Lane line topology connection marking method and system

Abstract

The invention discloses a lane line topological connection marking method and system, and relates to the technical field of lane line topological connection marking, and the method comprises the steps: S1, data collection and preprocessing; s2, lane line detection and tracking; s3, constructing a topological relation of the lane lines; s4, labeling and verification are carried out; and S5, outputting and applying. According to the lane line topological connection marking method and system, manual participation is not needed, lane line detection and tracking are carried out based on the three-dimensional data of the collected road scene, the lane line topological relation is constructed according to the lane line detection and tracking, lane line topological connection marking is automatically generated, meanwhile, the automatic verification function is achieved, and the working efficiency is improved. Therefore, efficient and accurate lane line topological structure marking is achieved, compared with the tedious steps of manual point marking and then connection in a traditional method, the efficiency and precision of lane line topological connection marking are obviously improved, the method can be well suitable for the road environment with the complex lane line connection condition, and wide application prospects and commercial value are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of lane line topology connection labeling, and in particular to a lane line topology connection labeling method and a system thereof. Background Art

[0002] With the rapid development of autonomous driving technology, high-precision maps play a vital role in autonomous driving systems. High-precision maps not only provide geometric information of roads, but also contain detailed information such as lane lines and traffic signs. Among this information, the annotation of lane line topology is particularly important, which directly affects the path planning and decision-making capabilities of autonomous vehicles. Traditional lane line topology connection annotation methods mainly rely on manual annotation. However, since manual annotation usually relies on straight line connections between the marked discrete path points, and the distribution of path points is often not dense enough, curvature discontinuities are prone to occur at the path connection, resulting in an unsmooth path, affecting the driving decision of the autonomous vehicle and further affecting driving stability. And since manual annotation of lane line topology requires manual punctuation and connection of each lane line endpoint, this process is very time-consuming, especially for large-scale road networks. This method is very inefficient and cannot meet the needs of quickly generating high-precision maps.

[0003] Secondly, the operating standards and styles of different annotators may differ, and even the same annotator may produce inconsistent annotation results in different time periods. This inconsistency will affect the overall quality and reliability of high-precision maps. Although some annotation rules and standards can be formulated, due to the uncontrollability of human operations, the degree of standardization in the actual annotation process is still low, and it is difficult to ensure that the quality of the annotation results meets a unified standard.

[0004] In addition, in the actual road environment, the connection of lane lines may be very complex, such as multiple lanes merging or diverging, curves, intersections, etc. Manual labeling is often unable to handle these complex situations, and it is easy to have connection errors or unreasonable labeling results.

[0005] In summary, although the traditional manual annotation method can complete the task of annotating the lane topology structure, its low efficiency, easy errors, poor consistency, difficulty in handling complex situations, high maintenance and update costs, lack of standardization, and difficulty in data integration have greatly limited its application in map annotation. These defects urgently need to be solved by more automated, efficient and accurate methods to meet the high requirements of autonomous driving technology for high-precision maps.

[0006] Therefore, there is an urgent need to improve this shortcoming. The present invention studies and improves the existing technology and its shortcomings, and provides a lane line topology connection marking method and system. Summary of the invention

[0007] The purpose of the present invention is to provide a lane line topology connection labeling method and system thereof to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] On the one hand, a lane line topology connection labeling method is provided, comprising the following steps:

[0010] S1. Data collection and preprocessing: Collect three-dimensional data of road scenes through sensors and preprocess the collected data, including data cleaning, noise removal, format unification, etc.

[0011] S2. Lane detection and tracking: Use image processing technology and machine learning algorithms to accurately detect and track lanes on pre-processed data, including lane edge detection, line extraction, line segment fitting, and type classification.

[0012] S3, Lane line topology relationship construction: Based on the detected lane lines, the connection relationship between the lane lines is determined using the topological connection reasoning algorithm, and the topological relationship between the lane lines is constructed accordingly;

[0013] S4, labeling and verification: Based on the constructed lane line topology, the labeling generation algorithm is used to automatically generate lane line topology connection labels, clearly mark the starting point and end point of each lane line and the connection relationship with other lane lines, and automatically verify to ensure the accuracy and reliability of the labeling;

[0014] S5. Output and application: Output the marked lane line topology connection information into a standard format, such as JSON, XML, etc., and apply the lane line topology connection information to multiple fields such as autonomous driving, path planning, and traffic management.

[0015] Furthermore, in step S1, the sensors include a lidar and a camera, and the three-dimensional data collected by them include information such as the position, shape, and color of the lane lines. For the data collected by different sensors, timestamp alignment is performed using hard alignment or soft synchronization to ensure data synchronization.

[0016] Furthermore, in step S1, the preprocessing specifically includes:

[0017] Data cleaning: remove invalid data and reduce noise interference. Specific operations: check missing values, abnormal values ​​or duplicate values ​​in the data (these values ​​may be caused by sensor failure, occlusion, reflection, etc.), directly delete or replace invalid data with reasonable estimated values, and verify the integrity and consistency of the data to ensure that each data conforms to the expected format and range, for example, check whether the position coordinates of the lane line are within a reasonable range, whether the type of the lane line conforms to traffic regulations, etc.;

[0018] Noise removal: Reduce the interference of noise on lane detection. Specific operations: Use filtering algorithms (common filtering algorithms include Gaussian filtering, median filtering, statistical filtering, etc.) to smooth the 3D point cloud or image data to remove noise points. For 3D point cloud data, point cloud denoising, point cloud simplification and other technologies are used to reduce noise points. For example, use voxel grid filtering to divide the point cloud data into voxel grids, and then perform statistical analysis within each voxel to remove sparse or scattered noise points.

[0019] Format unification: Ensure that data collected from different sources or at different times have a unified format. Specific operations: Through operations such as coordinate transformation and coordinate alignment, the data collected by different sensors are converted into a unified coordinate system to ensure the spatial consistency of the data, and according to actual application requirements, the data is converted into a unified format, for example, converting 3D point cloud data into PCD (Point Cloud Data) format, converting image data into JPEG or PNG format, etc.

[0020] Furthermore, the step S2 specifically includes the following sub-steps:

[0021] S21, edge detection: use the Canny edge detection algorithm to extract the edge information of the lane line;

[0022] S22, line extraction: use the Hough transform algorithm to extract the straight or curved features of the lane line from the edge information;

[0023] S23, line segment fitting: using the least squares method or RANSAC algorithm fitting method to fit the extracted lane edge points into smooth line segments;

[0024] S24. Type classification: Use a convolutional neural network to classify the detected lane lines according to their shape, color and other features, and identify their types, such as single solid line, single dashed line, double solid line, etc.

[0025] Furthermore, the step S3 specifically includes the following sub-steps:

[0026] S31. Determine the lane center line: determine the center line of each lane according to the position and shape of the lane line;

[0027] S32, identifying laterally associated lanes: analyzing the relationship between adjacent lanes and identifying lanes that are laterally associated with the current lane;

[0028] S33, calculating the longitudinal overlap length: calculating the longitudinal overlap length between the center line of the lateral associated lane and the center line of the current lane according to the position of the center line of the lane;

[0029] S34. Determine the topological connection relationship: Based on the longitudinal overlap length and other geometric features of the lane lines (such as lane width, angle, etc.), and combined with the semantic information of the lane lines (such as lane type, traffic rules, etc.), introduce explainable reasoning methods such as TopoLogic, which comprehensively considers the characteristics of the lane lines in geometric space and semantic space to determine the topological connection relationship between the lane lines.

[0030] Furthermore, the topological connection relationship includes the merging, bifurcation and intersection of lane lines;

[0031] Lane merging: Lanes that were originally parallel or separated merge into one or a few lanes in a certain area. This topological connection relationship is common in the entrance ramps of highways, the intersection of main and auxiliary roads, and the narrowing of roads. In these areas, vehicles from different directions or different lanes need to merge into the same or a few lanes to continue driving;

[0032] Lane bifurcation: A lane splits into two or more lanes in a certain area. This topological connection relationship is common in highway exit ramps, main and auxiliary road separations, road widening areas, etc. In these areas, vehicles need to choose the appropriate lane to continue driving according to their driving direction;

[0033] Lane crossing: Two or more lanes intersect or stagger in a certain area. This topological connection relationship is common in road intersections, the intersection of upper and lower levels of overpasses, etc. In these areas, vehicles from different directions need to pass through the intersection area in an orderly manner according to the instructions of traffic lights, traffic signs and markings.

[0034] Furthermore, in step S4, the annotation generation algorithm includes a rule engine and template matching, wherein the rule engine generates annotation information according to predefined rules based on the detected lane lines and the inferred topological connection relationship, and the template matching uses the known lane line template to match the detected lane line to determine its type and position, and generate corresponding annotations.

[0035] Furthermore, in step S4, the automatic verification specifically includes geometric consistency verification, semantic consistency verification and cross-modal verification, and after the verification is completed, the effect of the automatic verification is quantitatively evaluated by calculating indicators such as accuracy, recall rate, F1 score, etc., and the annotated data that failed the verification is corrected;

[0036] The geometric consistency verification is used to check whether the marked lane lines maintain geometric consistency, such as the smoothness and continuity of the lines and the relative position relationship with other lane lines. The operation steps are as follows:

[0037] 1) Use geometric algorithms (such as least squares method, RANSAC, etc.) to fit the marked lane lines to obtain smooth curves or straight lines;

[0038] 2) Calculate the error (such as mean square error MSE) between the fitting curve and the original marked points to determine whether the error is within an acceptable range;

[0039] 3) Check whether there are unreasonable intersections, breaks or overlaps between lane lines;

[0040] The semantic consistency verification is used to check whether the marked lane lines are consistent with the actual traffic rules of the road, lane types and other semantic information. The operation steps are:

[0041] 1) Using a pre-defined traffic rules knowledge base or map information, perform semantic analysis on the marked lane lines;

[0042] 2) Compare the marked lane line types (such as straight, left turn, right turn, etc.) with the information in the traffic rules knowledge base to see if they are consistent;

[0043] 3) Verify whether the marked lane lines match the connection relationship of adjacent lanes (such as merging, bifurcation, etc.);

[0044] The cross-modal verification is used to cross-validate lane marking using multi-source data (such as images, point clouds, GPS trajectories, etc.) to improve the accuracy and robustness of the verification. The operation steps are as follows:

[0045] 1) Compare the marked lane lines with other modal data in the same scene (e.g., overlay the lane line markings with the image captured by the camera);

[0046] 2) Check whether the lane lines in different modal data are consistent, especially in complex scenarios such as occlusion and lighting changes;

[0047] 3) Use GPS trajectory data to verify whether the actual driving trajectory of the lane line is consistent with the marking.

[0048] On the other hand, a lane line topology connection labeling system is provided, which is applied to the lane line topology connection labeling method as described above. The system is applied to a server and includes:

[0049] A data acquisition module, which is responsible for collecting three-dimensional data of road scenes through sensors;

[0050] A data preprocessing module, which is responsible for preprocessing the collected data;

[0051] A detection and tracking module, which is responsible for detecting and tracking lane lines on the preprocessed data;

[0052] A topological relationship building module, which is responsible for determining the connection relationship between lane lines and building the topological relationship between lane lines accordingly;

[0053] An automatic labeling module, which is responsible for automatically generating lane line topology connection labels;

[0054] An automatic verification module, which is responsible for verifying the labeled data;

[0055] An information output module, which is responsible for outputting the marked lane line topology connection information into a standard format;

[0056] The information application module is responsible for applying lane line topology connection information to the fields of autonomous driving, path planning, and traffic management.

[0057] Furthermore, the system comprises:

[0058] A processor, used to execute computer instructions to enable the system to execute: a data acquisition module, a data preprocessing module, a detection and tracking module, a topological relationship building module, an automatic annotation module, an automatic verification module, an information output module and an information application module;

[0059] The memory is used to store computer instructions. A computer program is stored in the memory. When the computer program is executed by the processor, the method described above is implemented.

[0060] The present invention provides a lane line topology connection marking method and system thereof, which has the following beneficial effects:

[0061] The present invention proposes a fully automatic lane line topology connection labeling method. The entire process does not require human intervention. The method detects and tracks lane lines based on the collected three-dimensional data of the road scene, constructs lane line topology relationships based on this, automatically generates lane line topology connection labeling, and has an automatic verification function, thereby achieving efficient and accurate lane line topology structure labeling. Compared with the tedious steps of manual markup and reconnection in traditional methods, the efficiency and accuracy of lane line topology connection labeling are significantly improved, and the method can be well applied to road environments with complex lane line connection conditions, and has broad application prospects and commercial value. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of the steps of a lane line topology connection marking method according to the present invention;

[0063] Figure 2 This is a schematic diagram of the operation flow of step S2 of a lane line topology connection labeling system of the present invention;

[0064] Figure 3 This is a schematic diagram of the operation flow of step S3 of a lane line topology connection labeling system of the present invention;

[0065] Figure 4 This is a logic block diagram of a lane line topology connection marking system of the present invention. DETAILED DESCRIPTION

[0066] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0067] Embodiment 1

[0068] like Figure 1-Figure 3 As shown, a lane line topology connection labeling method includes the following steps:

[0069] S1. Data collection and preprocessing: The three-dimensional data of the road scene is collected by sensors, including laser radar and camera. In the specific implementation process, the high-precision ranging capability of the laser radar and the rich color and texture information of the camera are combined to achieve a comprehensive perception of the road scene. The collected three-dimensional data includes the position, shape, color and other information of the lane line. For the data collected by different sensors, the timestamps are aligned by hard alignment or soft synchronization to ensure the synchronization of the data. Then, the collected data is preprocessed. In this embodiment, the preprocessing specifically includes the following operations:

[0070] Data cleaning: Check the missing values, abnormal values ​​or duplicate values ​​in the data. For invalid data, delete it directly or replace it with a reasonable estimated value. Verify the integrity and consistency of the data to ensure that each piece of data conforms to the expected format and range. For example, check whether the location coordinates of the lane line are within a reasonable range and whether the type of the lane line conforms to traffic regulations.

[0071] Noise removal: Use Gaussian filtering, median filtering or statistical filtering to smooth the 3D point cloud or image data and remove noise points. For 3D point cloud data, point cloud denoising, point cloud simplification and other technologies are used to reduce noise points. For example, use voxel grid filtering to divide the point cloud data into voxel grids, and then perform statistical analysis within each voxel to remove sparse or scattered noise points.

[0072] Format unification: Through operations such as coordinate transformation and coordinate alignment, the data collected by different sensors are converted into a unified coordinate system to ensure the spatial consistency of the data. In addition, the data is converted into a unified format according to actual application requirements. For example, 3D point cloud data is converted into PCD (Point Cloud Data) format, and image data is converted into JPEG or PNG format.

[0073] S2, Lane Detection and Tracking: Use image processing technology and machine learning algorithms to accurately detect and track lanes on pre-processed data. Specifically, it includes the following sub-steps:

[0074] S21, edge detection: using the Canny edge detection algorithm to extract the edge information of the lane line; in this embodiment, the Canny edge detection algorithm calculates the amplitude and direction of the image gradient and uses a dual threshold algorithm to detect the edge, and the implementation steps are:

[0075] Image grayscale: Convert color images into grayscale images to simplify calculations;

[0076] Gaussian filtering: Use a Gaussian filter to smooth the grayscale image to reduce noise;

[0077] Gradient calculation: calculate the gradient magnitude and direction of the image;

[0078] Non-maximum suppression: retain the local maximum point in the gradient direction to refine the edge;

[0079] Dual threshold detection: set two thresholds, high and low, and determine the edge point based on the relationship between the gradient amplitude and the threshold;

[0080] S22, line extraction: extract the straight line or curve features of the lane line from the edge information using the Hough transform algorithm; in this embodiment, the Hough transform algorithm realizes line extraction by mapping points in the image space to lines in the parameter space (for straight line detection), and the implementation steps are as follows:

[0081] Parameter space transformation: transform points in the image to lines in the parameter space;

[0082] Peak detection: Find local peaks of the accumulator in parameter space, which correspond to straight lines in the image;

[0083] S23, line segment fitting: The extracted lane edge points are fitted into smooth line segments using the least squares method or RANSAC algorithm fitting method. The least squares method is applicable to data points with noise, and the best fitting line segment is found by minimizing the sum of squared errors. The implementation steps are as follows:

[0084] Construct model: Establish a straight line model y=ax+b;

[0085] Calculate parameters: Determine parameters a and b by solving the system of equations that minimize the sum of squared errors;

[0086] The RANSAC algorithm robustly estimates the model parameters and can obtain good fitting results even in the presence of a large number of outliers. The implementation steps are as follows:

[0087] Random sampling: randomly select a small number of data points as inliers;

[0088] Model estimation: Use these inliers to estimate model parameters;

[0089] Verify the model: calculate the matching degree of all data points with the current model and count the number of inliers;

[0090] Iterative optimization: Repeat the above steps multiple times and select the model with the largest number of inliers as the final model;

[0091] S24, type classification: Use convolutional neural network to classify the detected lane lines according to their shape, color and other features, and identify their types, such as single solid line, single dashed line, double solid line, etc.;

[0092] S3, lane line topology relationship construction: based on the detected lane lines, the connection relationship between the lane lines is determined by using the topological connection inference algorithm, and the topological relationship between the lane lines is constructed accordingly, which specifically includes the following sub-steps:

[0093] S31. Determine the lane center line: determine the center line of each lane according to the position and shape of the lane line;

[0094] S32, identifying laterally associated lanes: analyzing the relationship between adjacent lanes and identifying lanes that are laterally associated with the current lane;

[0095] S33, calculating the longitudinal overlap length: calculating the longitudinal overlap length between the center line of the lateral associated lane and the center line of the current lane according to the position of the center line of the lane;

[0096] S34. Determine the topological connection relationship: Based on the longitudinal overlap length and other geometric features of the lane lines (such as lane width, angle, etc.), combined with the semantic information of the lane lines (such as lane type, traffic rules, etc.), introduce an explainable reasoning method, such as TopoLogic, to comprehensively consider the characteristics of the lane lines in the geometric space and semantic space to determine the topological connection relationship between the lane lines; among them, the topological connection relationship includes the merging of lane lines (the originally parallel or separated lanes merge into one or a few lanes in a certain area. This topological connection relationship is common in the entrance ramps of highways, the intersection of main and auxiliary roads, and the narrowing of roads. In these areas, lane lines from different Vehicles in different directions or lanes need to merge into the same or a few lanes to continue driving), bifurcation (a lane line is separated into two or more lanes in a certain area. This topological connection relationship is common in the exit ramps of expressways, the separation of main and auxiliary roads, and the widening of roads. In these areas, vehicles need to choose the appropriate lane to continue driving according to their own driving direction), intersection (two or more lane lines intersect or stagger in a certain area. This topological connection relationship is common in road intersections, the upper and lower interweaving of overpasses, etc. In these areas, vehicles from different directions need to pass through the intersection area in an orderly manner according to the instructions of traffic lights, traffic signs and markings);

[0097] S4. Labeling and verification: Based on the constructed lane line topology, the labeling generation algorithm is used to automatically generate lane line topology connection labels, clearly marking the starting point and end point of each lane line and the connection relationship with other lane lines.

[0098] In this embodiment, the annotation generation algorithm includes a rule engine and template matching. The rule engine generates annotation information according to predefined rules based on the detected lane lines and the inferred topological connection relationship. The template matching uses a known lane line template to match the detected lane line, determine its type and position, and generate corresponding annotations.

[0099] Then, automatic verification is performed. In this embodiment, the automatic verification specifically includes geometric consistency verification, semantic consistency verification, and cross-modal verification. After the verification is completed, the effect of the automatic verification is quantitatively evaluated by calculating indicators such as accuracy, recall rate, and F1 score, and the labeled data that failed the verification is corrected;

[0100] Geometric consistency verification is used to check whether the marked lane lines maintain geometric consistency, such as the smoothness and continuity of the lines and their relative position to other lane lines. The operation steps are as follows:

[0101] 1) Use geometric algorithms (such as least squares method, RANSAC, etc.) to fit the marked lane lines to obtain smooth curves or straight lines;

[0102] 2) Calculate the error (such as mean square error MSE) between the fitting curve and the original marked points to determine whether the error is within an acceptable range;

[0103] 3) Check whether there are unreasonable intersections, breaks or overlaps between lane lines;

[0104] Semantic consistency verification is used to check whether the marked lane lines are consistent with the actual traffic rules, lane types and other semantic information of the road. The operation steps are as follows:

[0105] 1) Using a pre-defined traffic rules knowledge base or map information, perform semantic analysis on the marked lane lines;

[0106] 2) Compare the marked lane line types (such as straight, left turn, right turn, etc.) with the information in the traffic rules knowledge base to see if they are consistent;

[0107] 3) Verify whether the marked lane lines match the connection relationship of adjacent lanes (such as merging, bifurcation, etc.);

[0108] Cross-modal verification is used to cross-validate lane marking using multi-source data (such as images, point clouds, GPS trajectories, etc.) to improve the accuracy and robustness of verification. The steps are as follows:

[0109] 1) Compare the marked lane lines with other modal data in the same scene (e.g., overlay the lane line markings with the image captured by the camera);

[0110] 2) Check whether the lane lines in different modal data are consistent, especially in complex scenarios such as occlusion and lighting changes;

[0111] 3) Use GPS trajectory data to verify whether the actual driving trajectory of the lane line is consistent with the marking;

[0112] S5. Output and application: Output the marked lane line topology connection information into a standard format, such as JSON, XML, etc., and apply the lane line topology connection information to multiple fields such as autonomous driving, path planning, and traffic management.

[0113] Embodiment 2

[0114] like Figure 4As shown, a lane line topology connection labeling system is applied to the lane line topology connection labeling method as described above. The system is applied to a server and includes:

[0115] Data acquisition module, which is responsible for collecting three-dimensional data of road scenes through sensors;

[0116] Data preprocessing module, the data preprocessing module is responsible for preprocessing the collected data;

[0117] Detection and tracking module,The detection and tracking module is responsible for detecting and tracking lane lines on the preprocessed data;

[0118] A topological relationship building module is responsible for determining the connection relationship between lane lines and building the topological relationship between lane lines accordingly;

[0119] Automatic labeling module, which is responsible for automatically generating lane line topology connection labels;

[0120] Automatic verification module, which is responsible for verifying the labeled data;

[0121] Information output module, which is responsible for outputting the marked lane line topology connection information into a standard format;

[0122] Information application module: The information application module is responsible for applying lane line topology connection information to the fields of autonomous driving, path planning, and traffic management;

[0123] A processor, used to execute computer instructions to enable the system to execute: a data acquisition module, a data preprocessing module, a detection and tracking module, a topological relationship building module, an automatic annotation module, an automatic verification module, an information output module and an information application module;

[0124] The memory is used to store computer instructions. A computer program is stored in the memory. When the computer program is executed by the processor, the lane line topology connection labeling method as described in the first embodiment is implemented.

[0125] The embodiments of the present invention are given for the purpose of illustration and description, and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

Claims

1. A lane line topology connection labeling method, characterized in that: The following steps are involved: S1. Data collection and preprocessing: Collect three-dimensional data of road scenes through sensors and preprocess the collected data; S2. Lane detection and tracking: Use image processing technology and machine learning algorithms to accurately detect and track lanes on pre-processed data, including lane edge detection, line extraction, line segment fitting, and type classification. S3, Lane line topology relationship construction: Based on the detected lane lines, the connection relationship between the lane lines is determined using the topological connection reasoning algorithm, and the topological relationship between the lane lines is constructed accordingly; S4, labeling and verification: Based on the constructed lane line topology relationship, the lane line topology connection labeling is automatically generated using the labeling generation algorithm, and automatic verification is performed to ensure the accuracy and reliability of the labeling; S5. Output and application: Output the marked lane line topology connection information into a standard format, and apply the lane line topology connection information to the fields of autonomous driving, path planning, and traffic management.

2. A lane line topology connection labeling method according to claim 1, characterized in that: In step S1, the sensors include a laser radar and a camera, and the three-dimensional data collected by the sensors include the position, shape, and color information of the lane line. For the data collected by different sensors, timestamp alignment is performed by hard alignment or soft synchronization.

3. A lane line topology connection labeling method according to claim 1, characterized in that: In step S1, the preprocessing specifically includes: Data cleaning: remove invalid data and reduce noise interference. Specific operations: check missing values, abnormal values ​​or duplicate values ​​in the data. For invalid data, directly delete or replace them with reasonable estimated values, and verify the integrity and consistency of the data to ensure that each piece of data conforms to the expected format and range; Noise removal: Reduce the interference of noise on lane detection. Specific operation: Use filtering algorithm to smooth the 3D point cloud or image data to remove noise points. For 3D point cloud data, use point cloud denoising and point cloud simplification technology to reduce noise points. Format unification: Ensure that data collected from different sources or at different times have a unified format. Specific operations: Convert data collected by different sensors into a unified coordinate system, and convert data into a unified format according to actual application requirements.

4. A lane line topology connection labeling method according to claim 1, characterized in that: The step S2 specifically includes the following sub-steps: S21, edge detection: use the Canny edge detection algorithm to extract the edge information of the lane line; S22, line extraction: use the Hough transform algorithm to extract the straight or curved features of the lane line from the edge information; S23, line segment fitting: The extracted lane edge points are fitted into smooth line segments using the least squares method or RANSAC algorithm fitting method. S24. Type classification: Use a convolutional neural network to classify the detected lane lines according to their features and identify their types.

5. The lane line topology connection labeling method according to claim 1, characterized in that: The step S3 specifically includes the following sub-steps: S31. Determine the lane center line: determine the center line of each lane according to the position and shape of the lane line; S32, identifying laterally associated lanes: analyzing the relationship between adjacent lanes and identifying lanes that are laterally associated with the current lane; S33, calculating the longitudinal overlap length: calculating the longitudinal overlap length between the center line of the lateral associated lane and the center line of the current lane according to the position of the center line of the lane; S34. Determine the topological connection relationship: Based on the longitudinal overlapping length and other geometric features of the lane lines, and combined with the semantic information of the lane lines, an explainable reasoning method is introduced to comprehensively consider the characteristics of the lane lines in the geometric space and the semantic space to determine the topological connection relationship between the lane lines.

6. A lane line topology connection labeling method according to claim 5, characterized in that: The topological connection relationship includes the merging, bifurcation and intersection of lane lines, and merging means that the originally parallel or separated lane lines merge into one or a few lanes in a certain area, bifurcation means that a lane line is separated into two or more lanes in a certain area, and intersection means that two or more lane lines intersect or stagger in a certain area.

7. A lane line topology connection labeling method according to claim 1, characterized in that: In step S4, the annotation generation algorithm includes a rule engine and template matching, wherein the rule engine generates annotation information according to predefined rules based on the detected lane lines and the inferred topological connection relationship, and the template matching uses the known lane line template to match the detected lane line to determine its type and position, and generate corresponding annotations.

8. The lane line topology connection labeling method according to claim 1, characterized in that: In step S4, the automatic verification specifically includes geometric consistency verification, semantic consistency verification and cross-modal verification. After the verification is completed, the effect of the automatic verification is quantitatively evaluated by calculating indicators, and the annotated data that failed the verification is corrected; The geometric consistency verification is used to check whether the marked lane lines maintain geometric consistency. The operation steps are: 1) Use geometric algorithms to fit the marked lane lines to obtain smooth curves or straight lines; 2) Calculate the error between the fitting curve and the original marked points to determine whether the error is within an acceptable range; 3) Check whether there are unreasonable intersections, breaks or overlaps between lane lines; The semantic consistency verification is used to check whether the marked lane lines are consistent with the actual traffic rules of the road and the semantic information of the lane type. The operation steps are as follows: 1) Using a pre-defined traffic rules knowledge base or map information, perform semantic analysis on the marked lane lines; 2) Compare the marked lane line type with the information in the traffic rules knowledge base to see if they are consistent; 3) Verify whether the marked lane lines match the connection relationship of adjacent lanes; The cross-modal verification is used to cross-validate lane marking using multi-source data. The operation steps are as follows: 1) Compare the marked lane lines with other modal data in the same scene; 2) Check whether the lane lines in different modal data are consistent, especially in complex scenarios such as occlusion and lighting changes; 3) Use GPS trajectory data to verify whether the actual driving trajectory of the lane line is consistent with the marking.

9. A lane line topology connection labeling system, applied to the lane line topology connection labeling method according to any one of claims 1 to 8, characterized in that: Applied to a server, the system comprises: A data acquisition module, which is responsible for collecting three-dimensional data of road scenes through sensors; A data preprocessing module, which is responsible for preprocessing the collected data; A detection and tracking module, which is responsible for detecting and tracking lane lines on the preprocessed data; A topological relationship building module, which is responsible for determining the connection relationship between lane lines and building the topological relationship between lane lines accordingly; An automatic labeling module, which is responsible for automatically generating lane line topology connection labels; An automatic verification module, which is responsible for verifying the labeled data; An information output module, which is responsible for outputting the marked lane line topology connection information into a standard format; The information application module is responsible for applying lane line topology connection information to the fields of autonomous driving, path planning, and traffic management.

10. A lane line topology connection labeling system according to claim 9, characterized in that: include: A processor, used to execute computer instructions to enable the system to execute: a data acquisition module, a data preprocessing module, a detection and tracking module, a topological relationship building module, an automatic annotation module, an automatic verification module, an information output module and an information application module; A memory for storing computer instructions, wherein a computer program is stored in the memory, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images