An OpenX-based automatic driving simulation scene construction system
Patent Information
- Application Number
- CN202310702227.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-13
- Publication Date
- 2026-09-15
- Estimated Expiration
- 2043-06-13
AI Technical Summary
[0006]本发明为克服上述现有技术所述的数据质量低、计算资源消耗大的缺陷，提供一种基于OpenX的自动驾驶仿真场景构建系统
本发明公开了一种基于OpenX的自动驾驶仿真场景构建系统，通过数据解耦分类模块与数据描述模块的配合设计，在保证解耦数据保真性高的同时，确保标注的准确性，实现原始数据的标注与分类，解决数据质量低、获取成本高的问题，降低数据处理的复杂度；本发明还基于openX对解耦后数据重新描述，从而避免了兼容性问题，减少算力成本的开销。
Smart Images
Figure CN116629019B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of autonomous driving data processing technology, and more specifically, to an autonomous driving simulation scenario construction system based on OpenX. Background Technology
[0002] In recent years, with the continuous development of autonomous driving simulation testing technology, the preprocessing of multimodal data has become increasingly important for the generation of simulation test scenarios. Multimodal data needs to be decoupled and classified before scheduling and use. Data decoupling refers to removing the coupling relationship between different data elements or data sets, making them independent and non-interfering, and easier to modify, expand, and maintain. Typically, when different data elements or data sets are coupled, a change in any one data element may affect other related data, leading to unexpected events and impacting the generation of simulation test scenarios.
[0003] Furthermore, the lack of unified standards for data at each stage of scenario-based autonomous driving simulation testing—from R&D to testing, and from sensors to functional modules—hinders the practical application of simulation testing. In actual development, various data providers and simulation tool vendors use a wide variety of data formats, making the formulation of unified simulation format standards imperative.
[0004] In response, Waymo has proposed an autonomous driving data system based on Scenario protocol in existing technologies. This system segments scene data collected by multi-camera cameras temporally, with each 9-second time step containing 1 second of historical data, 1 second of instantaneous data, and 8 seconds of future data sampled at 10 Hz, corresponding to 10 historical samples, 1 instantaneous sample, and 80 future samples, totaling 91 samples. These samples are stored in Scenario protocol buffers format, providing ground truth values for the future trajectories of up to 8 objects within the scene. For point cloud data returned by radar, the first two returns of each frame are processed and stored in Scenario protocol buffers format. Subsequently, the temporal data from radar and cameras within the buffers are fused to generate a Scenario protocol model. The Scenario protocol contains the trajectories of a set of objects, as well as static and dynamic map features for each time step. Finally, the Scenario protocol model is converted into the tf.Exampleprotos dataset for storage. Motional proposed an autonomous driving data system based on the nuScenes devkit. First, it classifies and grades the raw data collected by radar and cameras, designs multiple class mapping models, and uses a fusion model method to decouple the preprocessed multimodal data. Second, it stores annotations and raw data (such as positioning, timestamps, and calibration data) in a relational database and allows for fast access.
[0005] However, in the aforementioned existing technologies, the autonomous driving data system based on Scenario protocol has a small processing capacity for raw point cloud data. When noise or missing data exists in the point cloud data, it leads to low quality of labeled data and difficulty in generating simulation test models. Furthermore, it consumes significant memory and computing resources by caching labeled point cloud data and frame data through Scenario protocol buffers. Additionally, the generated tf.Example protos dataset does not classify models within the scene, making subsequent data retrieval difficult. In contrast, the autonomous driving data system based on nuScenes devkit, due to its use of a fusion model method for data decoupling, requires complex data preprocessing and feature extraction. This results in an imbalance in the quantity and distribution of data from different modalities, leading to weaker learning capabilities for some modalities and impacting model performance and accuracy. Moreover, the fusion model method requires considering feature extraction and fusion from multiple modalities, resulting in a complex model structure and consuming substantial computing resources. Summary of the Invention
[0006] To overcome the shortcomings of low data quality and high computational resource consumption in the prior art, this invention provides an autonomous driving simulation scenario construction system based on OpenX.
[0007] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: An OpenX-based autonomous driving simulation scenario construction system includes: The data acquisition module is used to acquire multimodal data; the multimodal data includes raw image data and raw point cloud data based on the same road scene; The data decoupling and classification module is used to decouple and classify the multimodal data to obtain static road network data, dynamic behavior data, and road table data; The data description module is used to describe the static road network data, the dynamic behavior data, and the road table data based on OpenX, and to construct a data resource set as elements. The scenario building module is used to obtain the call requirements and filter the elements in the data resource set according to the call requirements to build the simulation scenario.
[0008] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention discloses an autonomous driving simulation scenario construction system based on OpenX. Through the coordinated design of the data decoupling classification module and the data description module, it ensures high fidelity of decoupled data while ensuring the accuracy of annotation, realizing the annotation and classification of original data, solving the problems of low data quality and high acquisition cost, and reducing the complexity of data processing. This invention also re-describes the decoupled data based on OpenX, thereby avoiding compatibility issues and reducing computing power costs. Attached Figure Description
[0009] Figure 1 This is a schematic diagram of the autonomous driving simulation scenario construction system described in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the hierarchical description process in the Open Drive unit described in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the dynamic behavior data description process in the Open Scenario unit described in Embodiment 1 of the present invention. Detailed Implementation
[0010] The terms "first," "second," etc., used 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 terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.
[0011] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.
[0012] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0013] Example 1 This embodiment proposes an OpenX-based autonomous driving simulation scenario construction system, as shown in Figure 1, including: The data acquisition module is used to acquire multimodal data; the multimodal data includes raw image data and raw point cloud data based on the same road scene; The data decoupling and classification module is used to decouple and classify the multimodal data to obtain static road network data, dynamic behavior data, and road table data; The data description module is used to describe the static road network data, the dynamic behavior data, and the road table data based on OpenX, and to construct a data resource set as elements. The scenario building module is used to obtain the call requirements and filter the elements in the data resource set according to the call requirements to build the simulation scenario.
[0014] This embodiment decouples multimodal data through a data decoupling classification module, effectively preventing overfitting risks and solving modal imbalance problems. In the data description module, the data is redescribed based on OpenX, which improves compatibility when calling data later, enhances the efficiency of data management, reduces computing power costs, and helps to quickly build simulation scenarios.
[0015] In some examples, the raw image data is acquired via a camera; In some examples, the raw point cloud data is acquired using lidar.
[0016] In a preferred embodiment, the decoupling and classification of the multimodal data includes: The original image data is preprocessed, and the YOLO target detection algorithm is used to identify targets based on the preprocessed original image data to obtain labeled image data with target annotation information; wherein the labeled image data includes static road network data, dynamic behavior data and road table data; The original point cloud data is subjected to target recognition using the RangeNet++ algorithm to obtain a reconstructed point cloud with target annotation information; the reconstructed point cloud includes static road network data, dynamic behavior data, and road table data.
[0017] This preferred embodiment uses the YOLO object detection algorithm to process the raw image data and the RangeNet++ algorithm to process the raw point cloud data, which helps to achieve rapid annotation and decoupling of the raw data and reduce the consumption of computing resources.
[0018] In one alternative embodiment, the preprocessing includes scaling and pixel normalization of the raw image data.
[0019] In some examples, bilinear interpolation is used for image scaling; In some examples, nearest neighbor interpolation is used for image scaling; In other examples, bicubic interpolation is used for image scaling.
[0020] In an optional embodiment, the target recognition based on the preprocessed original image data using the YOLO target detection algorithm includes: For any frame of the preprocessed original image data, a grid is divided using at least one resolution, and a neural network model is used to predict any target in the image based on the grid to obtain at least one candidate bounding box for the target; wherein the attribute values configured for the candidate bounding box include center coordinates, width, height, and class confidence. For candidate bounding boxes of the same target, the non-maximum suppression algorithm is used iteratively to predict the candidate bounding boxes until the final target bounding box is confirmed: the coverage confidence of the candidate bounding boxes is calculated, the candidate bounding box with the highest coverage confidence is selected as the current target bounding box, and the candidate bounding boxes are deleted according to the IOU (Intersection Over Union) of the area between the remaining candidate bounding boxes and the target bounding box. The candidate bounding boxes remaining after the deletion operation are taken as the final target bounding boxes. Among them, when the IOU is greater than a predetermined threshold, the corresponding candidate bounding box is deleted; when the IOU does not exceed the predetermined threshold, the deletion operation is not performed. Based on the category confidence of the target bounding box, the semantic label of the target is determined. Based on expert experience, the target is identified, classified and labeled according to the semantic label to form labeled image data. The classification results include static road network data, dynamic behavior data and road surface data.
[0021] It should be noted that the coverage confidence indicates the accuracy of describing the complete area of the target object that the corresponding candidate bounding box can cover, and the category confidence indicates the accuracy of the target object selected by the target bounding box belonging to a certain object category (such as curb or pedestrian).
[0022] In some examples, ;in, Represents the category probability vector, Indicates category confidence. This represents the average IOU of the remaining candidate bounding boxes after the deletion operation is performed.
[0023] In some examples, the semantic label of a target is determined by whether the category confidence exceeds a specified threshold.
[0024] In an optional embodiment, target recognition is performed on the raw point cloud data using the RangeNet++ algorithm, including: For any frame of point cloud in the original point cloud data, the third frame in three-dimensional space is represented by spherical coordinates. i Points Projecting the image onto a two-dimensional space represented by the distance image, and obtaining the corresponding index relationship, the transformation process is as follows:
[0025] In the formula, Represents the x and y coordinates of the distance from a pixel in the image; Represents the width and height of the distance image; This represents the vertical field of view of the acquisition device corresponding to the raw point cloud data, expressed as the upper field of view in the vertical direction. and lower field of view composition; This represents the distance value at each coordinate point; Based on the index relationship, convert any frame of point cloud into a size. The tensor is semantically segmented using a fully convolutional semantic segmentation network to obtain the semantic label and corresponding semantic image of each pixel in the distance image. i This is the default value; Based on expert experience, the corresponding pixels of the semantic tags are classified and labeled. The classification results include static road network data, dynamic behavior data, and road table data. Based on the index relationship, the semantic image is mapped to the corresponding point cloud to obtain the reconstructed point cloud.
[0026] In some examples, the YOLO algorithm is used for target recognition on the raw image data, with preset values. i The value is determined based on the number of attribute values configured for the target bounding box.
[0027] As a non-limiting example, the semantic labels of the target include lane lines, sidewalks, curbs, drivable areas, road barriers and / or road signs, and the corresponding classification results are static road network data; As a non-limiting example, the semantic label of the target includes information on the category, speed, size, orientation, and depth of dynamic targets such as pedestrians, vehicles, and / or animals, and the corresponding classification result is dynamic behavior data; As a non-limiting example, the semantic label of the target may also include road surface details, such as road potholes and road undulations, and the corresponding classification result is road surface data.
[0028] Those skilled in the art should understand that when a dynamic target is identified from multimodal data, its speed, orientation, depth, and other information are known.
[0029] In some examples, the pedestrian's speed is calculated using the time difference and positional changes between consecutive frames; the speed can be instantaneous or average. In some examples, the vehicle's depth is calculated by identifying pairs of points in the point cloud and using triangulation methods.
[0030] Furthermore, after the reconstructed point cloud is constructed, the nearest neighbor search algorithm is used to correct the reconstructed point cloud.
[0031] Based on the above technical means, the "shadow" problem that occurs during the reconstruction of point clouds can be solved.
[0032] In a preferred embodiment, the data description module includes an OpenDrive unit for describing the static road network data, an Open Scenario unit for describing the dynamic behavior data, and an Open CRG unit for describing the road table data; The description of the static road network data, the dynamic behavior data, and the road table data based on OpenX includes: The static road network data is transmitted to the Open Drive unit, and the static road network data is described hierarchically based on XML to obtain a first set of elements consisting of elements related to the static road network data. The dynamic behavior data is transmitted to the Open Scenario unit, and the dynamic behavior data is described based on the Storyboard framework to obtain a second set of elements consisting of elements related to the dynamic behavior data; The road surface data is transmitted to the Open CRG unit, and visualization elements about road surface details are constructed based on preset visualization tools, and combined to obtain a set of visualization elements; The data resource set is obtained by combining the first set of elements, the second set of elements, and the set of visualization elements.
[0033] Those skilled in the art should understand that, in the above preferred embodiments, static road information is reorganized and generated based on XML and a hierarchical data model through the Open Drive unit; the description of vehicle behavior control information is realized through the Open Scenario unit based on the Storyboard framework; and the detailed processing information of the physical road surface is restored through the Open CRG unit.
[0034] In some examples, Matlab is used as the visualization tool.
[0035] In an optional embodiment, the static road network data is described in layers, as shown in Figure 2, including: Create geometry elements to describe the road reference line, laneOffset and laneSection elements to describe lanes along the road reference line, and junction elements to describe the road connection method; The geometry element has attributes. S describes the position of the starting point of the road reference line in the s direction in the reference line coordinate system; X describes the position of the starting point of the road reference line in the x direction in the inertial coordinate system; Y describes the position of the starting point of the road reference line in the y direction in the inertial coordinate system; Hdg describes the orientation of the road reference line; and Length describes the length of the road reference line. The laneOffset element has attributes. S is used to describe the current lane's position from the s direction, and a, b, c, and d represent the parameters in the cubic polynomial used to describe the shape of the road reference line. The laneSection element has attributes. left and right are used to describe lane numbers and types, while width is used to describe lane width variations. (Road markings) are used to describe the solid / dashed type, details, color, and height of lane markings. (Speed) is used to describe the maximum speed of the lane; The junction element includes a connection sub-element, which has attributes. ; Used to describe the road numbers that need to be connected. (Incoming Road) describes the road ID before the connection point, connectingRoad describes the road ID after the connection point, and contactPoint describes the road connection point. Based on expert experience, values are assigned to the geometry element, laneSection element, and junction element according to the static road network data, and combined to obtain the first element set.
[0036] It should be noted that road reference lines can be modeled as a series of adjacent cubic polynomial curve segments. By adjusting the parameters of each curve segment (i.e., a, b, c, d), the shape and direction of the curve can be controlled. By connecting multiple curve segments, complex road networks can be formed to simulate real-world road systems.
[0037] In an optional embodiment, the description of the dynamic behavior data, referring to Figure 3, includes: Create at least one Storyboard element, which includes the scene content Story, action set Act, action group ManeuverGroup, and action Maneuver; For any Storyboard element: Based on expert experience, initialize at least one scene content Story according to the dynamic behavior data, including setting weather conditions, setting simulators, and defining vehicles; complete the initialization of the Storyboard element based on the scene content Story, including: defining at least one action set Act, including the conditions for starting the behavior, the conditions for ending the behavior, and at least one of the operation groups ManeuverGroup; the operation group ManeuverGroup defines the affected target vehicles and at least one of the operation Maneuver; the operation Maneuver consists of several events, used to describe a complete behavior; Combine all Storyboard elements to obtain the second set of elements.
[0038] In a preferred embodiment, the step of filtering elements in the data resource set according to the calling requirements and constructing a simulation scenario includes: Based on expert experience, a topological relationship between elements in a simulation scenario is constructed to determine the decision-making scenario element information in the corresponding simulation scenario; wherein, the topological relationship includes the association weight between elements; Based on the call requirements and the information of decision-making scenario elements, elements with association weights greater than a preset threshold are extracted, and an initial simulation scenario is constructed based on the topological relationship. The initial simulation scenario is tested and verified. Based on expert experience, the simulation scenario is updated or directly output to obtain the final simulation scenario.
[0039] It should be noted that the decision-making scenario element information refers to scenario element information that plays a decisive role in the driving strategy of autonomous driving, such as road, pedestrian, and vehicle flow information. Extracting corresponding elements from the data resource set based on scenario element information can improve the generalization ability of the simulation scenario. For example, elements containing vehicle position and motion status data can be extracted from the data resource set based on vehicle flow information. Compared with the existing technology that uses perception-based scenario elements as the core for simulation scenario construction, this preferred embodiment can ensure the effectiveness of scenario construction and improve data retrieval efficiency.
[0040] In some examples, decision-making scenario elements are mainly distributed in static road network data and dynamic behavior scenarios. Furthermore, perception-related scenario elements and execution-related scenario elements in the corresponding simulation scenarios were identified to adapt to the "perception-decision-execution" architecture of autonomous driving algorithms.
[0041] In some examples, the testing and verification process includes determining whether the simulation scenario is constructed correctly based on one or more of the evaluation indexes proposed by Zhang Chenyuan and Tao Fei in the published paper "Evaluation Index System for Digital Twin Models" (Computer Integrated Manufacturing Systems, 2021, 27(08):2171-2186.DOI:10.13196 / j.cims.2021.08.001.). Example 2 This embodiment proposes a method for constructing autonomous driving simulation scenarios based on OpenX, applying the system proposed in Embodiment 1, including: Multimodal data is input into the data acquisition module; wherein, the multimodal data includes raw image data and raw point cloud data based on the same road scene; A data decoupling and classification module is used to decouple and classify the multimodal data to obtain static road network data, dynamic behavior data, and road table data; Using the data description module, based on OpenX, the static road network data, dynamic behavior data, and road table data are described respectively, and used as elements to construct a data resource set; Based on the call requirements, the simulation scenario is constructed by filtering elements from the data resource set through the scenario construction module.
[0042] It is understood that the method in this embodiment corresponds to the apparatus in Embodiment 1 above, and the options in Embodiment 1 above are also applicable to this embodiment, so they will not be described again here.
[0043] Example 3 This embodiment proposes a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the method described in Embodiment 2.
[0044] By way of example, the storage medium includes, but is not limited to, USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks, and other media capable of storing program code.
[0045] By way of example, the instructions, programs, code sets, or instruction sets may be implemented in programming languages such as Java, Python, C++, R, or Golang.
[0046] By way of example, the processor includes, but is not limited to, smartphones, personal computers, servers, network devices, etc., for performing all or part of the steps of the method described in Example 2.
[0047] This embodiment also provides a computer program product including instructions for constructing autonomous driving simulation scenarios, which, when run on a computer, causes the computer to perform all or part of the steps in the method described in the aforementioned embodiment 2.
[0048] The same or similar labels correspond to the same or similar parts; The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent. Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. The various parts of this specification are described in a progressive manner, and similar or identical parts between different embodiments can be referred to mutually. Each embodiment focuses on the differences from other embodiments. For those skilled in the art, other variations or modifications can be made based on the above description. Functional modules or units can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.
Claims
1. An autonomous driving simulation scenario construction system based on OpenX, characterized in that, include: The data acquisition module is used to acquire multimodal data; the multimodal data includes raw image data and raw point cloud data based on the same road scene; The data decoupling and classification module is used to decouple and classify the multimodal data to obtain static road network data, dynamic behavior data, and road table data; The data description module is used to describe the static road network data, the dynamic behavior data, and the road table data based on OpenX, and to construct a data resource set as elements. The scenario building module is used to obtain the call requirements and filter the elements in the data resource set according to the call requirements to build the simulation scenario; Decoupling and classifying the multimodal data includes: The original image data is preprocessed, and the YOLO target detection algorithm is used to identify targets based on the preprocessed original image data to obtain labeled image data with target annotation information; wherein the labeled image data includes static road network data, dynamic behavior data and road table data; The RangeNet++ algorithm is used to identify targets in the original point cloud data to obtain a reconstructed point cloud with target annotation information; the reconstructed point cloud includes static road network data, dynamic behavior data, and road table data. The RangeNet++ algorithm is used to perform target recognition on the raw point cloud data, including: For any frame of point cloud in the original point cloud data, the third frame in three-dimensional space is represented by spherical coordinates. i Points Projecting the image onto a two-dimensional space represented by the distance image, and obtaining the corresponding index relationship, the transformation process is as follows: In the formula, Represents the x and y coordinates of the distance from a pixel in the image; Represents the width and height of the distance image; This represents the vertical field of view of the acquisition device corresponding to the raw point cloud data, expressed as the upper field of view in the vertical direction. and lower field of view composition; This represents the distance value at each coordinate point; Based on the index relationship, convert any frame of point cloud into a size. The tensor is semantically segmented using a fully convolutional semantic segmentation network to obtain the semantic label and corresponding semantic image for each pixel in the distance image; where... As a preset value, Based on expert experience, the corresponding pixels of the semantic tags are classified and labeled. The classification results include static road network data, dynamic behavior data, and road table data. Based on the index relationship, the semantic image is mapped to the corresponding point cloud to obtain the reconstructed point cloud.
2. The autonomous driving simulation scenario construction system based on OpenX according to claim 1, characterized in that, The preprocessing includes scaling and pixel normalization of the original image data.
3. The autonomous driving simulation scenario construction system based on OpenX according to claim 1, characterized in that, The target recognition based on the preprocessed original image data uses the YOLO target detection algorithm, including: For any frame of the preprocessed original image data, a grid is divided using at least one resolution, and a neural network model is used to predict any target in the image based on the grid to obtain at least one candidate bounding box for the target; wherein the attribute values configured for the candidate bounding box include center coordinates, width, height, and class confidence. For candidate bounding boxes of the same target, the non-maximum suppression algorithm is used iteratively to predict the candidate bounding boxes until the final target bounding box is confirmed: the coverage confidence of the candidate bounding boxes is calculated, the candidate bounding box with the highest coverage confidence is selected as the current target bounding box, and the candidate bounding boxes are deleted according to the IOU between the area of the remaining candidate bounding boxes and the target bounding box. The candidate bounding boxes remaining after the deletion operation are taken as the final target bounding boxes. Among them, when the IOU is greater than a predetermined threshold, the corresponding candidate bounding box is deleted; when the IOU does not exceed the predetermined threshold, the deletion operation is not performed. Based on the category confidence of the target bounding box, the semantic label of the target is determined. Based on expert experience, the target is identified, classified and labeled according to the semantic label to form labeled image data. The classification results include static road network data, dynamic behavior data and road surface data.
4. The autonomous driving simulation scenario construction system based on OpenX according to claim 1, characterized in that, After the reconstructed point cloud is constructed, the nearest neighbor search algorithm is used to correct the reconstructed point cloud.
5. The autonomous driving simulation scenario construction system based on OpenX according to claim 1, characterized in that, The data description module includes an Open Drive unit for describing the static road network data, an Open Scenario unit for describing the dynamic behavior data, and an Open CRG unit for describing the road table data; The description of the static road network data, the dynamic behavior data, and the road table data based on OpenX includes: The static road network data is transmitted to the Open Drive unit, and the static road network data is described hierarchically based on XML to obtain a first set of elements consisting of elements related to the static road network data. The dynamic behavior data is transmitted to the Open Scenario unit, and the dynamic behavior data is described based on the Storyboard framework to obtain a second set of elements consisting of elements related to the dynamic behavior data; The road surface data is transmitted to the Open CRG unit, and visualization elements about road surface details are constructed based on preset visualization tools, and combined to obtain a set of visualization elements; The data resource set is obtained by combining the first set of elements, the second set of elements, and the set of visualization elements.
6. The autonomous driving simulation scenario construction system based on OpenX according to claim 5, characterized in that, The static road network data is described in layers, including: Create a geometry element to describe the road reference line, a laneOffset element and a laneSection element to describe the lanes along the road reference line, and a junction element to describe the road connection method; The geometry element has attributes. S describes the position of the starting point of the road reference line in the s direction in the reference line coordinate system; X describes the position of the starting point of the road reference line in the x direction in the inertial coordinate system; Y describes the position of the starting point of the road reference line in the y direction in the inertial coordinate system; Hdg describes the orientation of the road reference line; and Length describes the length of the road reference line. The laneOffset element has attributes. S is used to describe the current lane's position from the s direction, and a, b, c, and d represent the parameters in the cubic polynomial used to describe the shape of the road reference line. The laneSection element has attributes. left and right are used to describe lane numbers and types, while width is used to describe lane width variations. Used to describe the solid / dark type, details, color, and height of lane lines. Used to describe the maximum speed of a lane; The junction element includes a connection sub-element, which has attributes. ; Used to describe the road numbers that need to be connected. The connectingRoad is used to describe the road ID before the connection point, the connectingRoad is used to describe the road ID after the connection point, and the contactPoint is used to describe the road connection point. Based on expert experience, values are assigned to the geometry element, laneSection element, and junction element according to the static road network data, and combined to obtain the first element set.
7. The autonomous driving simulation scenario construction system based on OpenX according to claim 5, characterized in that, The description of the dynamic behavior data includes: Create at least one Storyboard element, which includes the scene content Story, the action set Act, the action group ManeuverGroup, and the action Maneuver; For any Storyboard element: Based on expert experience, initialize at least one scene content Story according to the dynamic behavior data, including setting weather conditions, setting simulators, and defining vehicles; complete the initialization of the Storyboard element based on the scene content Story, including: defining at least one action set Act, including the conditions for starting the behavior, the conditions for ending the behavior, and at least one of the operation groups ManeuverGroup; the operation group ManeuverGroup defines the affected target vehicles and at least one of the operation Maneuver; the operation Maneuver consists of several events, used to describe a complete behavior; Combine all Storyboard elements to obtain the second set of elements.
8. A system for constructing autonomous driving simulation scenarios based on OpenX according to any one of claims 1-7, characterized in that, The step of filtering elements from the data resource set according to the calling requirements and constructing a simulation scenario includes: Based on expert experience, a topological relationship between elements in a simulation scenario is constructed to determine the decision-making scenario element information in the corresponding simulation scenario; wherein, the topological relationship includes the association weight between elements; Based on the call requirements and the information of decision-making scenario elements, elements with association weights greater than a preset threshold are extracted, and an initial simulation scenario is constructed based on the topological relationship. The initial simulation scenario is tested and verified. Based on expert experience, the simulation scenario is updated or directly output to obtain the final simulation scenario.