embodiment, the method described above effectively eliminates invalid geometric data located outside the camera's field of view by filtering the initial projection points and determining the target projection points using the image's visible field of view. This ensures that the projection features only include the portion actually observable by the image sensor. This mechanism avoids subsequent semantic query errors or wasted computational resources caused by projection points falling outside the image boundary, improving the accuracy and effectiveness of projection feature construction. Simultaneously, retaining the projection features formed by the target projection points accurately reflects the visibility of vector features in the image.

[0088] As an optional implementation, the projection element consists of multiple projection points. Step S204 involves performing an offset operation on the projection element, including: determining the offset direction of the offset operation to be performed on the projection element; and performing the offset operation on the multiple projection points according to the offset direction to obtain the offset projection points.

[0089] In this embodiment, during the offset operation on the projected features, the offset direction can be determined. Multiple projection points can be offset according to this offset direction to obtain the offset projection points. The offset direction refers to the displacement path direction set in the image plane coordinate system to correct the geometric deviation when a vector feature point is projected onto the image. The offset direction can be divided into horizontal and vertical directions, used to guide the movement of the projection point along the X-axis or Y-axis. The selection of this direction is based on the local geometry of the vector feature in the image.

[0090] This embodiment proposes a specific implementation process for performing an offset operation on a projected feature composed of multiple projection points. This process aims to compensate for geometric deviations caused by camera calibration, positioning, or synchronization errors by dynamically adjusting the positions of the projection points. The process may include determining a suitable offset direction based on the local geometric features of the projected feature, displacing each projection point along the offset direction, thereby generating offset projection points for subsequent semantic queries. This mechanism enhances the fault tolerance of the projected feature in image space.

[0091] Optionally, in determining the offset direction for the projected feature, the coordinate difference between adjacent projected points can be calculated, and the absolute values ​​of the abscissa difference and ordinate difference can be compared. If the absolute value of the ordinate difference is greater than the absolute value of the abscissa difference, it indicates that the projected feature is mainly vertical in the image, and the offset direction is determined to be horizontal, i.e., displacement along the X-axis of the image. If the absolute value of the abscissa difference is greater than or equal to the absolute value of the ordinate difference, it indicates that the projected feature is mainly horizontal in the image, and the offset direction is determined to be vertical, i.e., displacement along the Y-axis of the image. Through the above judgment based on local geometry, it is ensured that the offset operation is always perpendicular to the main extension direction of the feature, thereby maximizing the coverage of possible misaligned areas at the edge of the feature.

[0092] Optionally, during the process of performing offset operations on multiple projection points according to the offset direction to obtain the offset projection points, the coordinates of each projection point can be adjusted incrementally or subtractively within a set tolerance range, based on the determined offset direction and a specific offset step size. If the offset direction is horizontal, the x-coordinate value is increased or decreased sequentially while keeping the y-coordinate unchanged; if the offset direction is vertical, the y-coordinate value is increased or decreased sequentially while keeping the x-coordinate unchanged. Each adjustment generates a new coordinate position, which is the offset projection point. By traversing multiple offsets, a sequence of candidate checkpoints distributed along the offset direction is generated for the original projection point. These candidate checkpoints constitute a local search space for the original projection position, used to address potential positioning deviations.

[0093] In this embodiment, the method effectively solves the problem of mismatch between the vector feature projection points and the actual observed positions of the image caused by camera calibration errors, vehicle positioning errors, or time synchronization errors by determining an adaptive offset direction and performing corresponding offset operations. This mechanism, by searching along a direction perpendicular to the feature's orientation, can cover the largest range of geometric deviations with minimal computational cost, significantly improving the hit rate of subsequent semantic queries.

[0094] As an optional implementation, the vector feature consists of multiple vector feature points. Obtaining the offset direction for the offset operation to be performed on the projected feature includes: determining the line segment direction of the line segment formed by two adjacent vector feature points in the vector feature; determining the offset direction as the horizontal axis direction in response to the line segment direction being vertical; and determining the offset direction as the vertical axis direction in response to the line segment direction being horizontal.

[0095] In this embodiment, during the process of obtaining the offset direction for performing an offset operation on the projected feature, the direction of the line segment formed by two adjacent vector feature points in the vector feature can be determined. If the line segment direction is vertical, the offset direction can be determined as the horizontal axis direction. If the line segment direction is horizontal, the offset direction is determined as the vertical axis direction. The line segment direction can refer to the extension trend of the geometric connection line determined by two adjacent vector feature points in the vector feature on the image plane. The line segment direction can reflect the orientation characteristics of the local vector feature in space. The vertical direction can refer to the state where the line connecting two adjacent points in the vector feature extends vertically in the image coordinate system, which can be expressed as the absolute value of the difference between the vertical coordinates of the two points being greater than the absolute value of the difference between the horizontal coordinates. When the line segment direction is vertical, it indicates that the local feature is distributed along the image height direction. The horizontal axis direction can refer to the horizontal displacement path in the image coordinate system, that is, the axis parallel to the image width direction. When the line segment direction is vertical, the offset direction is determined as the horizontal axis (X) direction, which aims to compensate for positioning errors by searching the visible area near the vertically oriented feature through horizontal displacement.

[0096] Optionally, the aforementioned horizontal direction can refer to the state where the line connecting two adjacent points in a vector feature extends laterally in the image coordinate system, which can be represented by the absolute value of the difference in the horizontal coordinates between the two points being greater than or equal to the absolute value of the difference in the vertical coordinates. When the line segment direction is horizontal, it indicates that the local feature is distributed along the width direction of the image. The aforementioned vertical axis (Y) direction can refer to the vertical displacement path in the image coordinate system, that is, the axis parallel to the height direction of the image. When the line segment direction is horizontal, the offset direction is determined as the vertical axis direction, aiming to compensate for positioning errors by searching the visible area near the horizontally oriented feature through vertical displacement.

[0097] This embodiment proposes a logical flow for determining the offset direction of projected features. This flow aims to adaptively select the optimal offset path based on the local geometry of the vector features. By analyzing the direction of the line segments formed by adjacent points in the vector features, it is determined whether the direction of these line segments is vertical or horizontal. Based on the direction of these line segments, the offset direction is set as the horizontal axis or the vertical axis, thereby ensuring that the offset operation is perpendicular to the feature extension direction to maximize the coverage of potential geometric deviation areas.

[0098] Optionally, in determining the direction of the line segment formed by two adjacent vector element points in a vector feature, the two-dimensional coordinate positions of any two adjacent vector element points in the vector feature can be extracted in the image coordinate system. The absolute values ​​of the differences in the x-coordinates and y-coordinates between these two points are calculated and compared. If the absolute value of the y-coordinate difference is greater than the absolute value of the x-coordinate difference, the line segment is determined to extend vertically in the image, i.e., the line segment direction is vertical; if the absolute value of the x-coordinate difference is greater than or equal to the absolute value of the y-coordinate difference, the line segment is determined to extend horizontally in the image, i.e., the line segment direction is horizontal. By comparing the coordinate differences, the geometric orientation characteristics of the local line segment are quantified.

[0099] Optionally, when the line segment direction is vertical and the offset direction is determined to be the horizontal axis, if the determination result indicates that the line segment direction is vertical, it means that the local vector feature is mainly distributed along the image height direction. To effectively compensate for potential lateral positioning errors or projection deviations and avoid invalid searches along the feature extension direction (vertical direction), the offset direction is determined to be the horizontal axis. The horizontal axis corresponds to the horizontal axis of the image coordinate system. Offset operations along this direction mean adjusting the horizontal coordinate while keeping the vertical coordinate unchanged, thereby generating candidate checkpoints on both sides of the vertical feature to find the correct semantic matching position.

[0100] Optionally, when the line segment direction is horizontal and the offset direction is determined to be the vertical axis, if the determination result indicates that the line segment direction is horizontal, it means that the local vector feature is mainly distributed along the image width direction. To effectively compensate for potential vertical positioning errors or projection deviations and avoid invalid searches along the feature extension direction (horizontal direction), the offset direction is determined to be the vertical axis. The vertical axis corresponds to the vertical axis of the image coordinate system. Offset operations along this direction mean adjusting the vertical coordinate while keeping the horizontal coordinate unchanged, thereby generating candidate checkpoints on both the upper and lower sides of the horizontal feature to find the correct semantic matching position.

[0101] In this embodiment, the method dynamically determines the offset direction based on the line segment direction, achieving adaptive matching between the offset strategy and the geometric shape of the vector features. This mechanism ensures that the offset operation is always perpendicular to the main extension direction of the feature, thereby covering potentially misaligned areas at the feature edges with maximum efficiency within a limited offset tolerance range. Compared to fixed-direction offset searches, this significantly reduces redundant computation, improves the hit rate and accuracy of semantic queries, and enhances adaptability to different road feature shapes (e.g., straight lines and curves, straight and turning road segments), thereby improving the robustness and computational efficiency of the overall occlusion recognition method.

[0102] As an optional implementation, offset operations are performed on multiple projection points according to the offset direction to obtain offset projection points, including: determining the allowable offset amount for each of the multiple projection points to be constituted as projection features based on the category attribute of the vector features; performing offset operations on the projection points according to the offset direction and offset amount to generate offset projection points; and / or, step S204, generating at least one candidate checkpoint corresponding to the offset projection features based on the offset projection features, including: determining the offset projection points as candidate checkpoints.

[0103] In this embodiment, during the process of performing offset operations on multiple projection points according to the offset direction, the allowable offset amount for each projection point to be constituted as a projection element can be determined based on the category attribute of the vector element. The offset operation can be performed on the projection points according to the offset direction and offset amount to generate offset projection points. During the process of generating at least one candidate checkpoint corresponding to the offset projection element, the offset projection point can be identified as a candidate checkpoint. Here, the aforementioned category attribute can refer to the classification identifier of the vector element, used to distinguish different types of road geometric elements. The category attribute determines the meaning of the vector element in the semantic space and its corresponding geometric features, such as lane lines, curbs, ground markings, or stop lines.

[0104] Optionally, the aforementioned offset can refer to the maximum allowable displacement distance of the projected point along the offset direction in the image coordinate system, also known as the maximum offset (max_offset). This offset defines the boundary of the candidate checkpoint search range, i.e., the farthest coordinate position that can be reached from the original projected point along the offset direction. The magnitude of the offset is dynamically determined based on the category attributes of the vector features. For example, thinner lane line features typically correspond to smaller offsets, while wider curb features correspond to larger offsets. By setting a reasonable offset, the aim is to cover projection deviations caused by camera calibration errors, positioning errors, or time synchronization errors, while avoiding excessive noise interference introduced by an overly large search range.

[0105] This embodiment proposes a process for dynamically determining the offset range based on vector feature category attributes and performing offset operations to generate candidate checkpoints. The process aims to adapt to the geometric characteristics and error sensitivities of different road features through a differentiated offset tolerance mechanism. The maximum offset is determined based on the feature type, then offset projection points are generated along a specified direction. These offset projection points are then transformed into candidate checkpoints for semantic queries. Finally, the candidate checkpoints are projected back to obtain the final coordinate positions used for analysis.

[0106] Optionally, in determining the allowable offset of multiple projection points to form a projection feature based on the category attributes of vector features, different maximum offset values ​​are set for different types of vector features. For example, for ordinary lane line features, due to their narrow geometric width and high positioning accuracy requirements, the maximum offset `max_offset` is set to 3; for curb-type features, due to their wider physical structure and potentially blurred edge features, the maximum offset `max_offset` is set to 30; for other feature types not explicitly listed, the maximum offset `max_offset` is set to 10. This classification mapping mechanism ensures a moderate search range for fine-grained and small-sized features, while providing sufficient search space for large-area features, thereby optimizing the balance between search efficiency and accuracy.

[0107] Optionally, during the process of performing an offset operation on the projected points according to the offset direction and offset amount to generate the offset projected points, the coordinates of the original projected points can be adjusted incrementally or subtractively according to the previously determined offset direction (horizontal or vertical axis) and the corresponding maximum offset amount, within a set offset step size (e.g., 0.1 meters). If the offset direction is the horizontal axis, the horizontal coordinate values ​​are adjusted sequentially while keeping the vertical coordinate unchanged to generate a series of new coordinate points distributed along the X-axis; if the offset direction is the vertical axis, the vertical coordinate values ​​are adjusted sequentially while keeping the horizontal coordinate unchanged to generate a series of new coordinate points distributed along the Y-axis. The points after the coordinate displacement processing are the offset projected points, which constitute a local search grid around the original projected position to cover potential geometric deviation areas.

[0108] Optionally, in determining candidate checkpoints from the offset projection points, the generated offset projection points can be directly mapped to a sequence of candidate checkpoints. Each candidate checkpoint represents a semantic query location to be verified, aiming to tolerate errors from camera calibration, localization, or time synchronization by expanding the search range. These candidate checkpoints retain the topological association of the original projection points but are spatially expanded, forming a dense set of sampling points for specific vector feature fragments, serving as the input coordinate basis for subsequent semantic segmentation models to query semantic categories.

[0109] Optionally, for each candidate checkpoint, the projection process described above can be repeated to obtain a candidate projection point. For each candidate checkpoint, a projection calculation process is performed, including coordinate transformation and image projection. For example, the 3D world coordinates corresponding to the candidate checkpoint are transformed to the camera coordinate system, and then projected onto the image plane according to the camera intrinsic parameters, thus obtaining the final two-dimensional coordinates of the candidate checkpoint in the image coordinate system, i.e., the candidate projection point. This process ensures the precise position of the candidate checkpoint in the image space, enabling the direct reading of the semantic classification information of the corresponding pixel from the semantic segmentation results, completing the mapping from the geometric offset position to the semantic observation position.

[0110] In this embodiment, the method described above effectively solves the problem of the failure of a unified offset strategy caused by differences in geometric shape and error sensitivity among different road features by dynamically determining the offset based on category attributes and performing the offset operation. This method provides high-precision search within a small range for fine features such as lane lines and robust search within a large range for features such as road edges, significantly improving the hit rate of semantic queries. Simultaneously, converting the offset points into candidate checkpoints and reprojecting them ensures the accuracy of the search location in the image space.

[0111] As an optional implementation, step S206 involves using the category attributes of the candidate checkpoints and the category attributes of the vector features to check the candidate checkpoints and obtain a check result, including: in response to the fact that the category attributes of the candidate checkpoints are the same as the category attributes of the vector features, determining that the check result indicates that the visibility of the candidate checkpoints is greater than the visibility threshold; in response to the fact that the category attributes of the candidate checkpoints are different from the category attributes of the vector features, determining that the check result indicates that the visibility of the candidate checkpoints is less than or equal to the visibility threshold.

[0112] In this embodiment, during the inspection of candidate checkpoints using both the category attributes of the candidate checkpoints and the category attributes of the vector features, if the category attributes of the candidate checkpoints are the same as those of the vector features, the inspection result indicates that the visibility of the candidate checkpoints is greater than a visibility threshold. If the category attributes of the candidate checkpoints are different from those of the vector features, the inspection result indicates that the visibility of the candidate checkpoints is less than or equal to a visibility threshold. The aforementioned visibility threshold can refer to a critical value used to determine the semantic consistency of candidate checkpoints, distinguishing between "visible" and "occluded" states.

[0113] This embodiment proposes a specific logical process for inspecting candidate checkpoints and generating inspection results based on semantic consistency. This process aims to determine whether a location is visible by comparing the semantic category of a candidate checkpoint obtained in the image segmentation result with the preset semantic category of the vector element. Based on the matching of category attributes, the inspection result is mapped to a visibility level higher or lower than a specific threshold, thereby achieving automated identification and quantification of vector element occlusion.

[0114] Optionally, in response to the candidate checkpoint's category attribute being the same as the vector feature's category attribute, the check result indicates that the candidate checkpoint's visibility is greater than a visibility threshold. When the semantic classification of the candidate checkpoint obtained from the semantic segmentation result belongs to the expected semantic type set corresponding to the vector feature, the two are determined to have the same category attribute. This means that the semantic content observed in the image is consistent with the geometric feature type at that location in the vector map, indicating that the geometric location is not occluded by non-map features (such as vehicles, pedestrians, vegetation, etc.), or is occluded by similar map features but the semantic features remain continuous. Based on this semantic consistency judgment, the visibility of the candidate checkpoint is determined to be at a high level, i.e., greater than the preset visibility threshold, and the check result is marked as a valid visible point for subsequent calculation of the number of visible points and generation of occlusion labels.

[0115] Optionally, in response to a discrepancy between the category attribute of a candidate checkpoint and the category attribute of a vector feature, the check result indicates that the visibility of the candidate checkpoint is less than or equal to a visibility threshold. When the semantic classification of a candidate checkpoint obtained from the semantic segmentation result does not belong to the expected semantic type set corresponding to the vector feature, the two category attributes are determined to be different. This may mean that the location is obscured by other types of map features (e.g., lane lines are obscured by guardrails) or by non-map foreground objects, causing a change in the semantic information observed in the image. Based on this semantic inconsistency judgment, the visibility of the candidate checkpoint is determined to be at a low level, i.e., less than or equal to a preset visibility threshold. The check result is then marked as an obscured point or an invalid point, and is not included in the number of visible points in subsequent statistics, thus accurately reflecting the true visibility of vector features in the image.

[0116] In this embodiment, the method described above achieves automated occlusion recognition without manual annotation by performing semantic consistency checks based on category attribute matching. This method leverages the semantic understanding capabilities of image segmentation models to accurately distinguish the visibility and occlusion states of vector elements, effectively solving the problem in related technologies where geometric projection methods cannot identify occlusion of non-map elements. By setting a visibility threshold as the criterion, the complex semantic matching process is transformed into a clear binary judgment logic, improving the accuracy and robustness of occlusion recognition.

[0117] As an optional implementation, the method further includes: determining a set of category attributes for vector features based on the type information of the vector features, wherein the set of category attributes includes the category attributes of at least one vector feature.

[0118] In this embodiment, a set of category attributes for vector elements can be determined based on the type information of the vector elements. This set of category attributes refers to the expected semantic type set (T_expected) determined based on the type information of the vector elements, used to verify semantic consistency. This set of category attributes can consist of semantic categories output by at least one image segmentation model, representing the legal semantic classification range that a specific type of vector element should present during image observation. For example, for lane line elements, the set of category attributes could include "road," "lane line," or "ground marking"; for curb elements, the set of category attributes could include "road," "lane line," "ground marking," or "curb." During occlusion recognition, the actual semantic classification of candidate checkpoints is compared with this set of category attributes. If the actual classification falls within the set's range, it is determined to be semantically consistent and visible; otherwise, it is determined to be semantically inconsistent and possibly occluded. This set is the key rule definition for realizing the mapping from geometric projection to semantic visibility.

[0119] Optionally, based on the type information of the vector features, a set of category attributes for the vector features is determined. For the vector features to be processed, a type identifier is extracted. This type identifier is used to distinguish different road geometric features, such as ordinary lane lines, curbs, ground markings, or stop lines. According to predefined mapping rules, the type identifiers of the vector features are associated with a specific semantic classification list. For ordinary lane line features, the determined set of category attributes includes "road," "lane line," and "ground markings"; for curb type features, the determined set of category attributes includes "road," "lane line," "ground markings," and "curb"; for ground marking features, the determined set of category attributes includes "lane line" and "ground markings"; and for stop line features, the determined set of category attributes includes "lane line" and "ground markings." Through this type-information-based mapping mechanism, a set of category attributes containing at least one category attribute is generated. This set is the expected semantic type set (T_expected), which defines the semantic range that the vector feature may present when visible in the image.

[0120] In this embodiment, the method described above dynamically determines the set of category attributes based on vector feature type information, thereby standardizing and automating semantic verification rules. This method avoids the lack of flexibility caused by hard coding, enabling the system to adaptively adjust visibility judgment criteria according to the geometric and semantic features of different road features. For example, considering that lane lines may be identified as road or ground markings in an image, including them in the set of category attributes improves the tolerance and accuracy of the judgment. The above mechanism ensures the tight coupling between occlusion recognition logic and road feature types, improving the overall solution's adaptability and recognition accuracy to complex road scenarios, while reducing the cost of manually configuring rules.

[0121] As an optional implementation, step S208, generating label information for environmental objects based on the inspection results, includes: determining the visibility ratio of vector elements based on the inspection results, wherein the visibility ratio is used to represent the proportion of the visible portion of a vector element to the total portion of the vector elements; determining label information as first label information in response to a visibility ratio greater than a first visibility ratio threshold; determining label information as second label information in response to a visibility ratio less than or equal to the first visibility ratio threshold and greater than a second visibility ratio threshold, wherein the occlusion degree of the environmental object indicated by the second label information is greater than the occlusion degree of the environmental object indicated by the first label information, and the second visibility ratio threshold is less than the first visibility ratio threshold; and determining label information as third label information in response to a visibility ratio less than or equal to the second visibility ratio threshold, wherein the occlusion degree of the environmental object indicated by the third label information is greater than the occlusion degree of the environmental object indicated by the second label information.

[0122] In this embodiment, during the process of generating label information for environmental objects based on the inspection results, the visibility ratio of vector elements can be determined based on the inspection results. If the visibility ratio is greater than a first visibility ratio threshold, the label information is determined to be first label information. If the visibility ratio is less than or equal to the first visibility ratio threshold and greater than a second visibility ratio threshold, the label information is determined to be second label information. If the visibility ratio is less than or equal to the second visibility ratio threshold, the label information is determined to be third label information. The aforementioned visibility ratio can refer to the ratio of the number of visible points in a vector element to the total number of points, used to quantify the visibility of the vector element in the image. The first visibility ratio can be obtained by statistically analyzing the ratio of the number of visible projection points to the total number of projection points; a higher value indicates a lower degree of occlusion of the element. The aforementioned first visibility ratio threshold can be a high-level critical value (e.g., 0.8) used to determine whether a vector element is in a fully visible state. When the visibility ratio is higher than the first visibility ratio threshold, it is determined that the vast majority of the vector element is not occluded. The aforementioned first label information can represent a label indicating that the vector element is in a fully visible state (e.g., "fully visible"). The aforementioned first label information can indicate the lowest degree of occlusion of environmental objects, and is suitable for training scenarios that are not sensitive to occlusion or require high confidence.

[0123] Optionally, the aforementioned second visibility ratio threshold can be used as a median threshold (e.g., 0.4) to determine whether a vector feature is in a state between partially visible and invisible. The second visibility ratio threshold is less than the first visibility ratio threshold, serving as the boundary between partially visible and invisible. The aforementioned second label information can represent the label of a vector feature in a partially visible state (e.g., "partially visible"). The aforementioned second label information can indicate a moderate degree of occlusion of the environment object, between fully visible and invisible, used to prompt the model that the feature is partially occluded. The aforementioned third label information can represent the label of a vector feature in an invisible state (e.g., "invisible"). The aforementioned third label information can indicate the highest degree of occlusion of the environment object, indicating that the vector feature is essentially invisible in the image, and can be used for filtering or as negative samples for training.

[0124] This embodiment proposes a quantitative process for generating environmental object label information based on statistical information from semantic inspection results. The process aims to discretize continuous visibility states into three levels of label information by calculating the visibility ratio of vector elements and comparing it with preset thresholds (e.g., a first visibility ratio threshold, a second visibility ratio threshold, etc.). The process first counts visible points to calculate the visibility ratio, then sequentially determines whether the ratio falls within the range of fully visible, partially visible, or invisible, and generates corresponding labels to achieve a refined classification of occlusion levels.

[0125] Optionally, in determining the visibility ratio of vector features based on the inspection results, the inspection results of each projection point of the vector feature in semantic inspection are statistically analyzed. Points with visibility greater than a visibility threshold are counted as visible points, while points with visibility less than or equal to the visibility threshold are counted as invisible points. The ratio of the number of visible points to the total number of projection points of the vector feature is calculated to obtain the visibility ratio. This visibility ratio quantifies the proportion of the visible portion of the vector feature in the image relative to the total portion of the vector feature. The value typically ranges from 0 to 1; a higher value indicates a greater visible proportion and a smaller occluded proportion of the vector feature in the image.

[0126] Optionally, in response to a visible ratio greater than a first visible ratio threshold, the label information is determined to be the first label information. The calculated visible ratio is compared with the first visible ratio threshold, which is preferably set to 0.8. When the visible ratio is strictly greater than 0.8, the vector element is determined to have extremely high visibility and extremely low occlusion, therefore the label information is determined to be the first label information. In a preferred embodiment, the first label information specifically represents "fully visible," indicating that the environmental object is not occluded in most areas of the image, suitable for training scenarios that are insensitive to occlusion or require high-confidence positive samples.

[0127] Optionally, in response to a visible ratio being less than or equal to a first visible ratio threshold and greater than a second visible ratio threshold, the label information is determined to be second label information. When the visible ratio satisfies a value greater than or equal to 0 and less than or equal to 0.8, and is strictly greater than the second visible ratio threshold, it enters a second determination interval. The second visible ratio threshold is preferably set to 0.4, and this second visible ratio threshold is less than the first visible ratio threshold. Within the aforementioned second determination interval, the visibility of the vector element is at a moderate level, and its occlusion degree is greater than the occlusion degree indicated by the first label information, but less than the invisible state. Therefore, the label information is determined to be second label information. In a preferred embodiment, the second label information is specifically represented as "partially visible," indicating that the environmental object has significant occlusion, but some features are still visible, used to indicate the incompleteness of the element in the model.

[0128] Optionally, in response to a visible ratio being less than or equal to a second visible ratio threshold, the label information is determined to be third label information. When the visible ratio is less than or equal to the second visible ratio threshold (e.g., a visible ratio less than or equal to 0.4), the visibility of the vector element is determined to be extremely low, with the highest degree of occlusion. In this case, the label information is determined to be third label information. In a preferred embodiment, the third label information is specifically represented as "invisible," indicating that the environmental object is essentially completely occluded or cannot be effectively observed in the image, and the degree of occlusion indicated is greater than that indicated by the second label information. This label can be used to filter invalid data or as a negative sample to participate in model training, thereby enhancing the model's robustness to occluded scenes.

[0129] In this embodiment, the method described above achieves refined hierarchical labeling of vector element occlusion states through a multi-level threshold determination mechanism based on visibility ratio. This method maps continuous visibility statistics to three distinct label levels: "fully visible," "partially visible," and "invisible." This not only preserves the quantitative information of occlusion degree but also simplifies the label structure, making it easier for the model to use directly. This hierarchical strategy effectively distinguishes samples with different occlusion degrees, avoiding semantic ambiguity caused by a one-size-fits-all binary labeling approach, thereby improving the quality of training data. Furthermore, by dynamically adjusting the threshold, this method can adapt to the occlusion tolerance requirements of different scenarios, significantly improving the training effect and generalization ability of the perception model in complex occlusion environments.

[0130] As an optional implementation, determining the visibility ratio of vector features based on the inspection results includes: determining the number of visible points corresponding to at least one candidate inspection point and the total number of points based on the inspection results, wherein the number of visible points is used to represent the number of candidate inspection points whose visibility is greater than the visibility threshold, and the total number of points is used to represent the number of candidate inspection points; and determining the ratio of the number of visible points to the total number of points as the visibility ratio.

[0131] In this embodiment, during the process of determining the visibility ratio of vector features based on the detection results, the number of visible points corresponding to at least one candidate checkpoint and the total number of points can be determined based on the inspection results. The proportion of visible points to the total number of points can be determined as the visibility ratio. The number of visible points can refer to the total number of candidate checkpoints in the candidate checkpoint set of the vector feature that are determined to be visible by the semantic inspection results. The total number of points can refer to the total number of candidate checkpoints used to evaluate the visibility of the vector feature, i.e., the total number of points. The total number of points can include all candidate checkpoints of the vector feature after projection, offset search, and semantic query, regardless of whether the category attribute is consistent with the category attribute of the vector feature.

[0132] Optionally, in determining the number of visible points and the total number of points corresponding to at least one candidate checkpoint based on the inspection results, each candidate checkpoint corresponding to the vector element can be traversed, and the inspection result of each candidate checkpoint can be judged. If the inspection result of a candidate checkpoint indicates that the visibility is greater than the visibility threshold, then the candidate checkpoint is counted in the number of visible points. The number of visible points is used to represent the number of candidate checkpoints with visibility greater than the visibility threshold, reflecting the scale of effective observation points of the vector element in the image that are not occluded. At the same time, the total number of each candidate checkpoint participating in the inspection is counted to determine the total number of points. The total number of points is used to represent the number of candidate checkpoints, representing the total scale of points sampled by the vector element in the image space.

[0133] Optionally, in determining the visibility ratio as the proportion of visible points to the total number of points, the aforementioned number of visible points can be divided by the total number of points, and the resulting quotient is the visibility ratio. This visibility ratio represents the proportion of the visible portion of a vector feature to the total portion of the vector feature, and its value ranges from 0 to 1. A higher visibility ratio indicates that more of the vector feature is visible in the image and less is occluded; a lower visibility ratio indicates that the vector feature is occluded to a greater extent in the image.

[0134] In this embodiment, the method described above achieves a quantitative assessment of the visibility of vector elements by statistically analyzing the number of visible points and the total number of points based on the inspection results and calculating the visibility ratio. This method avoids errors caused by subjective judgment and accurately reflects the actual visibility status of vector elements in the image using objective pixel-level statistical results.

[0135] The technical solutions of the embodiments of this application will be illustrated below with reference to preferred embodiments.

[0136] When training a large-scale perceptual model, each training sample needs to include the original image and its corresponding ground truth vector map. The ground truth vector map, typically derived from manually drawn or map-derived survey data, contains precise location information of road geometry elements such as lane lines, curbs, and ground markings, and is generally quite complete. However, the original image may have some gaps due to viewing angle and occlusion. This inconsistency interferes with the training process.

[0137] Existing perceptual model training methods mainly rely on the following techniques.

[0138] (1) Manual annotation method: The image is checked frame by frame by manual annotators to identify which vector features are visible or occluded in the image. This method has high accuracy, but low efficiency and high cost, and it is difficult to meet the needs of large-scale data annotation.

[0139] (2) Geometric projection-based method: The occlusion relationship is inferred by using elevation information (such as guardrail height): If vector features A and B coincide in the line of sight, they are considered to be occluded. However, this method has obvious limitations: it can only determine the occlusion between map features. For occlusion caused by non-map features such as pedestrians, vehicles, and vegetation in the image, geometric projection cannot identify it, and therefore cannot correctly remove the occluded vector features.

[0140] The existing methods have the following problems. (1) They cannot automatically identify occlusion: The existing methods cannot automatically identify whether vector features are occluded in the image, which requires manual annotation, which is inefficient and costly. (2) They lack robustness, have low recall, and cannot identify all possible occlusions.

[0141] This application provides a method for element occlusion recognition in vector ground truth maps based on image segmentation, aiming to solve the following technical problems: how to automatically identify whether elements in a vector ground truth map are visible or occluded in an image without manual annotation, thus improving annotation efficiency; how to generate high-precision occlusion labels to support weakly supervised or unsupervised model training; and how to support visibility detection for various road feature types (lane lines, curbs, ground markings, etc.).

[0142] The methods of the embodiments of this application will be further illustrated below.

[0143] Figure 3 This is a flowchart of a vector ground map occlusion recognition method based on image segmentation according to an embodiment of this application, such as... Figure 3 As shown, the method may include the following steps.

[0144] Step S301, Data preparation and initialization.

[0145] In this embodiment, the vector ground truth map data to be processed, the original image and its corresponding semantic segmentation results, as well as necessary parameters such as camera intrinsic and extrinsic parameters are acquired. The vector ground truth map contains geometric elements composed of 3D coordinate points, such as lane lines, curbs, and ground markings. The original image is in RGB format, the semantic segmentation results provide pixel-level semantic classification, and the camera parameters are used for subsequent spatial coordinate transformation to ensure that all input data are in a uniform format and are ready for processing.

[0146] Optionally, vector ground truth map data: a vector ground truth map containing road geometry features, with the coordinate system being the world coordinate system. Feature types include, but are not limited to: lane line features, including lane center lines, lane boundary lines, etc., each feature consisting of a series of 3D coordinate points; curb features, including curb stones, guardrails, etc., also represented by a sequence of 3D coordinate points; ground marking features, including pedestrian crossings, ground arrows, stop lines, etc., represented by polygons or point sequences.

[0147] Optionally, the image data may include the original image and the corresponding semantic segmentation result, wherein the original image is an RGB image captured by the camera; the semantic segmentation result is a pixel-level semantic classification result, with each pixel corresponding to a semantic category, including: road, lane line, curb, ground marking, vehicle, pedestrian, vegetation, etc.

[0148] Optionally, camera parameters may include intrinsic and extrinsic parameters, wherein the intrinsic parameters include focal lengths fx and fy and principal point coordinates cx and cy; and the extrinsic parameters include the rotation matrix R from the world coordinate system to the camera coordinate system and the translation vector t.

[0149] Step S302: Project the vector points onto the image coordinates.

[0150] In this embodiment, each 3D world coordinate point in the vector features is extracted, transformed to the camera coordinate system using camera extrinsic parameters, and then projected onto the image plane using camera intrinsic parameters to obtain the corresponding image pixel coordinates. Simultaneously, a validity check is performed on the projected points, retaining only those within the effective field of view of the image and discarding invalid points beyond the image boundaries, thus providing an accurate positioning basis for subsequent semantic queries.

[0151] Optionally, for each feature in the vector truth map, extract its 3D coordinate point sequence P={p1,p2,...,p...} n}, where each point p i =(x i ,y i ,z i () represents the 3D coordinates in the world coordinate system. For each 3D point p i Perform the following projection procedure.

[0152] Optionally, a coordinate transformation can be performed. The 3D point p in the world coordinate system can be transformed. i Transform to camera coordinate system: p_camera=R·p_world+t Where p_camera=(x_c,y_c,z_c) are the coordinates in the camera coordinate system, R is a 3×3 rotation matrix, and t is a 3×1 translation vector.

[0153] Optionally, image projection is performed. Points in the camera coordinate system are projected onto the image plane to obtain image coordinates (u,v): u = -fx·(y_c / x_c) + cx v = -fy·(z_c / x_c) + cy Where fx and fy are the camera focal lengths, and cx and cy are the coordinates of the camera principal point.

[0154] Optionally, a validity check is performed. The projected point is checked to see if it is within the valid range of the image. If u∈[0,W] and v∈[0,H], where W is the image width and H is the image height, then the projected point is valid; otherwise, the point is not within the image's field of view and is marked as invisible.

[0155] Step S303, semantic query.

[0156] In this embodiment, considering projection errors and positioning deviations, an offset tolerance mechanism is implemented for effective projection points. Multiple candidate checkpoints are generated within a preset maximum offset range based on the line segment direction. For each candidate checkpoint, the projection process is repeated to obtain its position in the semantic segmentation result, and the semantic classification corresponding to that position is queried to obtain semantic information reflecting the actual observed content of the image, thereby addressing the problem of projection points deviating from the true semantic region due to errors.

[0157] Optionally, due to factors such as camera calibration errors, positioning errors, and time synchronization errors, there may be deviations when vector points are projected onto image coordinates. To solve this problem, embodiments of this application employ the following offset tolerance mechanism.

[0158] Optionally, the direction of the line segment can be determined. For two adjacent points p in the element... i and p i+1 Determine the direction of the line segment. If |y i+1 -y i |>|x i+1 -x i If |, then the line segment is in the vertical direction; otherwise, the line segment is in the horizontal direction.

[0159] Optionally, the offset direction can be selected. The offset direction is selected based on the direction of the line segment. For example, if the line segment is vertical, the offset is made in the X direction; if the line segment is horizontal, the offset is made in the Y direction.

[0160] Optionally, candidate checkpoints are generated. An offset search is performed on the original projected points within the tolerance range to generate candidate checkpoints. For offsets ∈ [-max_offset, max_offset], where max_offset is the maximum offset, candidate points are generated. If the line segment is vertical, p_candidate=(x i +offset·Δs,y i ,z i If the line segment is horizontal, p_candidate=(x i ,y i +offset·Δs,z i ), where Δs is the offset step size, which increases in increments of 0.1 meters.

[0161] Optionally, different maximum offsets (max_offset) can be used for different feature types. For ordinary lane line features, max_offset=3; for curb-type features, max_offset=30; and for other features, max_offset=10.

[0162] Optionally, a semantic query can be performed. For each candidate checkpoint, the projection process described above is repeated to obtain candidate projection points, and the semantic classification of the point is queried from the semantic segmentation results.

[0163] Step S304, semantic consistency verification.

[0164] In this embodiment, a set of expected semantic types is defined based on the type of the vector element. This set contains the legal semantic classifications that the element may present in the image. The actual semantic classification retrieved by the candidate checkpoint is compared with the set of expected semantic types. If the actual classification belongs to the expected set, the point is determined to be visible and semantically consistent; if it does not belong, the point is determined to be occluded or semantically inconsistent, thereby achieving a preliminary logical judgment on the visibility of the vector element.

[0165] Optionally, a corresponding expected semantic type set T_expected is defined based on the type of vector feature. For ordinary lane line features, the expected semantic type set includes: T_expected={road, lane line, ground marking}; for curb type features, the expected semantic type set includes: T_expected={road, lane line, ground marking, curb}; for ground marking features, the expected semantic type set includes: T_expected={lane line, ground marking}.

[0166] Optionally, semantic matching is performed. For each projection point (or candidate projection point), its semantic classification T_actual is obtained from the semantic segmentation results, and semantic consistency is judged. If T_actual∈T_expected, the point is considered visible; otherwise, the point is considered occluded.

[0167] Step S305, Visibility level determination.

[0168] In this embodiment, the visibility ratio is calculated by comparing the number of semantically consistent visible points among each candidate checkpoint of the statistical vector element with the total number of points. Based on a preset first and second visibility ratio thresholds, the visibility ratio is divided into three intervals, corresponding to three visibility levels: fully visible, partially visible, and invisible. For specific elements such as curbs, the situation where they are occluded by other curbs needs to be considered to further refine the visibility level and detail the description of the occlusion state.

[0169] Optionally, the occlusion ratio is calculated. For each point in the feature, the number of visible points and the total number of points are counted, and the occlusion ratio is calculated: Occlusion-visibility ratio = Number of visible points / Total number of points.

[0170] Optionally, a three-level visibility determination is performed. The visibility level of the feature is determined based on the occlusion ratio. If the occlusion visibility ratio > 0.8, it is determined to be fully visible (kVisible); if 0.4 < occlusion visibility ratio ≤ 0.8, it is determined to be partially visible (kFuzzyVisible); if the occlusion visibility ratio ≤ 0.4, it is determined to be invisible (kInvisible).

[0171] Optionally, a judgment logic for special elements can be implemented. For curb-type elements, in addition to considering the occlusion ratio, it is also necessary to consider whether they are occluded by other curbs: calculate the proportion of the curb element that is occluded by other curbs, and update the visibility level again based on the proportion.

[0172] Step S306: Obscuring label generation.

[0173] In this embodiment, based on the determined visibility level and related statistical information, a final occlusion label is generated for each vector feature. The label content includes data such as visibility level (fully visible, partially visible, or invisible), number of visible points, total number of points, visible proportion, and visible length. The generated occlusion labels are directly used for training the perception model. By providing accurate visibility supervision signals, they help the model learn to identify occluded map features and improve the model's robustness in complex scenes.

[0174] Optionally, an occlusion label is generated for each vector feature. The label information includes: visibility level and statistical information. The visibility level can be one of three levels: fully visible, partially visible, or invisible. The statistical information can include the total number of points, the number of visible points, the occlusion-to-visibility ratio, and the degree of visibility. The total number of points can be the total number of points contained in the feature. The number of visible points can be the number of points visible in the image. The occlusion-to-visibility ratio can be the proportion of visible points to the total number of points. The visible length can be the total length (in meters) of the visible portion. The generated occlusion labels can be used for weakly supervised or unsupervised model training. By filtering occluded features, the quality of training data and the model training effect are improved.

[0175] For example, the projection error tolerance parameter may include at least one of the following: Offset step size Δs: preferably 0.1 meters, which can be adjusted within the range of 0.05-0.2 meters according to the actual application scenario; Maximum offset of ordinary lane lines max_offset: preferably 3, corresponding to a tolerance range of 0.3 meters; Maximum offset of curb type max_offset: preferably 30, corresponding to a tolerance range of 3.0 meters; Other maximum offset max_offset: preferably 10, corresponding to a tolerance range of 1.0 meter.

[0176] For example, the visibility thresholds can be adjusted according to actual application requirements. In more demanding application scenarios, the fully visible threshold can be increased to 0.9. For instance, the fully visible threshold is defined as: occlusion ratio > 0.8; the partially visible threshold is defined as: 0.4 < occlusion ratio ≤ 0.8; and the invisible threshold is defined as: occlusion ratio ≤ 0.4.

[0177] As an optional example, the semantic mapping of feature types is as follows: Ordinary lane lines: expected semantic type set is {road, lane lines, ground markings}; Curb type: expected semantic type set is {road, lane lines, ground markings, curb}; Stop line: expected semantic type set is {lane lines, ground markings}.

[0178] Figure 4 This is a schematic diagram of a vector truth map occlusion recognition system based on image segmentation according to an embodiment of this application, such as... Figure 4 As shown, the vector ground truth map occlusion recognition system 40 based on image segmentation may include a data input module 41, a projection calculation module 42, an error tolerance module 43, a semantic query module 44, a consistency verification module 45, a visibility determination module 46, and a label generation module 47. The data input module 41 receives vector ground truth map data, image data, and camera parameters, and performs data format conversion and verification. The projection calculation module 42 projects 3D vector points onto image coordinates, including coordinate transformation, image projection, and validity checks. The error tolerance module 43 implements a projection error tolerance mechanism, including functions such as line segment direction determination, offset direction selection, and tolerance range search. The semantic query module 44 queries the semantic classification of projected points from the semantic segmentation results. The consistency verification module 45 verifies the consistency between the image semantic classification and the vector feature type, and determines visibility. The visibility determination module 46 determines the visibility level based on the proportion of visible points, including functions such as occlusion ratio calculation and three-level visibility determination. The aforementioned tag generation module 47 generates occlusion tags, including visibility level, statistical information, occlusion ratio, etc.

[0179] In the embodiments of this application, the technical effects verified through practical application are as follows: Automation: The embodiments of this application automate the process without manual intervention. Compared to manual annotation methods, annotation efficiency is improved by more than 10 times, significantly reducing annotation costs. Improved Accuracy: Through a projection error tolerance mechanism and multi-level visibility determination, the accuracy of occlusion labels reaches over 95%, meeting the needs of weakly supervised and unsupervised training. Compared to methods based solely on geometric projection, accuracy is improved. Enhanced Robustness: Through an offset tolerance mechanism, alignment errors such as camera calibration errors and positioning errors are effectively handled. Extensibility: Supports multiple road element types, including lane lines, curbs, stop lines, arrows, pedestrian crossings, center lines, etc. It is easy to extend to new element types; only the corresponding semantic type set and determination threshold need to be defined, resulting in low extension costs.

[0180] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0181] According to another aspect of the embodiments of this application, and the above... Figure 2 Corresponding to the embodiments of the vector map processing method shown, this specification also provides a vector map processing system.

[0182] Figure 5 This is a schematic diagram of a vector map processing system according to an embodiment of this application, such as... Figure 5As shown, the vector map processing system 50 may include: a data input module 51, a projection calculation module 52, an error tolerance module 53, a consistency verification module 54, a visibility determination module 55, and a label generation module 56. The system includes the following modules: a data input module 51, which acquires a vector map of the vehicle's environment and image data of the environment; a projection calculation module 52, which projects at least one vector feature from the vector map of the vehicle's environment onto the plane containing the image data of the environment to obtain the projection feature corresponding to the vector feature, wherein the vector feature represents the geometric structure of environmental objects in the environment, and the image content of the image data includes environmental objects; an error tolerance module 53, which performs an offset operation on the projection feature and generates at least one candidate checkpoint corresponding to the offset projection feature, wherein the candidate checkpoint characterizes the degree of offset allowed for the vector feature when it is projected onto the plane; a consistency verification module 54, which checks the candidate checkpoint using the category attributes of the candidate checkpoint and the category attributes of the vector feature to obtain a check result, wherein the check result represents the visibility of the candidate checkpoint; a visibility determination module 55, which determines the visibility level of the vector feature based on the check result; and a label generation module 56, which generates label information for the environmental object based on the visibility level, wherein the label information represents the degree to which the environmental object is occluded.

[0183] According to another aspect of the embodiments of this application, and the above... Figure 2 Corresponding to the embodiments of the vector map processing method shown, this specification also provides a vector map processing apparatus.

[0184] Figure 6 This is a schematic diagram of a vector map processing apparatus according to an embodiment of this application, such as... Figure 6As shown, the vector map processing device 60 may include: a projection module 62, an offset module 64, an inspection module 66, and a generation module 68. The projection module 62 projects at least one vector feature from the vector map of the vehicle's environment onto the plane containing the image data of the environment, obtaining a projected feature corresponding to the vector feature. The vector feature represents the geometric structure of an environmental object, and the image content of the image data includes the environmental object. The offset module 64 performs an offset operation on the projected feature and generates at least one candidate checkpoint corresponding to the offset projected feature. The candidate checkpoint characterizes the degree of offset allowed for the vector feature when projected onto the plane. The inspection module 66 inspects the candidate checkpoint using the category attributes of the candidate checkpoint and the category attributes of the vector feature, obtaining an inspection result. The inspection result indicates the visibility of the candidate checkpoint. The generation module 68 generates label information for the environmental object based on the inspection result. The label information indicates the degree of occlusion of the environmental object.

[0185] Figure 7 This is a schematic diagram of an electronic device according to an embodiment of this application, such as... Figure 7 As shown, an electronic device 70 is also provided, including: a memory 71 storing an executable program; and a processor 72 for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0186] Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0187] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0188] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0189] Embodiments of this application also provide a computer program that, when executed by a processor, implements the methods described in the various embodiments of this application.

[0190] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0191] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection can be through some interfaces; the indirect coupling or communication connection between units or modules can be electrical or other forms.

[0192] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0193] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0194] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0195] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for processing vector maps, characterized in that, include: At least one vector feature in the vector map of the vehicle's environment is projected onto the plane where the image data of the environment is located to obtain the projection feature corresponding to the vector feature. The vector feature is used to represent the geometric structure of environmental objects in the environment, and the image content of the image data includes the environmental objects. An offset operation is performed on the projected feature, and at least one candidate checkpoint is generated based on the offset projected feature, wherein the candidate checkpoint is used to characterize the degree of offset of the vector feature when the vector feature is projected onto the plane; The candidate checkpoints are inspected using the category attributes of the candidate checkpoints and the category attributes of the vector features to obtain inspection results, wherein the inspection results are used to represent the visibility of the candidate checkpoints. Based on the inspection results, label information of the environmental object is generated, wherein the label information is used to indicate the degree to which the environmental object is occluded.

2. The method according to claim 1, characterized in that, The step of projecting at least one vector feature from the vector map of the vehicle's environment onto the plane containing the image data of the environment to obtain the projected feature corresponding to the vector feature includes: Obtain the position information corresponding to the multiple vector element points to be constituted; Based on the location information, multiple vector feature points are projected onto the plane to obtain multiple projection points, wherein each projection point corresponds one-to-one with a multiple vector feature point. The projection points are used to form the projection element.

3. The method according to claim 2, characterized in that, The location information includes first location information, second location information, and third location information. The first location information indicates the position of the vector feature point in the world coordinate system. The second location information indicates the position of the vector feature point in the camera coordinate system of the vehicle. The third location information indicates the position of the projection point on the plane. The step of projecting multiple feature points onto the plane based on the location information to obtain multiple projection points includes: Based on the first location information and the second location information, the multiple vector feature points are transformed from the world coordinate system to the camera coordinate system. Based on the second position information and the third position information, multiple vector feature points are projected from the camera coordinate system onto the plane to obtain multiple projection points.

4. The method according to claim 2, characterized in that, The step of obtaining the projection element composed of multiple projection points includes: Using the image field of view of the image data, multiple projection points are detected respectively to obtain multiple detection results corresponding to the multiple projection points; Based on the multiple detection results corresponding to the multiple projection points, multiple target projection points are determined from the multiple projection points, wherein the detection results corresponding to the target projection points are used to indicate that the target projection points are in the image's visible field of view; The projection elements are formed by combining multiple target projection points.

5. The method according to claim 1, characterized in that, The projection feature is composed of multiple projection points, and the offset operation on the projection feature includes: Determine the offset direction on which the offset operation is to be performed on the projected feature; The offset operation is performed on multiple projection points according to the offset direction to obtain the offset projection points.

6. The method according to claim 5, characterized in that, The vector feature is composed of multiple vector feature points, and obtaining the offset direction for performing the offset operation on the projected feature includes: Determine the direction of the line segment formed by two adjacent vector element points in the vector element; In response to the fact that the direction of the line segment is vertical, the offset direction is determined to be the horizontal axis direction; In response to the line segment being horizontal, the offset direction is determined to be the vertical axis direction.

7. The method according to claim 5, characterized in that, The step of performing the offset operation on multiple projection points according to the offset direction to obtain the offset projection points includes: Based on the category attribute of the vector feature, determine the allowable offset amount for each of the multiple projection points to be constituted as the projection feature; The offset operation is performed on the projection point according to the offset direction and the offset amount to generate the offset projection point; And / or, Based on the offset projected features, at least one candidate checkpoint is generated corresponding to the offset projected features, including: The offset projection point is determined as the candidate checkpoint.

8. The method according to claim 7, characterized in that, The process of using the category attributes of the candidate checkpoints and the category attributes of the vector features to inspect the candidate checkpoints and obtain inspection results includes: In response to the fact that the category attribute of the candidate checkpoint is the same as the category attribute of the vector feature, the inspection result is determined to indicate that the visibility of the candidate checkpoint is greater than the visibility threshold; In response to the fact that the category attribute of the candidate checkpoint is different from the category attribute of the vector feature, the inspection result is determined to indicate that the visibility of the candidate checkpoint is less than or equal to the visibility threshold.

9. The method according to claim 8, characterized in that, The method further includes: Based on the type information of the vector features, a set of category attributes for the vector features is determined, wherein the set of category attributes includes at least one category attribute of the vector feature.

10. The method according to any one of claims 1 to 9, characterized in that, The step of generating tag information for the environmental object based on the inspection results includes: Based on the inspection results, the visibility ratio of the vector element is determined, wherein the visibility ratio is used to represent the proportion of the visible part of the vector element to the total part of the vector element; In response to the visibility ratio being greater than a first visibility ratio threshold, the label information is determined to be first label information; In response to the visibility ratio being less than or equal to a first visibility ratio threshold and greater than a second visibility ratio threshold, the label information is determined to be second label information, wherein the degree of occlusion of the environmental object indicated by the second label information is greater than the degree of occlusion of the environmental object indicated by the first label information, and the second visibility ratio threshold is less than the first visibility ratio threshold; In response to the visibility ratio being less than or equal to the second visibility ratio threshold, the label information is determined to be third label information, wherein the degree of occlusion of the environmental object indicated by the third label information is greater than the degree of occlusion of the environmental object indicated by the second label information.

11. The method according to claim 10, characterized in that, Determining the visibility ratio of the vector features based on the inspection results includes: Based on the inspection results, the number of visible points corresponding to at least one of the candidate inspection points and the total number of points are determined, wherein the number of visible points is used to represent the number of candidate inspection points whose visibility is greater than the visibility threshold, and the total number of points is used to represent the number of candidate inspection points; The visible ratio is defined as the proportion of the number of visible points to the total number of points.

12. A vector map processing system, characterized in that, include: The data input module is used to acquire a vector map of the vehicle's environment and image data of the environment; The projection calculation module is used to project at least one vector feature from the vector map of the vehicle's environment onto the plane where the image data of the environment is located, to obtain the projection feature corresponding to the vector feature. The vector feature is used to represent the geometric structure of the environmental object in the environment, and the image content of the image data includes the environmental object. An error tolerance module is used to perform an offset operation on the projected feature and generate at least one candidate checkpoint corresponding to the offset projected feature based on the offset projected feature, wherein the candidate checkpoint is used to characterize the degree of offset of the vector feature when the vector feature is projected onto the plane; The consistency verification module is used to check the candidate checkpoints using the category attributes of the candidate checkpoints and the category attributes of the vector features, and obtain the check results, wherein the check results are used to represent the visibility of the candidate checkpoints. The visibility determination module is used to determine the visibility level of the vector element based on the inspection results; A tag generation module is used to generate tag information for the environmental object based on the visibility level, wherein the tag information is used to indicate the degree to which the environmental object is occluded.

13. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 11.

Images