Vehicle test scene determination method and device, electronic equipment and storage medium

Abstract

The invention discloses a vehicle test scene determination method and device, electronic equipment and a storage medium. The method comprises the steps that multiple frames of test scene images are extracted from a real scene video of a vehicle, and the real scene video is used for representing a dynamic scene and a static scene where the vehicle is located in the driving process of the vehicle in an actual road environment; extracting scene node information of the vehicle from the multiple frames of test scene images; inputting the scene node information into a test scene generation model of the vehicle for analysis to obtain an original test scene of the vehicle; and performing a risk test on the original test scene to obtain a risk test result, and determining the original test scene as a target test scene based on the risk test result. The technical problem of low test scene generation efficiency is solved.

Description

Technical Field

[0001] This invention relates to the field of vehicle technology, and more specifically, to a method, apparatus, electronic device, and storage medium for determining a vehicle's test scenario. Background Technology

[0002] As autonomous driving technology continues to advance, the requirements for testing scenarios are also increasing. These scenarios need to cover various complex and extreme situations that vehicles may encounter during operation, in order to ensure the reliability and safety of autonomous driving technology in practical applications.

[0003] Currently, traditional methods for modeling test scenarios involve either manually designing scenarios or constructing them based on real road test data, but these methods suffer from the technical problem of low efficiency in generating test scenarios.

[0004] There is currently no effective solution to the aforementioned technical problem of low efficiency in generating test scenarios. Summary of the Invention

[0005] This invention provides a method, apparatus, electronic device, and storage medium for determining test scenarios for vehicles, thereby at least solving the technical problem of low efficiency in test scenario generation.

[0006] According to one aspect of the invention, a method for determining a vehicle's test scenario is provided. The method may include: extracting multiple frames of test scenario images from a real-world video of the vehicle, wherein the real-world video represents the dynamic and static scenarios in which the vehicle operates during actual road driving; extracting scene node information of the vehicle from the multiple frames of test scenario images, wherein the scene node information represents the vehicle's dynamic entity behavior, static environment layout, and external environment state in the test scenario; inputting the scene node information into a vehicle test scenario generation model for analysis to obtain the vehicle's original test scenario, wherein the test scenario generation model is trained using scene node information samples, the scene node information samples represent historical information of the scene node information, and the original test scenario represents a traffic environment state with a target testing purpose; performing a risk test on the original test scenario to obtain a risk test result; and determining the original test scenario as a target test scenario based on the risk test result, wherein the risk test result represents a risk event occurring during the vehicle's operation in the original test scenario, and the similarity between the target test scenario and the vehicle's actual application scenario is higher than the similarity between the original test scenario and the actual application scenario.

[0007] Optionally, scene node information of the vehicle is extracted from multiple test scene images, including: performing object detection on the multiple test scene images to obtain participant node information and static environment node information of the scene node information, wherein the participant node information is used to characterize moving objects appearing during the vehicle's driving process, and the static environment node information is used to characterize static objects appearing during the vehicle's driving process; and / or, obtaining the illumination intensity in the multiple test scene images; and analyzing the illumination intensity to obtain external environment node information of the scene node information, wherein the external environment node information is used to characterize the external environment in which the vehicle is located during the vehicle's driving process.

[0008] Optionally, the scene node information is input into the vehicle's test scene generation model for analysis to obtain the vehicle's original test scene, including: determining the vehicle's target scene graph based on the scene node information, wherein the target scene graph is used to represent the relationship between various entities of the vehicle in the test scene in the time and space dimensions respectively; and inputting the target scene graph into the test scene generation model for analysis to obtain the original test scene.

[0009] Optionally, determining the target scene map of the vehicle based on scene node information includes: constructing a first scene map based on participant node information in the scene node information, wherein the first scene map is used to represent the relationship between multiple moving objects appearing during the vehicle's operation; updating the first scene map based on static environment node information in the scene node information to obtain a second scene map, wherein the second scene map is used to represent the relationship between multiple moving objects and multiple static objects appearing during the vehicle's operation; updating the second scene map based on external environment node information in the scene node information to obtain an initial scene map, wherein the initial scene map represents the relationship between multiple moving objects, multiple static objects, and different external environments in which the vehicle is located during its operation; and updating the initial scene map based on vehicle disturbance information to obtain a target scene map, wherein the disturbance information is used to represent sudden traffic behavior of moving objects, visual interference or information tampering of static objects, and changing external environments.

[0010] Optionally, a first scene graph is constructed based on participant node information in the scene node information, including: determining at least a first participant entity and a second participant entity from the participant node information; determining the first current position and first velocity of the first participant entity, and the second current position and second velocity of the second participant entity, based on a vehicle scene library; determining the target relative displacement based on the first and second current positions; and determining the target relative velocity based on the first and second velocities; obtaining the first equivalent radius corresponding to the first participant entity and the second equivalent radius corresponding to the second participant entity, and determining the minimum safe distance between the first participant entity and the second participant entity based on the first and second equivalent radii; in response to the target relative displacement being greater than a target safe distance threshold, determining the collision duration based on the target relative displacement, target relative velocity, and minimum safe distance, wherein the collision duration is used to characterize the time required for a collision to occur between the first participant entity and the second participant entity; in response to the collision duration being less than a duration threshold, determining the first target edge information between the first participant entity and the second participant entity; and constructing the first scene graph based on the first target edge information and the participant node information.

[0011] Optionally, the first scene graph is updated based on the static environment node information in the scene node information to obtain a second scene graph, including: determining at least one target static object from the static environment node information and determining the center position of the target static object; determining the target participant entity corresponding to the target static object from the first scene graph and determining the current position of the target participant entity; obtaining the maximum perceptual distance between the target static object and the target participant entity, wherein the maximum perceptual distance is used to characterize the degree of influence of the target static object on the target participant entity; determining the target distance based on the center position of the target static object and the current position of the target participant entity; determining the second target edge information between the target static object and the target participant entity in response to the target distance being less than the maximum perceptual distance; and determining the second scene graph based on the second target edge information and the static environment node information.

[0012] Optionally, the target scene graph is input into the test scene generation model for analysis to obtain the original test scene, including: evaluating the target scene graph using the scene evaluation index of the test scene generation model to obtain the evaluation result, wherein the evaluation result is used to characterize that the target scene graph meets the scene evaluation index; and based on the evaluation result, calling the test scene generation model to generate the original test scene.

[0013] Optionally, risk testing is performed on the original test scenario to obtain risk test results, including: acquiring the vehicle's false recognition rate, path deviation, and emergency braking frequency in the original test scenario, wherein the false recognition rate is used to characterize the number of times the vehicle misidentifies a moving object in the original test scenario, the path deviation is used to characterize the degree of path deviation of the vehicle in the original test scenario, and the emergency braking frequency is used to characterize the number of times the vehicle brakes suddenly in the original test scenario; in response to the false recognition rate being greater than a false recognition rate threshold, the path deviation being greater than a path deviation threshold, and the emergency braking frequency being greater than an emergency braking frequency threshold, the risk test results are determined.

[0014] According to one aspect of the present invention, a test scenario determination device for a vehicle is provided. The device may include: an acquisition unit for extracting multiple frames of test scene images from real-world video of the vehicle, wherein the real-world video represents the dynamic and static scenes in which the vehicle is located during its driving in a real road environment; an extraction unit for extracting scene node information of the vehicle from the multiple frames of test scene images, wherein the scene node information represents the dynamic entity behavior, static environment layout, and external environment state of the vehicle in the test scene; an analysis unit for inputting the scene node information into a test scene generation model of the vehicle for analysis to obtain the original test scene of the vehicle, wherein the test scene generation model is trained using scene node information samples, the scene node information samples represent the historical information of scene node information, and the original test scene represents the traffic environment state with the target testing purpose; and a determination unit for performing risk testing on the original test scene to obtain risk test results, and determining the original test scene as the target test scene based on the risk test results, wherein the risk test results represent that there are risk events in the original test scene during the vehicle's driving, and the similarity between the target test scene and the vehicle's actual application scene is higher than the similarity between the original test scene and the actual application scene.

[0015] According to another aspect of the present invention, an electronic device is also provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods of various embodiments of the present invention during runtime.

[0016] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is executed, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of the present invention.

[0017] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of the present invention.

[0018] According to another aspect of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods of various embodiments of the present invention.

[0019] According to another aspect of the present invention, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of the present invention.

[0020] According to another aspect of the present invention, a vehicle is also provided that implements the methods of the various embodiments of the present invention when executed.

[0021] In this embodiment of the invention, multiple frames of test scene images can be extracted from real-world vehicle scene videos. Then, scene node information of the vehicle is extracted from these multiple frames. This scene node information characterizes the vehicle's dynamic entity behavior, static environment layout, and external environment state within the test scene. The obtained scene node information is input into the vehicle's test scene generation model for analysis to obtain the original test scene. A risk test is then performed on the original test scene to obtain risk test results. These risk test results characterize the presence of risk events during vehicle operation within the original test scene. Based on the risk test results, the original test scene can be identified as the target test scene. Because the scene node information can be analyzed using the test scene generation model after extraction from the multiple frames of test scene images, the obtained original test scene can encompass various scene types of the vehicle in real-world scenarios. Further analysis and testing of the original test scene yields risk test results characterizing the presence of risk events during vehicle operation within the original test scene, thus achieving the goal of identifying the target test scene. This solves the technical problem of low test scene generation efficiency and improves the technical effect of increasing test scene generation efficiency. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0023] Figure 1 This is a flowchart of a method for determining a vehicle test scenario according to an embodiment of the present invention;

[0024] Figure 2 This is a flowchart of an automatic generation method for autonomous driving adversarial test scenarios based on D-GAN according to an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of an adversarial scene generation framework according to an embodiment of the present invention;

[0026] Figure 4 This is a schematic diagram of a vehicle test scenario determination device according to an embodiment of the present invention. Detailed Implementation

[0027] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] According to an embodiment of the present invention, a method for determining a test scenario for a vehicle is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] The method for determining the test scenario for vehicles according to an embodiment of the present invention will be described below.

[0031] Figure 1 This is a flowchart of a method for determining a vehicle test scenario according to an embodiment of the present invention, such as... Figure 1 As shown, the method may include the following steps:

[0032] Step S101: Extract multiple frames of test scene images from the real scene video of the vehicle.

[0033] In the technical solution provided by step S101 of the present invention, a real scene video of the vehicle can be obtained, and multiple test scene images can be extracted from the real scene video. The real scene video is used to characterize the dynamic and static scenes in which the vehicle is located during its driving in the actual road environment.

[0034] Optionally, the real-world scene video may include, but is not limited to, dynamic and static scenes such as vehicles, pedestrians, non-motorized vehicles, road traffic signs, traffic lights, lane lines, weather conditions, and changes in lighting. The test scene image may be referred to as a scene image. The real-world scene video may be referred to as an autonomous driving test scene video.

[0035] For example, real-world video footage of a vehicle driving in a real road environment is collected, and test scene images are extracted frame by frame from the real-world video footage to perform the following steps.

[0036] It should be noted that this is only a preferred implementation method for acquiring multiple frames of test scene images, and the process and method for acquiring multiple frames of test scene images are not specifically limited.

[0037] Step S102: Extract the scene node information of the vehicle from the multi-frame test scene images.

[0038] In the technical solution provided by step S102 of the present invention, after obtaining multiple test scene images, the scene node information of the vehicle can be extracted from the multiple test scene images using a deep learning model (convolutional neural network, etc.). The scene node information is used to characterize the dynamic entity behavior, static environment layout and external environment state of the vehicle in the test scene.

[0039] Optionally, a deep learning model is used to analyze and understand the autonomous driving test scenario videos of vehicles. This model can quickly and accurately identify and classify various traffic elements in the videos, providing a solid foundation for constructing dynamic and complex test scenario maps, thereby promoting the automatic generation of high-fidelity autonomous driving test scenarios.

[0040] Optionally, scene node information may include participant node information, static environment node information, and external environment node information. Participant node information can be referred to as dynamic traffic participant node information, and dynamic traffic participants can be referred to as moving objects, such as vehicles, pedestrians, and bicycles. Static environment node information can be referred to as static environment element node information, and the objects referred to by static environment elements can be referred to as static objects, such as lane lines, traffic lights, and signs. External environment node information can be referred to as external environment factor node information.

[0041] For example, test scene images are extracted frame by frame from real-world video, and a deep learning model is used to identify and extract target objects in the test scene images. These objects are then deconstructed into editable semantic elements, including dynamic traffic participants, static environmental elements, and external environmental factors, forming semantically structured data that can be used for training. This achieves the goal of obtaining participant node information, static environment node information, and external environment node information.

[0042] It is understood that this is only a preferred implementation of obtaining scene node information of the vehicle, and the process and method of obtaining scene node information of the vehicle are not specifically limited. As long as the process and method of extracting scene node information of the vehicle from multiple frames of test scene images are within the protection scope of this invention, they will not be described in detail here.

[0043] Step S103: Input the scene node information into the vehicle's test scene generation model for analysis to obtain the vehicle's original test scene.

[0044] In the technical solution provided by step S103 of the present invention, the obtained scene node information is input into the vehicle's test scene generation model for analysis, so as to obtain the original test scene of the vehicle. The test scene generation model is trained using scene node information samples, which represent the historical information of the scene node information. The original test scene represents the traffic environment state with a target testing objective. The target testing objective can be referred to as a specific testing objective.

[0045] Optionally, the test scene generation model can be constructed using a Deep Generative Adversarial Network (D-GAN), which can be referred to as the D-GAN adversarial generation framework. This deep generative adversarial network includes a generator and a discriminator. The generator includes local perturbators, global perturbators, and environmental perturbators. The generator is used to generate new, real-like samples from random noise or from the latent feature space. The discriminator is used to determine whether the samples generated by the generator are real, i.e., whether they can be confused with real data.

[0046] For example, participant node information, static environment node information, and external environment node information are input into a deep generative adversarial network for analysis to obtain the original test scenario of the vehicle. This original test scenario is used to characterize the traffic environment state of the vehicle with a specific test purpose.

[0047] It should be noted that this is only a preferred implementation method for obtaining the original test scene of the vehicle. The process and method of obtaining the original test scene of the vehicle are not specifically limited. As long as the scene node information is input into the vehicle test scene generation model for analysis to obtain the original test scene of the vehicle, the process and method are within the protection scope of this invention, and will not be listed here.

[0048] Step S104: Perform risk testing on the original test scenario to obtain the risk test results, and determine the original test scenario as the target test scenario based on the risk test results.

[0049] In the technical solution provided by step S104 of the present invention, the obtained original test scenario can be subjected to risk testing to obtain risk test results, and the original test scenario can be determined as the target test scenario based on the risk test results at this time.

[0050] Optionally, the risk test results are used to characterize the presence of risk events during vehicle operation in the original test scenario. The similarity between the target test scenario and the actual application scenario of the vehicle is higher than the similarity between the original test scenario and the actual application scenario.

[0051] Optionally, the target test scenario can be called an effective adversarial scenario. An effective adversarial scenario is used to characterize test scenarios that can simulate low-probability but high-risk events that autonomous vehicles may encounter in daily driving. The original test scenario is used to characterize test scenarios that simulate extreme situations that are almost impossible to occur or have minimal impact on driving safety.

[0052] For example, by conducting risk testing on the original test scenario, we can obtain risk test results that characterize the presence of risk events during vehicle operation under the original test scenario. This indicates that the original test scenario contains situations where risk events exist, and the original test scenario can be identified as an effective adversarial scenario.

[0053] It should be noted that this is only a preferred implementation method for determining the target test scenario, and the process and method for determining the target test scenario are not specifically limited. As long as it is based on the risk test results, the process and method of determining the original test scenario as the target test scenario are within the protection scope of this invention, and will not be listed here.

[0054] Steps S101 to S104 of this invention extract multiple frames of test scene images from real-world vehicle video. Then, scene node information of the vehicle is extracted from these multiple frames. This scene node information characterizes the vehicle's dynamic entity behavior, static environment layout, and external environment state within the test scene. The obtained scene node information is input into the vehicle's test scene generation model for analysis, yielding the original test scene. A risk test is then performed on the original test scene to obtain risk test results. These results characterize the presence of risk events during vehicle operation within the original test scene. Based on these risk test results, the original test scene can be identified as the target test scene. Because the scene node information can be analyzed using the test scene generation model after extraction from multiple frames of test scene images, the obtained original test scene can encompass various scene types encountered by the vehicle in real-world scenarios. Further analysis and testing of the original test scene yields risk test results characterizing the presence of risk events during vehicle operation within the original test scene, thus achieving the goal of identifying the target test scene. This solves the technical problem of low test scene generation efficiency and improves the technical effect of increasing test scene generation efficiency.

[0055] The method described in this embodiment will be further described below.

[0056] As an optional embodiment, the vehicle's scene node information is extracted from multiple test scene images, including: performing object detection on the multiple test scene images to obtain participant node information and static environment node information of the scene node information, wherein the participant node information is used to characterize moving objects appearing during the vehicle's driving process, and the static environment node information is used to characterize static objects appearing during the vehicle's driving process; and / or, obtaining the illumination intensity in the multiple test scene images; and analyzing the illumination intensity to obtain the external environment node information of the scene node information, wherein the external environment node information is used to characterize the external environment in which the vehicle is located during the vehicle's driving process.

[0057] In this embodiment, object detection is performed on multiple frames of test scene images to obtain participant node information and static environment node information of scene node information. Objects can be referred to as entities, such as vehicle information, pedestrian information, lane line information, traffic light information, and sign information.

[0058] For example, a deep learning model can be used to perform object detection on each frame of an image, identifying all entities of interest and their corresponding information. These entities include, but are not limited to, dynamic traffic participants such as vehicles, pedestrians, and bicycles. At the same time, static environmental elements such as lane lines, traffic lights, and signs should also be identified.

[0059] Optionally, the illumination intensity in multiple frames of test scene images can be obtained using a brightness histogram, and this illumination intensity can be analyzed to obtain the external environment node information of the scene node information. The external environment node information can include weather conditions, illumination intensity, and visibility, and these external environmental factors are typically global conditions that affect all entities in the entire scene.

[0060] For example, color and texture analysis in videos can determine weather conditions (sunny, cloudy, rainy, snowy, foggy) in a test scene, brightness histograms can be used to determine light intensity, and the field of view can be estimated. For special environmental factors (such as construction areas), Optical Character Recognition (OCR) technology can be used to read information from on-site signs to obtain information about external environmental nodes.

[0061] As an optional implementation method, the scene node information is input into the vehicle's test scene generation model for analysis to obtain the vehicle's original test scene, including: determining the vehicle's target scene graph based on the scene node information, wherein the target scene graph is used to represent the relationship between various entities of the vehicle in the test scene in the time and space dimensions respectively; and inputting the target scene graph into the test scene generation model for analysis to obtain the original test scene.

[0062] In this embodiment, a target scene map of the vehicle is constructed based on scene node information. This target scene map is then input into a test scene generation model for analysis to obtain the original test scene. The target scene map can be referred to as an adversarial scene map, and is generated using G... adv This is represented. Scene node information can be referred to as raw data.

[0063] For example, based on participant node information, static environment node information, and external environment node information, an adversarial scenario graph of the vehicle is constructed. The adversarial scenario graph is a graph structure with semantic information, which facilitates the learning and reasoning of the subsequent generative model. The adversarial scenario graph at this time is input into the D-GAN adversarial generation framework for analysis to obtain the original test scenario.

[0064] As an optional embodiment, determining the target scene map of the vehicle based on scene node information includes: constructing a first scene map based on participant node information in the scene node information, wherein the first scene map is used to represent the relationship between multiple moving objects appearing during the vehicle's movement; updating the first scene map based on static environment node information in the scene node information to obtain a second scene map, wherein the second scene map is used to represent the relationship between multiple moving objects and multiple static objects appearing during the vehicle's movement; updating the second scene map based on external environment node information in the scene node information to obtain an initial scene map, wherein the initial scene map represents the relationship between multiple moving objects, multiple static objects, and different external environments in which the vehicle is located during its movement; and updating the initial scene map based on vehicle disturbance information to obtain a target scene map, wherein the disturbance information is used to represent sudden traffic behavior of moving objects, visual interference or information tampering of static objects, and changing external environments.

[0065] In this embodiment, a first scene graph is constructed based on the participant node information in the scene node information. Then, the first scene graph is updated based on the static environment node information in the scene node information to obtain a second scene graph. The second scene graph is then updated based on the external environment node information in the scene node information to obtain an initial scene graph. Finally, the initial scene graph is updated based on the vehicle disturbance information to achieve the purpose of obtaining the target scene graph.

[0066] Optionally, the initial scene graph can be called the scene graph structure, through G base Representation is performed based on the generated scene graph structure G. base Combined with disturbance information, generate adversarial scenario diagrams that conform to traffic rules and are logically consistent.

[0067] Optionally, the disturbance information includes: disturbance information of dynamic traffic participants, disturbance information of static environmental elements, and disturbance information of weather environment, wherein the disturbance information of weather environment can be referred to as disturbance information of global environmental factors.

[0068] Optionally, the process of acquiring disturbance information of dynamic traffic participants is as follows: Under the premise of maintaining traffic rules and physical constraints, adversarial disturbances (position, speed, acceleration, etc.) of dynamic traffic participants are generated to obtain disturbance information. This information can then be used to test the perception and decision-making capabilities of the autonomous driving system by simulating abnormal driving behaviors (such as sudden lane changes and emergency braking). Specifically, local disturbances in the disturbance information of dynamic traffic participants include: minor adjustments to the trajectories of key participants, such as sudden speed changes (randomly increasing / decreasing speed within a reasonable range) and path deviations (randomly adjusting the horizontal coordinate within lane boundaries to ensure no boundary crossings). Global disturbances in the disturbance information of dynamic traffic participants include: introducing logically contradictory interactive behaviors, such as illegal lane changes (forced lane changes when lane changes are prohibited by traffic lights) and intersection conflicts (generating two vehicles simultaneously entering an intersection without priority right-of-way).

[0069] Optionally, the process of obtaining disturbance information of static environmental elements is as follows: adversarial disturbances are generated for static environmental elements (such as lane lines, traffic signs, and traffic lights) to obtain disturbance information. This information can then be used to test the perception robustness of the autonomous driving system by simulating visual interference or information tampering scenarios. Examples include wear simulation (reducing lane line visibility by changing the color from white to gray), simulating low-light scenarios such as rain or nighttime; visual interference (overlaying textured stickers, such as reflective stickers, on signs to simulate strong light interference); and traffic light status disturbances (abnormal signal flashing or malfunction).

[0070] Optionally, the process of obtaining weather disturbance information involves generating adversarial disturbances to global environmental factors (such as weather, illumination, and visibility) to obtain weather disturbance information. This allows for the testing of the autonomous driving system's adaptability under adverse conditions by simulating extreme or complex scenarios. Examples include weather disturbances (rain / snow simulation: increasing raindrop / snowflake density, adjusting fog concentration to blur distant objects, such as lane lines and signs, reducing visibility), illumination disturbances (strong light reflection: simulating direct sunlight or headlight interference, such as increasing the proportion of highlight areas in the brightness histogram, and low nighttime illumination: reducing overall light intensity), and visibility disturbances (dynamic occlusion: introducing random obstacles (such as billboards and trees) to occlude key elements (such as traffic lights and lane lines).

[0071] For example, a model is created for traffic participants (including vehicles, motor vehicles, non-motor vehicles, pedestrians, animals, etc.) to construct a graph structure in the spatiotemporal dimension to capture the behavioral patterns and interaction relationships of dynamic traffic participants. This graph structure is the first scene graph. Based on this, a node is created for each static environmental element, and these nodes are connected to each other and to other traffic participant nodes through edges to obtain the first scene graph. Then, nodes are created for each entity in the external environmental factors, and these nodes are connected to each other, to other traffic participant nodes, and to each node of the static environmental factors through edges to obtain the initial scene graph. Further, perturbation information is added to the initial scene graph to obtain the target scene graph.

[0072] As an optional embodiment, a first scene graph is constructed based on participant node information in the scene node information, including: determining at least a first participant entity and a second participant entity from the participant node information; determining the first current position and first speed of the first participant entity, and the second current position and second speed of the second participant entity, based on a vehicle scene library; determining the target relative displacement based on the first and second current positions; and determining the target relative speed based on the first and second speeds; obtaining the first equivalent radius corresponding to the first participant entity and the second equivalent radius corresponding to the second participant entity, and determining the minimum safe distance between the first participant entity and the second participant entity based on the first and second equivalent radii; determining a collision duration based on the target relative displacement, target relative speed, and minimum safe distance in response to the target relative displacement being greater than a target safe distance threshold, wherein the collision duration is used to characterize the time required for a collision to occur between the first participant entity and the second participant entity; determining the first target edge information between the first participant entity and the second participant entity in response to the collision duration being less than a duration threshold; and constructing the first scene graph based on the first target edge information and the participant node information.

[0073] In this embodiment, at least a first participant entity and a second participant entity can be determined from the participant node information. Then, based on the vehicle scene library, the first current position and first velocity of the first participant entity, and the second current position and second velocity of the second participant entity are determined. Based on the obtained first and second current positions, the target relative displacement is determined, and based on the obtained first and second velocities, the target relative velocity is determined. Then, the first equivalent radius corresponding to the first participant entity and the second equivalent radius corresponding to the second participant entity can be obtained respectively. Based on the first and second equivalent radii, the minimum safe distance between the first and second participant entities is determined. The target relative displacement at this time is compared with a target safe distance threshold. If the target relative displacement is greater than the target safe distance threshold, the collision duration can be determined based on the target relative displacement, target relative velocity, and minimum safe distance. The collision duration is compared with a duration threshold. If the collision duration is less than the duration threshold, the first target edge information between the first and second participant entities is determined. Multiple first target edge information between different entities can be determined according to the above steps. Finally, based on the first target edge information and participant node information, a first scene graph can be constructed. Here, the collision duration can be referred to as the collision time.

[0074] Optionally, the first participant entity and the second participant entity can each be two traffic participants, where the first participant entity can be represented by A, and the second participant entity can be represented by B. The node features of the first participant entity are as follows: lateral and longitudinal positions, lateral and longitudinal velocities, lateral and longitudinal accelerations, and yaw angles, where the lateral and longitudinal positions are represented by x... A and y A The lateral and longitudinal velocities are represented by v. xA and v yA The node features of the second participant entity are of the same type as those of the first participant entity. The horizontal and vertical positions of the second participant entity are represented by x... B and y B The lateral and longitudinal velocities are represented by v. xB and v yB To express.

[0075] Optionally, the first current position can be accessed via This can be represented. The first velocity can be obtained through... This can be represented. The second current position can be indicated through... This can be represented. The second speed can be achieved through... The target's relative displacement can be represented. This can be expressed as relative displacement. The target's relative velocity can be simply referred to as relative velocity, and can be expressed as... The first equivalent radius can be represented by r. A The second equivalent radius can be represented by r. B This is represented. The minimum safe distance can be expressed as d. safe The collision time can be represented as the sum of the first and second equivalent radii. Collision time can be called Time-To-Collision (TTC). The target safety distance threshold can be called the safety distance threshold, and the collision time threshold can be a preset threshold, represented by T_threshold.

[0076] Optionally, the participant node information is obtained by: using the trajectory data of traffic participants obtained from the scene database, and constructing a spatiotemporal graph of the scene. Nodes are used to represent entity elements in the scene, including the main vehicle and key motor vehicles, non-motor vehicles, pedestrians, animals, etc., that have important interactions with the main vehicle. Node features represent the entity's horizontal and vertical positions x and y, and its horizontal and vertical velocities v. x v y acceleration a x a y y-angle aw The system includes 5 element flag bits, where 10000 represents the main vehicle, 0X000 represents the Xth key motor vehicle entity, 00X00 represents the Xth key non-motor vehicle entity, 000X0 represents the Xth key pedestrian entity, and 0000X represents the Xth key animal entity, totaling 12 node features. Within the same frame, every two nodes are connected by an edge when certain conditions are met; the edge feature is the time interval. In different frames, only nodes with the same element flag bits are connected by an edge; the edge feature is the time interval.

[0077] Optionally, if the relative displacement of the target is greater than the target safe distance threshold, the collision duration is determined based on the target relative displacement, target relative velocity, and minimum safe distance. For example, TTC is only calculated when the relative motion trend of the two traffic participants is "approaching" and the current distance is greater than the safe distance threshold. The TTC is calculated as follows: nodes A and B are two traffic participants, and their positions are respectively... The speeds are respectively relative speed is Relative displacement is Assume the equivalent radius of the two vehicles is r A r B The minimum safe distance is d. safe =r A +r B The TTC calculation formula is:

[0078]

[0079] When TTC(A,B) < T_threshold is satisfied, an edge is established between node A and node B. Here, TTC(A,B) is used to represent the minimum time to collision between participants A and B.

[0080] As an optional embodiment, based on the static environment node information in the scenario node information, the first scenario graph is updated to obtain the second scenario graph, including: determining at least one target static object from the static environment node information and determining the center position of the target static object; determining the target participant entity corresponding to the target static object from the first scenario graph and determining the current position of the target participant entity; obtaining the maximum perception distance between the target static object and the target participant entity, where the maximum perception distance is used to characterize the influence degree of the target static object on the target participant entity; determining the target distance based on the center position of the target static object and the current position of the target participant entity; in response to the target distance being less than the maximum perception distance, determining the second target edge information between the target static object and the target participant entity; and determining the second scenario graph based on the second target edge information and the static environment node information.

[0081] In this embodiment, after obtaining the first scenario graph, at least one target static object is determined from the static environment node information and the center position of the target static object is determined. Then, the target participant entity corresponding to the target static object is determined from the first scenario graph and the current position of the target participant entity is determined. Furthermore, the maximum perception distance between the target static object and the target participant entity can be obtained. According to the center position of the target static object and the current position of the target participant entity, the target distance is determined. The target distance and the maximum perception distance are compared. If the target distance is less than the maximum perception distance, the second target edge information between the target static object and the target participant entity can be determined. Finally, based on the second target edge information and the static environment node information obtained above, the second scenario graph is determined.

[0082] Optionally, the center position of the target static object can be referred to as the static element center position and is represented by That is, it can be represented as where x s is used to represent the abscissa of the target static object. y s is used to represent the ordinate of the target static object. The target participant entity can be abbreviated as the participant, and the current position of the target participant entity can be referred to as the participant position and is represented by where x a is used to represent the abscissa of the target participant entity. y a is used to represent the ordinate of the target participant entity. The maximum perception distance can be represented by D maxThis is used for representation. A function association flag is set; this flag indicates the degree of influence of the static element on the participants, and the association flag is expressed via f. related To represent, and f related ∈{0,1}. Static environment node information can be represented as information about static environment elements, where static environment elements refer to road infrastructure and landmarks that do not change or change very slowly over time, such as lane lines, traffic signs, traffic lights, etc.

[0083] Optionally, the static environment node information is obtained as follows: a node is created for each static environment element, and these nodes are connected to each other and to other traffic participant nodes through edges. Node characteristics are determined based on the specific characteristics of the static environment elements. The traffic light color status directly affects the behavior of traffic participants, and its node characteristics are designed considering the following: traffic light color status includes: red (prohibited), yellow (warning), green (go), off (no signal), flashing (special warning); traffic light location information: coordinates; directional control range: east, south, west, north. Lane lines are used to divide lane boundaries and driving directions, and their type and function directly affect vehicle path planning. The basic attributes of lane lines include line type, function, color, and position. Lane type attributes include solid lines (no crossing), dashed lines (crossing permitted), double solid lines (crossing strictly prohibited), double dashed lines (temporary crossing permitted), and yellow solid / dashed lines (separating opposing lanes); functional attributes include lane dividers, road edge lines, stop lines, turning lane markings, and pedestrian crossing boundaries; color attributes include white (regular lanes), yellow (opposing lanes), orange (construction zone), and blue (bus lanes); location information is coordinates or road segment identification. Lane line extension attributes also include line width and wear condition. Traffic signs convey rules or warning information through shape, color, and pattern, and their type and function must be accurately described. Basic attributes of traffic signs include sign type, shape, color, and location information. Traffic sign type attributes include warning signs (e.g., "Caution: Pedestrians"), prohibitory signs (e.g., "No Parking"), instruction signs (e.g., "Go Straight"), guide signs (e.g., "Exit Ahead"), and construction signs; shape attributes include triangle (warning), circle (prohibition), rectangle (instruction), square (guide), and diamond (construction); color attributes include yellow (warning), red (prohibition), blue (instruction), green (highway), brown (tourist), white (auxiliary), and orange (construction). Location information is coordinates or road segment identification. Extended attributes of traffic signs include additional information such as the speed limit for speed limit signs and the distance to warning signs.

[0084] Optionally, edge connections between static environmental elements can be used to represent their spatial adjacency. For example, there are edge connections between lane lines and traffic signs on the same road segment. Edge connections also exist between static environmental elements and traffic participants when certain conditions are met, representing their spatiotemporal relationships or interaction possibilities: the participant's location is... The center position of static elements is The maximum sensing distance is D max The function association flag is f. related For elements ∈{0,1}, when the distance between a traffic participant and a static environmental element is less than the maximum perceptual distance, and there is a functional association between them, an edge connection is established:

[0085] Furthermore, external environmental factors are usually global conditions that can affect all entities in the entire scene, including weather conditions (sunny, cloudy, rainy, snowy, fog) and intensity levels (small, medium, large), light intensity (daytime, dusk, night), visibility (unit: meters), etc.

[0086] It should be noted that this is only a preferred embodiment for determining the second scene graph, and the process and method for determining the second scene graph are not specifically limited. As long as it is based on the second target edge information and static environment node information, the process and method for determining the second scene graph are within the protection scope of this invention, and will not be elaborated here.

[0087] As an optional implementation method, the target scene graph is input into the test scene generation model for analysis to obtain the original test scene, including: evaluating the target scene graph using the scene evaluation index of the test scene generation model to obtain the evaluation result, wherein the evaluation result is used to characterize that the target scene graph meets the scene evaluation index; and based on the evaluation result, calling the test scene generation model to generate the original test scene.

[0088] In this embodiment, the target scene graph is evaluated using the scene evaluation index of the test scene generation model to obtain the evaluation result. Then, based on the evaluation index, the test scene generation model is invoked to generate the original test scene.

[0089] Optionally, scenario evaluation metrics may include: complexity metrics, conflict probability metrics, trajectory rationality metrics, and environmental interference intensity metrics. The complexity metric characterizes the density of dynamic traffic participants and their interactions within the scenario, and can be simply referred to as complexity. The conflict probability metric characterizes the total probability of traffic conflicts or collisions occurring in the scenario, and can be simply referred to as conflict probability. The trajectory rationality metric characterizes whether the movement trajectories and behaviors of traffic participants conform to traffic rules and physical constraints, and can be simply referred to as trajectory rationality. The environmental interference intensity metric characterizes the degree of influence of external environmental factors (such as weather, lighting, and visibility) on the scenario, and can be simply referred to as environmental interference intensity.

[0090] Optionally, the complexity metric can be defined as scenario complexity (C), which is the ratio of the number of nodes to the number of edges. This ratio reflects the density of traffic participants and their interactions in the scene, and should be avoided to be too sparse or too dense.

[0091] Optionally, the conflict probability index is obtained by using the edge set (ε) of the scene graph. adv ), calculate the mean of the global TTC, which can be obtained using the following formula: Here, TTC(A,B) is used to represent the minimum collision time between participants A and B.

[0092] Optionally, the trajectory rationality index is obtained as follows: for each participant v∈V adv Calculate the violation rate of its trajectory. Where, N invalid Used to indicate violations of the dynamic model (e.g., acceleration exceeding a). max The number of trajectory segments, N, representing traffic rules (such as running a red light). total This represents the total number of trajectory segments. The target value is set to R. violation A margin of less than 30% ensures that most of the trajectory is reasonable, but retains a small number of high-risk behaviors (such as sudden lane changes and emergency braking).

[0093] Optionally, the environmental disturbance intensity index can define the environmental disturbance intensity (I) env This is the weighted sum of visibility (S(m)), light intensity (G(lux)), and weather interference coefficient (α): Here, w1, w2, and w3 all represent weights.

[0094] As an optional implementation method, risk testing is performed on the original test scenario to obtain risk test results, including: acquiring the vehicle's false recognition rate, path deviation, and emergency braking frequency in the original test scenario, wherein the false recognition rate is used to characterize the number of times the vehicle misidentifies a moving object in the original test scenario, the path deviation is used to characterize the degree of path deviation of the vehicle in the original test scenario, and the emergency braking frequency is used to characterize the number of times the vehicle brakes suddenly in the original test scenario; in response to the false recognition rate being greater than a false recognition rate threshold, the path deviation being greater than a path deviation threshold, and the emergency braking frequency being greater than an emergency braking frequency threshold, the risk test results are determined.

[0095] In this embodiment, the vehicle's false recognition rate, path deviation, and emergency braking frequency in the original test scenario can be obtained. The false recognition rate is compared with the false recognition rate threshold, the path deviation and the path deviation threshold, and the emergency braking frequency and the emergency braking frequency threshold, respectively. If the false recognition rate is greater than the false recognition rate threshold, the path deviation is greater than the path deviation threshold, and the emergency braking frequency is greater than the emergency braking frequency threshold, the risk test result can be directly determined.

[0096] Optionally, the false recognition rate threshold can be a preset threshold, such as 15%. The path deviation threshold can be a preset threshold, such as 2 meters (m). The emergency braking frequency threshold can be a preset threshold, such as 3 times / minute.

[0097] For example, after obtaining the vehicle's misidentification rate, path deviation, and emergency braking frequency in the original test scenario, if the misidentification rate is >15%, the path deviation is >2m, and the emergency braking frequency is >3 times / minute, then the original test scenario at this time is determined to be an effective adversarial scenario, that is, there are risk events in the vehicle during driving.

[0098] In this embodiment of the invention, multiple frames of test scene images can be extracted from real-world vehicle scene videos. Then, scene node information of the vehicle is extracted from these multiple frames. This scene node information characterizes the vehicle's dynamic entity behavior, static environment layout, and external environment state within the test scene. The obtained scene node information is input into the vehicle's test scene generation model for analysis to obtain the original test scene. A risk test is then performed on the original test scene to obtain risk test results. These risk test results characterize the presence of risk events during vehicle operation within the original test scene. Based on the risk test results, the original test scene can be identified as the target test scene. Because the scene node information can be analyzed using the test scene generation model after extraction from the multiple frames of test scene images, the obtained original test scene can encompass various scene types of the vehicle in real-world scenarios. Further analysis and testing of the original test scene yields risk test results characterizing the presence of risk events during vehicle operation within the original test scene, thus achieving the goal of identifying the target test scene. This solves the technical problem of low test scene generation efficiency and improves the technical effect of increasing test scene generation efficiency.

[0099] The technical solutions of the embodiments of the present invention will be illustrated below with reference to preferred embodiments.

[0100] As autonomous driving technology advances, the complexity of vehicle perception, decision-making, and control systems continues to increase, placing higher demands on testing environments. Specifically, testing scenarios need to cover various complex and extreme situations that vehicles may encounter during operation to ensure the reliability and safety of autonomous driving technology in real-world applications.

[0101] Currently, traditional methods for modeling test scenarios involve either manually designing scenarios or constructing them based on real road test data. Manually designing scenarios requires a large amount of human resources and is inefficient. While real road tests can provide authentic traffic data, their collection costs are high, the cycle is long, and they are difficult to cover rare extreme scenarios, resulting in the technical problem of low efficiency in generating test scenarios.

[0102] Therefore, in order to solve the above problems, the present invention provides a method for determining a vehicle test scenario. The method includes: extracting multiple frames of test scenario images from a real-world video of the vehicle; extracting scene node information of the vehicle from the multiple frames of test scenario images; wherein the scene node information is used to characterize the dynamic entity behavior, static environment layout, and external environment state of the vehicle in the test scenario; inputting the obtained scene node information into a vehicle test scenario generation model for analysis to obtain the original test scenario of the vehicle; performing a risk test on the original test scenario to obtain a risk test result; wherein the risk test result is used to characterize the presence of risk events during vehicle operation in the original test scenario; and determining the original test scenario as the target test scenario based on the risk test result. Because the scene node information can be extracted from multiple test scene images, and the test scene generation model can be called to analyze the scene node information, the obtained original test scene can contain multiple scene types of vehicles in real-world scenarios. Then, the original test scene is analyzed and tested to obtain risk test results that characterize the risk events that exist in the vehicle during driving under the original test scene, so as to determine the target test scene. This solves the technical problem of low test scene generation efficiency and achieves the technical effect of improving the test scene generation efficiency.

[0103] In this embodiment of the invention, a method for autonomous driving scenario modeling and adversarial test generation based on deep generative adversarial networks (D-GANs) is proposed. This method can learn the spatiotemporal distribution characteristics of complex traffic scenarios under unsupervised conditions and fuse multi-source heterogeneous sensor data to achieve the automated generation of high-fidelity, semantically consistent, and adversarial autonomous driving test scenarios, thereby effectively improving the safety and robustness of autonomous driving systems.

[0104] Optionally, a novel deep generative model, D-GAN, is proposed. Its core lies in combining GAN with VAE, introducing a spatiotemporal graph structure to model traffic scenarios, and designing a general feature fusion module to achieve collaborative learning of multi-source data. This method can efficiently learn the distribution characteristics of complex traffic scenarios under unsupervised conditions and generate semantically consistent test scenarios, thereby significantly improving the testing efficiency and safety of autonomous driving systems.

[0105] Figure 2 This is a flowchart illustrating an automatic generation method for autonomous driving adversarial test scenarios based on D-GAN, according to an embodiment of the present invention. Figure 2 As shown, the method includes the following steps:

[0106] Step S201: Test scenario data preprocessing.

[0107] In this embodiment, a real-scene video is input, and the video is used to identify and extract targets, which are then deconstructed into editable semantic elements, including dynamic traffic participants, static environmental elements, and external environmental factors, forming semantically structured data that can be used for training.

[0108] Optionally, images are extracted frame by frame from the video, and a deep learning model is used to perform object detection on each frame to identify all entities of interest, including but not limited to dynamic traffic participants such as vehicles, pedestrians, and bicycles; at the same time, static environmental elements, such as lane lines, traffic lights, and signs, should also be identified.

[0109] Optionally, a multi-object tracking algorithm (DeepSORT) can be used to track the trajectory of each entity throughout the video sequence, recording the changes in its position, velocity, acceleration, and other motion parameters over time.

[0110] Optionally, the colors and textures in the video are analyzed to determine the weather conditions (sunny, cloudy, rainy, snowy, foggy) in the test scene, the light intensity is determined using a brightness histogram, and the field of view is estimated. For special environmental factors (e.g., construction areas), information on on-site signs is read using text recognition technology.

[0111] Optionally, the extracted information can be organized into a unified data format for easy storage and subsequent processing. For each dynamic traffic participant, a time series containing attributes such as trajectory coordinates, speed, and acceleration is created; for static environmental elements, a map layer describing their geometry, location, and type is constructed; and for external environmental factors, the corresponding global or local attribute values ​​are saved.

[0112] Step S202, scene semantic modeling.

[0113] In this embodiment, traffic participants (including the host vehicle, motor vehicles, non-motor vehicles, pedestrians, animals, etc.) are modeled to construct a graph structure in the spatio-temporal dimension to capture the behavior patterns and interaction relationships of dynamic traffic participants. Based on the trajectory data of traffic participants obtained from the scenario library, a spatio-temporal graph of the scenario is constructed. The nodes are used to represent the entity elements in the scenario, including the host vehicle and key motor vehicles, non-motor vehicles, pedestrians, animals, etc. that have important interaction relationships with the host vehicle. The node features represent the horizontal and vertical positions x, y, the horizontal and vertical speeds vx, vy, the accelerations ax, ay, the yaw angle, and a 5-bit element flag. Among them, 10000 is used to represent the host vehicle, 0X000 is used to represent the Xth key motor vehicle entity, 00X00 is used to represent the Xth key non-motor vehicle entity, 000X0 is used to represent the Xth key pedestrian entity, and 0000X is used to represent the Xth key animal entity, with a total of 12-bit node features. In the same frame, there is an edge connection between every two nodes when the conditions are met, and the edge feature is the time interval. In different frames, there is an edge connection only between nodes with the same element flag, and the edge feature is the time interval.

[0114] Optionally, the condition for an edge connection between nodes in the same frame is that nodes A and B are two traffic participants, and their positions are respectively The speeds are respectively The relative speed is The relative displacement is Assume the equivalent radius of the two vehicles is r A , r B , then the minimum safety distance is d safe = r A + r B . The TTC calculation formula is:

[0115]

[0116] It should be noted that only when the relative motion trend of the two traffic participants is "approaching" and the current distance is greater than the safety distance threshold, the TTC is calculated. When TTC(A,B) < T_threshold, an edge is established between node A and node B. Among them, TTC(A,B) is used to represent the minimum collision time between participants A and B.

[0117] Optionally, static environmental elements refer to road infrastructure and landmarks that do not change or change very slowly over time, such as lane markings, traffic signs, and traffic lights. A node is created for each static environmental element, and these nodes are connected to each other and to other traffic participant nodes via edges. Node characteristics are determined based on the specific characteristics of the static environmental element. Traffic light color states directly affect the behavior of traffic participants, and their node characteristics are designed considering the following: traffic light color states include: red (prohibited), yellow (warning), green (go), off (no signal), flashing (special warning); traffic light location information: coordinates; directional control range: east, south, west, north. Lane markings are used to delineate lane boundaries and driving directions, and their type and function directly affect vehicle path planning. Basic lane marking attributes include line type, function, color, and location. Lane type attributes include solid lines (no crossing), dashed lines (crossing permitted), double solid lines (crossing strictly prohibited), double dashed lines (temporary crossing permitted), and yellow solid / dashed lines (separating opposing lanes); functional attributes include lane dividers, road edge lines, stop lines, turning lane markings, and pedestrian crossing boundaries; color attributes include white (regular lanes), yellow (opposing lanes), orange (construction zone), and blue (bus lanes); location information is coordinates or road segment identification. Lane line extension attributes also include line width and wear condition. Traffic signs convey rules or warning information through shape, color, and pattern, and their type and function must be accurately described. Basic attributes of traffic signs include sign type, shape, color, and location information. Traffic sign type attributes include warning signs (e.g., "Caution: Pedestrians"), prohibitory signs (e.g., "No Parking"), instruction signs (e.g., "Go Straight"), guide signs (e.g., "Exit Ahead"), and construction signs; shape attributes include triangle (warning), circle (prohibition), rectangle (instruction), square (guide), and diamond (construction); color attributes include yellow (warning), red (prohibition), blue (instruction), green (highway), brown (tourist), white (auxiliary), and orange (construction). Location information is coordinates or road segment identification. Extended attributes of traffic signs include additional information such as the speed limit for speed limit signs and the distance to warning signs.

[0118] Optionally, edge connections between static environmental elements can be used to represent their spatial adjacency. For example, there are edge connections between lane lines and traffic signs on the same road segment. Edge connections also exist between static environmental elements and traffic participants when certain conditions are met, representing their spatiotemporal relationships or interaction possibilities: the participant's location is... The center position of static elements is The maximum sensing distance is D max The function association flag is f. relatedFor elements ∈{0,1}, when the distance between a traffic participant and a static environmental element is less than the maximum perceptual distance, and there is a functional association between them, an edge connection is established:

[0119] Furthermore, external environmental factors are usually global conditions that can affect all entities in the entire scene, including weather conditions (sunny, cloudy, rainy, snowy, fog) and intensity levels (small, medium, large), light intensity (daytime, dusk, night), visibility (unit: meters), etc.

[0120] Step S203, adversarial scenario generation framework.

[0121] In this embodiment, an adversarial scenario generation framework can be constructed, specifically, Figure 3 This is a schematic diagram of an adversarial scene generation framework according to an embodiment of the present invention, such as... Figure 3 As shown, the original scene video can be used as input to the multi-source data preprocessing module. Then, target recognition, trajectory tracking, and environmental analysis are performed on the original scene video to output structured data. Scene semantic modeling is then performed on the structured data to construct a spatiotemporal graph. This spatiotemporal graph is constructed using participant entity nodes, static environment element nodes, and external environment global nodes. This spatiotemporal graph is then input into the D-GAN adversarial generative framework for generating test scenarios. This framework includes a generator and a discriminator. The generator includes local perturbators, global perturbators, and environmental interference devices. The discriminator includes four metrics: complexity, conflict probability, trajectory rationality, and environmental interference intensity. Finally, the test scenario is closed-loop verified through the perception module, decision module, and control module to obtain a final test scenario that meets the requirements.

[0122] Optionally, based on the generated scene graph structure G base Based on disturbance information, generate adversarial scenario graph G that conforms to traffic rules and is logically consistent. adv .

[0123] Optionally, while maintaining traffic rules and physical constraints, adversarial perturbations are generated on the trajectories (position, speed, acceleration, etc.) of dynamic traffic participants to obtain perturbation information. This information can then be used to test the perception and decision-making capabilities of the autonomous driving system by simulating abnormal driving behaviors (such as sudden lane changes and emergency braking). Local perturbations in the dynamic traffic participant perturbation information include: minor adjustments to the trajectories of key participants, such as sudden speed changes (randomly increasing / decreasing speed within a reasonable range) and path deviations (randomly adjusting the horizontal coordinate within lane boundaries to ensure no boundary crossings). Global perturbations in the dynamic traffic participant perturbation information include: introducing logically contradictory interactive behaviors, such as illegal lane changes (forced lane changes when lane changes are prohibited by traffic lights) and intersection conflicts (generating two vehicles simultaneously entering an intersection without right-of-way).

[0124] Optionally, adversarial perturbations can be generated for static environmental elements (such as lane lines, traffic signs, and traffic lights) to obtain perturbation information. This allows for testing the perception robustness of the autonomous driving system by simulating visual interference or information tampering scenarios. Examples include wear simulation (reducing lane line visibility by changing the color from white to gray), simulating low-light scenarios such as rain or nighttime; visual interference (overlaying textured stickers, such as reflective stickers, on signs to simulate strong light interference); and traffic light status perturbations (abnormal signal flashing or malfunction).

[0125] Optionally, adversarial perturbations can be generated for global environmental factors (such as weather, lighting, and visibility) to obtain weather and environmental perturbation information. This allows for the testing of the autonomous driving system's adaptability under adverse conditions by simulating extreme or complex scenarios. Examples include weather perturbations (rain and snow simulation: increasing raindrop / snowflake density, adjusting fog concentration to blur distant objects, such as lane lines and signs, reducing visibility), lighting perturbations (strong light reflection: simulating direct sunlight or headlight interference, such as increasing the proportion of highlight areas in the brightness histogram, and low nighttime illumination: reducing overall light intensity), and visibility perturbations (dynamic occlusion: introducing random obstacles (such as billboards and trees) to occlude key elements (such as traffic lights and lane lines).

[0126] Step S204, dynamic discriminator verification mechanism.

[0127] In this embodiment, based on scene evaluation metrics, the dynamic discriminator verification mechanism in the D-GAN adversarial generation framework can be invoked to evaluate the test scene. These scene evaluation metrics may include: complexity metrics, conflict probability metrics, trajectory rationality metrics, and environmental interference intensity metrics.

[0128] Optionally, the complexity and potential conflict probability of the scene can be quantified by the topological structure and global attributes of the scene graph, ensuring that the generated scene is highly risky under physical constraints.

[0129] Optionally, the complexity metric can be defined as scenario complexity (C), which is the ratio of the number of nodes to the number of edges. This ratio reflects the density of traffic participants and their interactions in the scene, and should be avoided to be too sparse or too dense.

[0130] Optionally, the conflict probability index is obtained by using the edge set (ε) of the scene graph. adv ), calculate the mean of the global TTC, which can be obtained using the following formula: Here, TTC(A,B) is used to represent the minimum collision time between participants A and B.

[0131] Optionally, it can be verified whether the trajectories and behaviors of local traffic participants conform to physical laws and traffic rules, while introducing logical contradictions to enhance the danger.

[0132] Optionally, the trajectory rationality index is obtained as follows: for each participant v∈V adv Calculate the violation rate of its trajectory. Where, N invalid Used to indicate violations of the dynamic model (e.g., acceleration exceeding a). max (or the number of trajectory segments of traffic rules, N) total This represents the total number of trajectory segments. The target value is set to R. violation A margin of less than 30% ensures that most of the trajectory is reasonable, but retains a small number of high-risk behaviors (such as sudden lane changes and emergency braking).

[0133] Optionally, the environmental disturbance intensity index can define the environmental disturbance intensity (I) env This is the weighted sum of visibility (S(m)), light intensity (G(lux)), and weather interference coefficient (α): Here, w1, w2, and w3 all represent weights.

[0134] Step S205, closed-loop test verification.

[0135] In this embodiment, the closed-loop test verification process is as follows: the abstract representation of the scene graph (nodes, edges, attributes) is transformed into three-dimensional traffic scene data that can be parsed by the physics engine, ensuring that the generated adversarial scene not only conforms to physical laws, but also triggers the adversarial response of the autonomous driving system.

[0136] Optionally, the node attribute mapping process is as follows: Vehicle nodes: Convert the node's dynamic attributes (position, speed, direction angle) into vehicle initialization parameters supported by the physics engine. Pedestrian / obstacle nodes: Map position, speed, and direction to parameters for the pedestrian controller. Environment nodes: Convert attributes such as weather conditions and light intensity into environment configuration parameters for the physics engine. Edge relationship mapping: Convert the correlation between traffic light states and lane line attributes into traffic signal control logic for the physics engine.

[0137] Optionally, by quantifying the decision changes of the autonomous driving system in adversarial scenarios, it can be determined whether the expected adversarial effects are triggered (e.g., sudden changes in steering angle, path deviation, increased false recognition rate). The false recognition rate can be obtained through the perception module: the number of times the model misidentifies targets such as pedestrians and vehicles. The path deviation rate can be obtained through the path planning module: calculating the average displacement error between the model-planned path and the ideal path. Sudden changes in steering angle: monitoring the instantaneous rate of change of the steering wheel angle. The emergency braking frequency can be obtained through the control module: the number of times the model triggers emergency braking per unit time.

[0138] Optionally, if the false recognition rate is greater than 15%, the path deviation is greater than 2m, and the emergency braking frequency is greater than 3 times / minute, then the test scenario at this time is determined to be an effective adversarial scenario. The above process realizes automated screening and feedback optimization of adversarial examples, improving testing efficiency and system robustness.

[0139] In this embodiment, multiple frames of test scene images can be extracted from real-world vehicle video. Then, scene node information of the vehicle is extracted from these frames. This scene node information characterizes the vehicle's dynamic entity behavior, static environment layout, and external environment state within the test scene. The obtained scene node information is input into the vehicle's test scene generation model for analysis, yielding the original test scene. A risk test is then performed on the original test scene to obtain risk test results. These results characterize the presence of risk events during vehicle operation within the original test scene. Based on these results, the original test scene can be identified as the target test scene. Because the scene node information can be extracted from the multiple frames of test scene images and then analyzed using the test scene generation model, the obtained original test scene can encompass various scene types encountered by the vehicle in real-world scenarios. Further analysis and testing of the original test scene yields risk test results characterizing the presence of risk events during vehicle operation within the original test scene, thus achieving the goal of identifying the target test scene. This solves the technical problem of low test scene generation efficiency and improves the technical effect of increasing test scene generation efficiency.

[0140] According to an embodiment of the present invention, a vehicle test scenario determination device is provided. It should be noted that this vehicle test scenario determination device can be used to execute a vehicle test scenario determination method according to one of the embodiments.

[0141] Figure 4 This is a schematic diagram of a vehicle test scenario determination device according to an embodiment of the present invention. Figure 4 As shown, a test scenario determination device 400 for a vehicle may include: an acquisition unit 401, an extraction unit 402, an analysis unit 403, and a determination unit 404.

[0142] The acquisition unit 401 is used to extract multiple frames of test scene images from the real scene video of the vehicle. The real scene video is used to represent the dynamic and static scenes in which the vehicle is driving in the actual road environment.

[0143] The extraction unit 402 is used to extract scene node information of the vehicle from multiple frames of test scene images. The scene node information is used to characterize the dynamic entity behavior, static environment layout and external environment state of the vehicle in the test scene.

[0144] The analysis unit 403 is used to input scene node information into the vehicle's test scene generation model for analysis to obtain the vehicle's original test scene. The test scene generation model is trained using scene node information samples, which represent the historical information of scene node information. The original test scene represents the traffic environment state with the target test purpose.

[0145] The determining unit 404 is used to perform risk testing on the original test scenario, obtain risk test results, and determine the original test scenario as the target test scenario based on the risk test results. The risk test results are used to characterize the existence of risk events during vehicle operation under the original test scenario. The similarity between the target test scenario and the actual application scenario of the vehicle is higher than the similarity between the original test scenario and the actual application scenario.

[0146] Optionally, the extraction unit 402 includes: a detection module, used to perform object detection on multiple frames of test scene images to obtain participant node information and static environment node information of scene node information, wherein the participant node information is used to characterize moving objects appearing during vehicle driving, and the static environment node information is used to characterize static objects appearing during vehicle driving; and / or, an analysis module, used to obtain the light intensity in multiple frames of test scene images; and analyze the light intensity to obtain external environment node information of scene node information, wherein the external environment node information is used to characterize the external environment in which the vehicle is located during vehicle driving.

[0147] Optionally, the analysis unit 403 may include: a first determining module, used to determine the target scene map of the vehicle based on scene node information, wherein the target scene map is used to characterize the relationship between various entities of the vehicle in the test scene in the time dimension and the spatial dimension respectively; and a first obtaining module, used to input the target scene map into the test scene generation model for analysis to obtain the original test scene.

[0148] Optionally, the first determining module further includes: a construction submodule, used to construct a first scene graph based on participant node information in the scene node information, wherein the first scene graph is used to represent the relationship between multiple moving objects appearing during vehicle travel; a first updating submodule, used to update the first scene graph based on static environment node information in the scene node information to obtain a second scene graph, wherein the second scene graph is used to represent the relationship between multiple moving objects and multiple static objects appearing during vehicle travel; a second updating submodule, used to update the second scene graph based on external environment node information in the scene node information to obtain an initial scene graph, wherein the initial scene graph represents the relationship between multiple moving objects, multiple static objects, and different external environments in which the vehicle is located during travel; and a third updating submodule, used to update the initial scene graph based on vehicle disturbance information to obtain a target scene graph, wherein the disturbance information is used to represent sudden traffic behavior of moving objects, visual interference or information tampering of static objects, and changing external environment.

[0149] Optionally, the construction submodule is further configured to: determine at least a first participant entity and a second participant entity from the participant node information; determine the first current position and first velocity of the first participant entity, and the second current position and second velocity of the second participant entity, based on the vehicle scene library; determine the target relative displacement based on the first and second current positions; and determine the target relative velocity based on the first and second velocities; obtain the first equivalent radius corresponding to the first participant entity and the second equivalent radius corresponding to the second participant entity, and determine the minimum safe distance between the first participant entity and the second participant entity based on the first and second equivalent radii; in response to the target relative displacement being greater than the target safe distance threshold, determine the collision duration based on the target relative displacement, the target relative velocity, and the minimum safe distance, wherein the collision duration is used to characterize the time required for a collision to occur between the first participant entity and the second participant entity; in response to the collision duration being less than the duration threshold, determine the first target edge information between the first participant entity and the second participant entity; and construct a first scene graph based on the first target edge information and the participant node information.

[0150] Optionally, the first update submodule is further configured to: determine at least one target static object from the static environment node information and determine the center position of the target static object; determine the target participant entity corresponding to the target static object from the first scene graph and determine the current position of the target participant entity; obtain the maximum perceptual distance between the target static object and the target participant entity, wherein the maximum perceptual distance is used to characterize the degree of influence of the target static object on the target participant entity; determine the target distance based on the center position of the target static object and the current position of the target participant entity; in response to the target distance being less than the maximum perceptual distance, determine the second target edge information between the target static object and the target participant entity; and determine the second scene graph based on the second target edge information and the static environment node information.

[0151] Optionally, the first acquisition module further includes: an evaluation submodule, used to evaluate the target scene graph through the scene evaluation index of the test scene generation model, and obtain an evaluation result, wherein the evaluation result is used to characterize that the target scene graph meets the scene evaluation index; and a generation submodule, used to call the test scene generation model based on the evaluation result to generate the original test scene.

[0152] Optionally, the determining unit 404 further includes: a second acquisition module, used to acquire the vehicle's misidentification rate, path deviation, and emergency braking frequency in the original test scenario, wherein the misidentification rate is used to characterize the number of times the vehicle misidentifies a moving object in the original test scenario, the path deviation is used to characterize the degree of path deviation of the vehicle in the original test scenario, and the emergency braking frequency is used to characterize the number of times the vehicle brakes suddenly in the original test scenario; and the second determining module is used to determine the risk test result in response to the misidentification rate being greater than a misidentification rate threshold, the path deviation being greater than a path deviation threshold, and the emergency braking frequency being greater than an emergency braking frequency threshold.

[0153] In this embodiment, the acquisition unit extracts multiple frames of test scene images from the vehicle's real-world scene video. The real-world scene video represents the dynamic and static scenes in which the vehicle operates during actual road driving. The extraction unit extracts scene node information from the multiple frames of test scene images. This scene node information represents the vehicle's dynamic entity behavior, static environment layout, and external environment state within the test scene. The analysis unit inputs the scene node information into the vehicle's test scene generation model for analysis, obtaining the vehicle's original test scene. The test scene generation model is trained using scene node information samples. Node information samples are used to represent historical information of scene node information, and the original test scene is used to represent the traffic environment state with the target test purpose. Risk testing is performed on the original test scene by determining the unit to obtain risk test results. Based on the risk test results, the original test scene is determined as the target test scene. The risk test results are used to represent the risk events that exist when the vehicle is driving under the original test scene. The similarity between the target test scene and the actual application scene of the vehicle is higher than that between the original test scene and the actual application scene. This solves the technical problem of low test scene generation efficiency and achieves the technical effect of improving the test scene generation efficiency.

[0154] According to embodiments of the present invention, an electronic device is also provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods of various embodiments of the present invention during runtime.

[0155] According to embodiments of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is executed, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of the present invention.

[0156] According to embodiments of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of the present invention.

[0157] According to embodiments of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the methods of various embodiments of the present invention.

[0158] According to embodiments of the present invention, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of the present invention.

[0159] According to embodiments of the present invention, a vehicle is also provided that implements the methods of various embodiments of the present invention when executed.

[0160] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0161] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0162] In the several embodiments provided by this invention, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed can be through some interfaces; the indirect coupling or communication connection of units or modules can be electrical or other forms.

[0163] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0164] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0165] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, optical disks, and other media capable of storing program code.

[0166] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for determining a vehicle's test scenario, characterized in that, include: Multiple test scene images are extracted from real-world video of the vehicle. The real-world video is used to characterize the dynamic and static scenes in which the vehicle is driving in a real road environment. Scene node information of the vehicle is extracted from the multi-frame test scene images, wherein the scene node information is used to characterize the dynamic entity behavior, static environment layout and external environment state of the vehicle in the test scene; The scene node information is input into the test scene generation model of the vehicle for analysis to obtain the original test scene of the vehicle. The test scene generation model is trained by scene node information samples, which are used to represent the historical information of the scene node information. The original test scene is used to represent the traffic environment state with the target test purpose. A risk test is performed on the original test scenario to obtain a risk test result. Based on the risk test result, the original test scenario is determined as the target test scenario. The risk test result is used to characterize that there are risk events in the vehicle during driving under the original test scenario. The similarity between the target test scenario and the actual application scenario of the vehicle is higher than the similarity between the original test scenario and the actual application scenario.

2. The method according to claim 1, characterized in that, From the multi-frame test scene images, the scene node information of the vehicle is extracted, including: Object detection is performed on the multi-frame test scene images to obtain the participant node information and static environment node information of the scene node information. The participant node information is used to represent the moving objects that appear when the vehicle is driving, and the static environment node information is used to represent the static objects that appear when the vehicle is driving. And / or, The illumination intensity in the multi-frame test scene images is obtained; and the illumination intensity is analyzed to obtain the external environment node information of the scene node information, wherein the external environment node information is used to characterize the external environment in which the vehicle is located during driving.

3. The method according to claim 1, characterized in that, The scene node information is input into the vehicle's test scene generation model for analysis to obtain the vehicle's original test scene, including: Based on the scene node information, a target scene map of the vehicle is determined, wherein the target scene map is used to characterize the relationship between various entities of the vehicle in the test scene in terms of time and space. The target scene image is input into the test scene generation model for analysis to obtain the original test scene.

4. The method according to claim 3, characterized in that, Based on the scene node information, the target scene map of the vehicle is determined, including: Based on the participant node information in the scene node information, a first scene graph is constructed, wherein the first scene graph is used to represent the association relationship between multiple moving objects that appear during the vehicle's driving process; Based on the static environment node information in the scene node information, the first scene graph is updated to obtain a second scene graph, wherein the second scene graph is used to represent the association relationship between the multiple moving objects and the multiple static objects that appear when the vehicle is driving. Based on the external environment node information in the scene node information, the second scene graph is updated to obtain an initial scene graph, wherein the initial scene graph represents the relationship between the multiple moving objects, the multiple static objects and the vehicle in different external environments during the driving process; Based on the disturbance information of the vehicle, the initial scene map is updated to obtain the target scene map, wherein the disturbance information is used to characterize the sudden traffic behavior of the moving object, the visual interference or information tampering of the static object, and the changing external environment.

5. The method according to claim 4, characterized in that, Based on the participant node information in the scene node information, a first scene graph is constructed, including: At least the first participant entity and the second participant entity are determined from the participant node information; Based on the scene library of the vehicle, the first current position and first speed of the first participant entity, and the second current position and second speed of the second participant entity are determined; Based on the first current position and the second current position, determine the target's relative displacement; and based on the first velocity and the second velocity, determine the target's relative velocity; Obtain the first equivalent radius corresponding to the first participant entity and the second equivalent radius corresponding to the second participant entity, and determine the minimum safe distance between the first participant entity and the second participant entity based on the first equivalent radius and the second equivalent radius. In response to the target relative displacement being greater than the target safe distance threshold, a collision duration is determined based on the target relative displacement, the target relative velocity, and the minimum safe distance, wherein the collision duration is used to characterize the time required for a collision to occur between the first participant entity and the second participant entity; In response to the collision duration being less than a duration threshold, the first target edge information between the first participant entity and the second participant entity is determined; The first scene graph is constructed based on the first target edge information and the participant node information.

6. The method according to claim 4, characterized in that, Based on the static environment node information in the scene node information, the first scene graph is updated to obtain a second scene graph, including: Determine at least one target static object from the static environment node information, and determine the center position of the target static object; The target participant entity corresponding to the target static object is determined from the first scene graph, and the current position of the target participant entity is determined; Obtain the maximum perceptual distance between the target static object and the target participant entity, wherein the maximum perceptual distance is used to characterize the degree of influence of the target static object on the target participant entity; The target distance is determined based on the center position of the target static object and the current position of the target participant entity; In response to the target distance being less than the maximum perception distance, second target edge information between the target static object and the target participant entity is determined; The second scene graph is determined based on the second target edge information and the static environment node information.

7. The method according to claim 3, characterized in that, The target scene graph is input into the test scene generation model for analysis to obtain the original test scene, including: The target scene graph is evaluated using the scene evaluation metrics generated by the test scene generation model to obtain an evaluation result, wherein the evaluation result is used to characterize that the target scene graph meets the scene evaluation metrics. Based on the evaluation results, the test scenario generation model is invoked to generate the original test scenario.

8. The method according to claim 1, characterized in that, Risk testing was performed on the original test scenario to obtain risk test results, including: The vehicle's misidentification rate, path deviation, and emergency braking frequency are obtained in the original test scenario. The misidentification rate is used to characterize the number of times the vehicle misidentifies a moving object in the original test scenario. The path deviation is used to characterize the degree of path deviation of the vehicle in the original test scenario. The emergency braking frequency is used to characterize the number of times the vehicle brakes suddenly in the original test scenario. In response to the false recognition rate being greater than a false recognition rate threshold, the path deviation being greater than a path deviation threshold, and the emergency braking frequency being greater than an emergency braking frequency threshold, the risk test result is determined.

9. A test scenario determination device for a vehicle, characterized in that, include: The acquisition unit is used to extract multiple frames of test scene images from real scene videos of the vehicle, wherein the real scene videos are used to characterize the dynamic and static scenes in which the vehicle is driving in the actual road environment. The extraction unit is used to extract scene node information of the vehicle from the multi-frame test scene images, wherein the scene node information is used to characterize the dynamic entity behavior, static environment layout and external environment state of the vehicle in the test scene. The analysis unit is used to input the scene node information into the test scene generation model of the vehicle for analysis to obtain the original test scene of the vehicle. The test scene generation model is used to be trained by scene node information samples, the scene node information samples are used to represent the historical information of the scene node information, and the original test scene is used to represent the traffic environment state with the target test purpose. The determining unit is used to perform risk testing on the original test scenario, obtain risk test results, and determine the original test scenario as a target test scenario based on the risk test results. The risk test results are used to characterize that there are risk events in the vehicle during driving under the original test scenario, and the similarity between the target test scenario and the actual application scenario of the vehicle is higher than the similarity between the original test scenario and the actual application scenario.

10. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the storage medium is located to perform the method according to any one of claims 1 to 8.

12. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 8.

13. A vehicle, characterized in that, The vehicle is used to perform the method according to any one of claims 1 to 8.

Images