Automatic road information labeling method and labeling all-in-one machine intelligent agent

Abstract

The invention belongs to the technical field of road information labeling, and particularly discloses an automatic labeling method for road information and a labeling all-in-one machine agent, and the method comprises the steps: receiving original road data submitted by a user through a Web interface, and carrying out the encryption, standardization and quality verification of the data in sequence; positioning and fusing the data subjected to standardization processing and quality verification with a preset static road base map, automatically extracting road elements based on a fusion result to generate structured annotation data, and simultaneously extracting the structured annotation data based on the fusion result of the data and the static road base map; performing reverse optimization on camera-by-camera pose parameters on the equipment for acquiring the original road data; and based on the optimized pose parameters, performing automatic screening on the generated structured annotation data, converting the screened data meeting the requirements into a standard vector format, and performing encryption output. According to the invention, automatic marking of road information can be completed efficiently and accurately, and data security is ensured.

Description

Technical Field

[0001] This application belongs to the field of road information labeling technology, and more specifically, relates to an automated road information labeling method and an integrated labeling machine intelligent agent. Background Technology

[0002] Currently, the acquisition and processing of road information is crucial for the development of Advanced Driver Assistance Systems (ADAS) and autonomous driving technology. In existing technologies, the collection and mapping of road information usually rely on manual annotation or processing methods with limited automation. These methods generally suffer from the dilemma of balancing accuracy and efficiency: (1) Insufficient accuracy: Road data (such as point clouds and images) acquired by ordinary acquisition equipment is difficult to meet the requirements of high-precision maps, especially in complex urban scenarios, where the accuracy of positioning and element recognition needs to be improved; (2) Low efficiency: Large-scale road data annotation work is time-consuming and labor-intensive, with high labor costs, and it is difficult to ensure the consistency and accuracy of annotation; (3) Calibration error: The calibration parameters between sensors (such as lidar and cameras) on the acquisition vehicle are easily affected by environmental factors, vibration, etc., resulting in errors, which in turn affect the accuracy and projection effect of multi-sensor fusion data; (4) Complex data processing flow and security risks: From raw data to usable structured road information, multiple processing links are involved, the degree of automation of the process is not high, and the data standardization and verification process is cumbersome. At the same time, the transmission and storage of geospatial data face security and confidentiality risks.

[0003] Therefore, how to efficiently and accurately automate the labeling of road information while ensuring data security is an urgent problem to be solved. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this application is to provide an automated road information labeling method and an integrated labeling machine intelligent agent, which can efficiently and accurately complete the automated labeling of road information and ensure data security.

[0005] To achieve the above objectives, firstly, this application provides an automated annotation method for road information, comprising the following steps: S10, receive raw road data submitted by the user through the Web interface, and encrypt the raw road data; the raw road data is at least one of trajectory, point cloud and image data acquired by the vehicle-mounted acquisition device; S20, standardizes and verifies the quality of encrypted data; S30, the standardized and quality-verified data is located and fused with a preset static road base map, and road elements are automatically extracted based on the fusion result to generate structured labeled data. At the same time, based on the fusion result of the data and the static road base map, the camera pose parameters on the device that collected the original road data are optimized in reverse. The preset static road base map includes at least one key static road element among lane lines, curbs and traffic signs. S40, based on the optimized pose parameters, automatically screen the generated structured annotation data, and convert the screened data that meets the requirements into a standard vector format before encrypting and outputting it.

[0006] The automated road information labeling method provided in this application has the following effects: (1) By encrypting the data when it is received and encrypting it again at the output stage, combined with the Web interface accessed by the internal network, the security and confidentiality of geospatial data in the entire process of transmission, storage and use can be ensured, eliminating compliance risks from the source and the outlet; (2) By automating the positioning, fusion and feature extraction of the collected data and the preset high-precision static road base map, the traditional method of relying on manual identification and labeling can be replaced, and structured labeling data can be directly generated, greatly reducing manual operation links and time consumption, realizing a high degree of automation of the processing process, thereby significantly improving labeling efficiency; (3) Furthermore, by introducing a reverse optimization mechanism based on the fusion result in the data processing process, the camera pose parameters of the collection device can be dynamically optimized, correcting the accuracy deviation caused by calibration error or environmental factors from the source, and applying this optimized parameter to the subsequent automated screening link, so that the system can more accurately identify and remove the erroneous labeling results caused by inaccurate pose, thereby improving the processing speed while ensuring that the final output vector labeling data has higher accuracy and reliability. The encrypted security mechanism, automated processing flow, and built-in accuracy optimization feedback mechanism work together to achieve efficient, accurate, and secure automated labeling of road information.

[0007] As a further preferred embodiment, in step S30, the static road base map is obtained through the following steps: Raw mapping data of the target road area is collected using at least one sensor from among the Global Navigation Satellite System receiver, inertial measurement unit, lidar, and camera mounted on a professional surveying vehicle. The original surveying data is preprocessed by time synchronization, coordinate system unification, and noise filtering. The preprocessed data is fused and solved to generate intermediate road data containing 3D point cloud and georeferenced image; From the road center data, identify and vectorize the contours and position information of lane lines, curbs, traffic sign poles and traffic sign panels; The vectorized road element information is associated with absolute geographic coordinates and formatted according to preset data specifications to generate the static road base map.

[0008] As a further preferred embodiment, in step S30, the step of reverse optimization of the camera-by-camera pose parameters on the device acquiring the original road data specifically includes: Feature points are extracted from image frames in data that has undergone standardization and quality verification. The feature points are matched with the reference image features of the corresponding geographical locations in the static road base map to obtain multiple matching point pairs; Based on the matching point pairs, a reprojection error function is constructed regarding the camera pose parameters. The reprojection error function represents the difference between the position of the feature point projected onto the image plane under the current pose parameters and the position of its matching point in the reference image. The camera pose parameters are iteratively adjusted using a nonlinear optimization algorithm to minimize the reprojection error function, thereby obtaining the optimized camera-by-camera pose parameters.

[0009] As a further preferred embodiment, in step S40, the automated screening specifically includes: establishing a data index for the structured annotation data; automatically determining the conformity of the structured annotation data with preset rules based on the data index; and automatically removing structured annotation data that does not meet the requirements due to changes in the actual road scene or inaccurate projection calibration caused by the pose parameters before optimization.

[0010] As a further preferred embodiment, in step S40, the step of converting to a standard vector format specifically involves converting the road elements that meet the requirements after screening into a standard vector data format according to preset specifications. As a further preferred embodiment, after step S40, the following is also included: S50 provides a quality control toolchain that includes manual verification and confirmation functions to process data that is questionable after automated screening.

[0011] As a further preferred embodiment, after step S40, the following is also included: S60 performs automated testing on encrypted output standard vector format data to ensure it meets preset quality and format requirements.

[0012] As a further preferred embodiment, in step S10, the received original road data is encrypted, specifically through the built-in geographic information encryption / decryption module.

[0013] Secondly, this application provides an intelligent agent for labeling all-in-one machines, including: The central processing unit (CPU) provides general computing capabilities and executes operating systems, web services, and task scheduling. A graphics processing unit, connected to the central processing unit, is used to provide parallel computing capabilities to accelerate point cloud processing, image analysis, localization and calibration optimization; The memory system, connected to the central processing unit and the graphics processing unit, includes memory and a storage device for persistently storing data, for storing software programs, static road base maps, received raw road data, intermediate processing data and final generated result data; The data input / output interface is connected to the central processing unit and is used to access the internal network to provide Web service interaction and data transmission. When the central processing unit executes program instructions stored in the memory system, it implements the automated labeling method for road information as described in any of the above-mentioned methods.

[0014] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements an automated labeling method for road information as described in any one of the above. It is understood that the beneficial effects of the second and third aspects mentioned above can be found in the relevant descriptions in the first aspect above, and will not be repeated here. Attached Figure Description

[0015] Figure 1 This is a flowchart of the automated road information annotation method provided in this application; Figure 2 This is an overall system flowchart of the annotation all-in-one machine intelligent agent provided in the embodiments of this application; Figure 3 This is a schematic diagram of the hardware structure and interaction of the annotation all-in-one machine intelligent agent provided in the embodiments of this application; Figure 4 This is a diagram showing the overall workflow and data flow of the internal software system of the annotation all-in-one machine provided in this application embodiment. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0017] like Figure 1 As shown, this application provides an automated annotation method for road information, including steps S10 to S40, which are detailed below: Step S10: Receive the raw road data submitted by the user through the web interface and encrypt the raw road data.

[0018] In step S10, the raw road data can be at least one of trajectory, point cloud, and image data acquired by the vehicle-mounted data acquisition device. The system provides a web-based graphical user interface, allowing authorized users to submit data conveniently and securely through internal network security. Data is encrypted upon entering the system to ensure its security and compliance throughout subsequent storage and processing.

[0019] Step S20: Standardize and verify the quality of the encrypted data.

[0020] In step S20, the system standardizes the data to ensure that its format and content meet the requirements of subsequent processing, and performs automatic verification to ensure the accuracy of data processing.

[0021] Step S30 involves locating and fusing the standardized and quality-verified data with a preset static road base map, automatically extracting road features based on the fusion result to generate structured labeled data, and simultaneously optimizing the camera pose parameters on the device that acquired the original road data based on the fusion result of the data and the static road base map.

[0022] In step S30, the preset static road base map may include at least one key static road element among lane lines, curbs, and traffic signs. The system performs precise spatial alignment and fusion of the collected data with the high-precision base map. Based on this high-precision positioning, it can automatically extract key static road elements from the base map and convert them into structured labeled data. Simultaneously, the system can intelligently optimize the camera-by-camera pose on the acquisition device using the precise matching results between the collected data and the static base map, achieving pixel-level beyond-line-of-sight labeling accuracy.

[0023] Specifically, a static road base map can be obtained through the following steps: First, raw mapping data of the target road area is collected collaboratively by multiple sensors mounted on a specialized surveying vehicle. These sensors include at least a Global Navigation Satellite System (GNSS) receiver, an Inertial Measurement Unit (IMU), a lidar system, and a camera. The GNSS receiver and IMU together provide continuous high-frequency position and attitude information for the vehicle; the lidar emits laser pulses and receives the echoes, acquiring dense 3D point cloud data of the road and surrounding environment; and the camera simultaneously acquires high-resolution sequential images as the vehicle moves. This heterogeneous data is typically initially synchronized during acquisition using hardware triggering or software timestamps.

[0024] Next, the collected raw mapping data undergoes rigorous preprocessing. The core tasks of preprocessing include: unifying data streams from different sensors to the same time reference using time interpolation or alignment algorithms to achieve precise time synchronization; unifying the coordinate systems of all data, such as LiDAR point clouds and image pixels, to a global geographic coordinate system (e.g., WGS-84) or a local projected coordinate system to form a unified spatial reference frame; furthermore, filtering the point cloud data to remove stray points and noise, and adjusting the photometric properties of the image data to mitigate the impact of illumination variations, thereby improving data quality.

[0025] Then, the preprocessed multi-source data is fused and solved to generate highly reliable intermediate road data. This process typically involves tightly coupled or loosely coupled fusion algorithms. Using the precise pose provided by GNSS / IMU as initial values, each frame of point cloud obtained from LiDAR scanning is transformed to the global coordinate system and stitched together to form a complete and coherent 3D point cloud map. Simultaneously, based on the synchronized pose information, camera images are corrected and projected onto this 3D scene to generate orthophotos or panoramic images with accurate georeferenced information. The final output is intermediate road data containing 3D point clouds and georeferenced images, which serves as a reliable foundation for subsequent feature extraction.

[0026] Subsequently, key static road elements are automatically identified and vectorized from the fused intermediate road data. For 3D point clouds, point clusters belonging to lane lines, curbs, traffic sign poles, etc., can be identified through point cloud segmentation and classification algorithms; for georeferenced images, targets such as traffic sign panels can be detected using computer vision algorithms. After identifying these elements, the geometric contours and spatial location information of lane lines, curbs, traffic sign poles, and traffic sign panels are accurately obtained through model fitting (such as fitting lane line point clouds into Bézier curves or polylines), contour extraction, and vectorization, and then converted into vector format.

[0027] Finally, all road feature information extracted from vectorization is stably correlated with absolute geographic coordinates and formatted and encapsulated according to preset data specifications to generate the static road base map required by the system. This step includes assigning a unique identifier to each feature, recording its type, geometry, precise coordinates, attributes (such as lane line type, sign content), and organizing it into a specific data file or database format. This base map is pre-loaded into the system as high-precision prior knowledge, providing a benchmark for subsequent automated positioning, fusion, and annotation.

[0028] Specifically, the step of reverse optimization of the camera-by-camera pose parameters on the device that acquires the original road data can be as follows: First, from single or multiple frames of images in the standardized and quality-verified data, visually significant and discriminative feature points are extracted. Specifically, feature extraction algorithms (such as SIFT, ORB, or SURF) can be used to process the images, identifying key pixel locations such as corners and edge intersections, and calculating their corresponding feature descriptors. These feature points constitute the foundational data for subsequent geometric matching.

[0029] Next, the feature points (and descriptors) extracted from the acquired images in the previous step are matched with the reference image features of the corresponding geographic area in the static road base map. The static road base map typically contains pre-prepared orthophotos or feature libraries with precise geographic coordinates. The matching process is performed by calculating the similarity between feature descriptors (such as Euclidean distance) to find the correspondence between the features of the acquired image and the reference features of the base map, thus obtaining a series of preliminary matching point pairs. To improve the reliability of the matching, a robust algorithm (such as RANSAC) can be used to filter the preliminary matching results, eliminating false matches, and finally obtaining a set of high-confidence matching point pairs. Each valid matching point pair represents the correspondence between a three-dimensional spatial point (whose coordinates can be obtained from the base map) and its two-dimensional projection point in the acquired image.

[0030] Then, based on the filtered reliable matching point pairs, a reprojection error function for the camera pose parameters is constructed. The camera pose parameters typically include the camera's 3D position (translation vector) and 3D rotation attitude (rotation matrix or quaternion representation) in the global coordinate system. For each matching point pair, it is assumed that the coordinates of the 3D point are... Pw The two-dimensional pixel coordinates observed in the image are pobs Based on the camera imaging model, Pw Using the current camera pose parameter estimates [ R | t Projecting the image onto the image plane yields the calculated coordinates of the two-dimensional projection points. pcalc Reprojection error is defined as the difference between the observed position and the calculated projected position, and is usually expressed as the square of the Euclidean distance, i.e. e =∥ pobs pcalc ∥ 2 The summation of the reprojection errors of all matching point pairs constitutes the overall objective optimization function. The physical meaning of this function intuitively reflects the overall degree of agreement between the projected positions of 3D map points and their actual shooting positions under the currently estimated camera pose parameters. The smaller the error, the higher the degree of agreement, and the more accurate the pose estimation.

[0031] Finally, a nonlinear optimization algorithm is used to iteratively adjust the camera pose parameters to minimize the aforementioned reprojection error function. Since the relationship between the camera projection model and the pose parameters is nonlinear, mature nonlinear least squares optimizers such as the Levenberg-Marquardt algorithm or the Gauss-Newton method are typically used for solving the problem. At the start of the optimization process, the initial pose information accompanying the acquired data (such as positioning and orientation data from the vehicle-mounted GNSS / IMU) is used as the initial value for iteration. The optimizer calculates the Jacobian matrix of the objective function with respect to the pose parameters, determining an increment direction and step size in each iteration to update the pose parameters. After multiple iterations, the algorithm converges when the parameter update amount is less than a preset threshold or the decrease in the objective function is no longer significant. The pose parameters obtained at this point are the optimized results. This optimized per-camera pose parameter essentially uses a high-precision static base map as an "absolute truth" reference to precisely correct and refine the sensor extrinsic parameters (i.e., the camera's position and attitude relative to the vehicle coordinate system, or its pose directly in the world coordinate system) at the original acquisition time.

[0032] Through the series of computational steps described above, from feature matching to model optimization, the system can automatically and accurately inversely calculate better camera pose parameters. These parameters can be directly used to improve the performance of the same batch of acquired data in subsequent stages (such as multi-view simulations). Figure 3 The geometric consistency of the data (dimensional reconstruction, target projection) can be used as prior knowledge to optimize the future data acquisition accuracy of the acquisition device, thereby fundamentally improving the input data quality and the absolute accuracy of the final result of the entire automated annotation process.

[0033] Step S40: Based on the optimized pose parameters, the generated structured annotation data is automatically screened, and the screened data that meets the requirements is converted into a standard vector format and then encrypted and output.

[0034] In step S40, automated screening may include establishing an efficient index for the processed data, automatically determining the conformity of the results, and automatically removing non-compliant results due to changes in reality or inaccurate projection calibration. The system converts the processed road elements into a standard vector data format according to preset specifications and encrypts the results in the output stage to achieve end-to-end data security.

[0035] The automated road information labeling method provided in this application has the following effects: (1) By encrypting the data when it is received and encrypting it again at the output stage, combined with the Web interface accessed by the internal network, the security and confidentiality of geospatial data in the entire process of transmission, storage and use can be ensured, eliminating compliance risks from the source and the outlet; (2) By automating the positioning, fusion and feature extraction of the collected data and the preset high-precision static road base map, the traditional method of relying on manual identification and labeling can be replaced, and structured labeling data can be directly generated, greatly reducing manual operation links and time consumption, realizing a high degree of automation of the processing process, thereby significantly improving labeling efficiency; (3) Furthermore, by introducing a reverse optimization mechanism based on the fusion result in the data processing process, the camera pose parameters of the collection device can be dynamically optimized, correcting the accuracy deviation caused by calibration error or environmental factors from the source, and applying this optimized parameter to the subsequent automated screening link, so that the system can more accurately identify and remove the erroneous labeling results caused by inaccurate pose, thereby improving the processing speed while ensuring that the final output vector labeling data has higher accuracy and reliability. The encrypted security mechanism, automated processing flow, and built-in accuracy optimization feedback mechanism work together to achieve efficient, accurate, and secure automated labeling of road information.

[0036] In one embodiment, the technical solution to achieve the above objective can be as follows: This embodiment aims to provide a data integration and processing system for an integrated labeling machine, its supporting electronic device, and related computer-readable storage media. This embodiment is dedicated to solving core problems in the prior art such as low automation of labeling and quality inspection, insufficient labeling efficiency and accuracy, inconsistent labeling results, and compliance of geospatial data storage and use.

[0037] The annotation all-in-one intelligent agent provided in this embodiment is a highly integrated dedicated electronic device. Its hardware core integrates a high-performance central processing unit (CPU) and a graphics processing unit (GPU) to provide powerful computing capabilities, and is equipped with a large-capacity storage system. This device has network communication capabilities and can easily connect to the user's internal network environment.

[0038] Data Processing System and its Core Operating Mechanism: The data processing system of this invention is deployed and runs on the intelligent agent hardware platform of this annotation all-in-one machine. This system, acting as the "brain" of the intelligent agent, endows it with automated processing and analysis capabilities. Its core operating mechanism is manifested as follows: Secure and convenient web-based interaction and data access: The system provides a web-based graphical user interface. Authorized users can access this web page via the internal network to securely and conveniently submit raw road data (such as trajectories, point clouds, and images) acquired by ordinary vehicle-mounted data collection equipment. Upon entering the system, the data is encrypted using a built-in geographic information encryption / decryption module, ensuring compliance throughout the subsequent storage and processing.

[0039] Intelligent positioning and feature extraction based on high-precision benchmarks: The system is pre-loaded or loaded with ultra-high-precision encrypted static road base maps. After receiving user data, the system automatically invokes the high-precision positioning function to accurately align and fuse the user-submitted collected data with the high-precision base map in space. Based on this high-precision positioning, the system further automatically extracts key static road features (such as lane lines, curbs, traffic signs, etc.) from the base map and converts them into a structured labeled data format.

[0040] Feedback-based sensor calibration self-optimization: The system can intelligently optimize the camera pose on the acquisition device by using the pixel-level accurate matching results of ordinary acquisition data and high-precision static base map, thereby achieving pixel-level labeling accuracy beyond line of sight.

[0041] A toolchain combining automated and manual quality inspection.

[0042] Automated Processing and Secure Output: After the aforementioned series of intelligent processing steps, the system can automatically generate accurate and compliant static road data. During the data output phase, the geographic information encryption / decryption module can also be invoked to encrypt the results, ensuring end-to-end data security.

[0043] The annotation all-in-one intelligent agent provided in this embodiment achieves the above functions by integrating a sophisticated automated data processing workflow, as shown in Table 1. The main steps include: secure reception of collected data through a web interface, standardization and quality verification of raw data, high-precision pixel-level positioning, automated screening (including efficient index creation, automatic judgment of result conformity, automatic removal of real-world changes and inaccurate projection calibration results), screening toolchain, generation and secure output of structured vector data, and automated testing and convenient delivery of the final deliverables.

[0044] Table 1

[0045] The electronic device (an integrated labeling machine intelligent agent) provided in this embodiment is a highly integrated, high-performance dedicated hardware unit designed to provide a powerful operating platform for data processing systems. Its core hardware components are shown in Table 2: Table 2

[0046] All the aforementioned hardware components are tightly integrated to form a collaborative whole. The processor executes data processing system program instructions (including Web service applications and geographic information security processing functions) stored in memory, driving the entire automated process and achieving high-precision, high-efficiency, and high-security road information processing.

[0047] Figure 2 This is an overall system flowchart provided for an embodiment of this application. The diagram illustrates the overall high-level process of the annotation all-in-one machine intelligent agent from user interaction to final output, highlighting key internal components and their interaction with external users. Figure 3 This is a schematic diagram of the hardware structure and interaction provided in an embodiment of this application. The diagram clearly illustrates the physical hardware composition of the "annotation all-in-one intelligent agent" and its direct interaction with the software system and the external environment. The geographic information security processing unit can be an independent hardware module, or its functions can be implemented in software via the main CPU / GPU. Figure 4 This is a software system workflow and data flow diagram provided in this application embodiment. The diagram details the overall workflow of the internal software system of the annotation all-in-one machine intelligent agent and clearly shows the data flow path and decision logic between various processing modules.

[0048] The key point of this embodiment is: (1) Comprehensive functions and optimization integration of the labeling all-in-one machine intelligent agent: A labeling all-in-one machine intelligent agent integrating a high-performance hardware platform and an efficient data processing system is provided. The intelligent agent integrates multiple functions such as data acquisition, data processing, sensor optimization, data encryption and secure output, and provides one-stop road information processing services through a unified hardware platform and software system.

[0049] (2) Hardware design of the labeling machine intelligent agent (device core): An integrated labeling machine intelligent agent is provided, which includes a central processing unit (CPU) and a graphics processing unit (GPU) working together to provide powerful computing power, and is equipped with a hardware platform with large-capacity storage and network communication capabilities. This hardware platform can efficiently process the raw road data from ordinary collection vehicle equipment and automatically execute data processing tasks. In addition, the provided labeling machine intelligent agent can run directly in the intranet environment, and further integrates a geographic information encryption and decryption module, which can conveniently ensure the security and confidentiality of raw data in any data reception, storage, processing and output process.

[0050] (3) Intelligent processing mechanism of data processing system: The data processing process is integrated and adapted to the hardware design, which can efficiently perform automatic data processing, labeling and quality inspection, and convert it into a structured data format.

[0051] The beneficial effects of this embodiment are: (1) High precision and high reliability: This invention achieves a one-stop high-precision annotation system for collected data by efficiently integrating a pixel-level positioning system, an automated annotation quality inspection algorithm, innovative geographic information hardware and software encryption plugins, and an efficient quality inspection toolchain. It integrates powerful computing power, large-capacity storage, and a complete data processing system (including Web services) into a single "annotation all-in-one intelligent body" device and provides a Web-based graphical user interface, which simplifies the data processing flow, reduces operational complexity, and improves the convenience of user operation and the flexibility of system deployment.

[0052] (2) High-efficiency automated processing: This system achieves a high degree of automation in the data processing process, including data preprocessing, quality assessment, automated labeling based on high-precision positioning and automated screening, which greatly reduces manual intervention, significantly improves the efficiency of large-scale road data labeling and reduces labor costs.

[0053] (3) Eliminating spatial data compliance risks: The system has a built-in geographic information encryption and decryption module to provide end-to-end encryption protection for raw data and final results. The system runs within an internal local area network, effectively preventing data leakage risks and meeting the security and confidentiality requirements for geospatial data transmission, storage, and use.

[0054] (4) Excellent computing performance: The integrated CPU and GPU provide powerful computing support for data processing algorithms, ensuring efficient data processing.

[0055] (5) Systematic quality control: Built-in multi-level verification and automated screening process can automatically judge the conformity of the results and eliminate inaccurate data, ensuring the high quality and consistency of the final delivered results.

[0056] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An automated labeling method for road information, characterized in that, Includes the following steps: S10, receive raw road data submitted by the user through the Web interface, and encrypt the raw road data; the raw road data is at least one of trajectory, point cloud and image data acquired by the vehicle-mounted acquisition device; S20, standardizes and verifies the quality of encrypted data; S30, the standardized and quality-verified data is located and fused with a preset static road base map, and road elements are automatically extracted based on the fusion result to generate structured labeled data. At the same time, based on the fusion result of the data and the static road base map, the camera pose parameters on the device that collected the original road data are optimized in reverse. The preset static road base map includes at least one key static road element among lane lines, curbs and traffic signs. S40, based on the optimized pose parameters, automatically screen the generated structured annotation data, and convert the screened data that meets the requirements into a standard vector format before encrypting and outputting it.

2. The automated road information labeling method as described in claim 1, characterized in that, In step S30, the static road base map is obtained through the following steps: Raw mapping data of the target road area is collected using at least one sensor from among the Global Navigation Satellite System receiver, inertial measurement unit, lidar, and camera mounted on a professional surveying vehicle. The original surveying data is preprocessed by time synchronization, coordinate system unification, and noise filtering. The preprocessed data is fused and solved to generate intermediate road data containing 3D point cloud and georeferenced image; From the road center data, identify and vectorize the contours and position information of lane lines, curbs, traffic sign poles, and traffic sign panels; The vectorized road element information is associated with absolute geographic coordinates and formatted according to preset data specifications to generate the static road base map.

3. The automated road information labeling method as described in claim 1, characterized in that, In step S30, the step of reverse optimization of the camera-by-camera pose parameters on the device that acquires the original road data specifically includes: Feature points are extracted from image frames in data that has undergone standardization and quality verification. The feature points are matched with the reference image features of the corresponding geographical locations in the static road base map to obtain multiple matching point pairs; Based on the matching point pairs, a reprojection error function is constructed regarding the camera pose parameters. The reprojection error function represents the difference between the position of the feature point projected onto the image plane under the current pose parameters and the position of its matching point in the reference image. The camera pose parameters are iteratively adjusted using a nonlinear optimization algorithm to minimize the reprojection error function, thereby obtaining the optimized camera-by-camera pose parameters.

4. The automated road information labeling method as described in claim 1, characterized in that, In step S40, the automated screening specifically includes: establishing a data index for the structured annotation data; automatically determining the conformity of the structured annotation data with preset rules based on the data index; and automatically removing structured annotation data that does not meet the requirements due to changes in the actual road scene or inaccurate projection calibration caused by the pose parameters before optimization.

5. The automated road information labeling method as described in claim 1, characterized in that, In step S40, the step of converting to a standard vector format specifically involves converting the road elements that meet the requirements after screening into a standard vector data format according to preset specifications.

6. The automated road information labeling method as described in claim 1, characterized in that, After step S40, the following is also included: S50 provides a quality control toolchain that includes manual verification and confirmation functions to process data that is questionable after automated screening.

7. The automated road information labeling method as described in claim 1, characterized in that, After step S40, the following is also included: S60 performs automated testing on encrypted output standard vector format data to ensure it meets preset quality and format requirements.

8. The automated road information labeling method as described in claim 1, characterized in that, In step S10, the received raw road data is encrypted, specifically through the built-in geographic information encryption / decryption module.

9. An intelligent labeling machine, characterized in that, include: The central processing unit (CPU) provides general computing capabilities and executes operating systems, web services, and task scheduling. A graphics processing unit, connected to the central processing unit, is used to provide parallel computing capabilities to accelerate point cloud processing, image analysis, localization and calibration optimization; The memory system, connected to the central processing unit and the graphics processing unit, includes memory and a storage device for persistently storing data, for storing software programs, static road base maps, received raw road data, intermediate processing data and final generated result data; The data input / output interface is connected to the central processing unit and is used to access the internal network to provide Web service interaction and data transmission. When the central processing unit executes program instructions stored in the memory system, it implements the automatic labeling method for road information as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for automatically labeling road information as described in any one of claims 1 to 8.

Images