Automatic driving multi-background vehicle high-risk test scene generation method

Abstract

The embodiment of the present disclosure provides a kind of automatic driving multi-background vehicle high-risk test scene generation method, belong to data processing technical field, specifically include: step 1, extract the space-time characteristics of all vehicles in the accident occurrence process;Step 2, combine the pre-crash timing characteristics of the collision vehicle and the timing interaction characteristics of the non-collision vehicle and the accident vehicle, fuse static characteristics, create scene space-time coupling matrix;Step 3, according to the preset sampling interval, sample the scene space-time coupling matrix, obtain multiple scene sections;Step 4, input all scene sections into GAN network, generalize to generate scene space-time coupling section;Step 5, based on the scene space-time coupling section, according to the distribution of pre-crash timing characteristics, timing interaction characteristics and static characteristics in the scene space-time coupling matrix, the complete space-time coupling test scene is obtained by expansion. Through the scheme of the present disclosure, the safety and reliability of the automatic driving vehicle are improved.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a method for generating high-risk test scenarios for autonomous vehicles with multiple backgrounds. Background Technology

[0002] Currently, when conducting scenario-based testing and evaluation of autonomous vehicles, the trajectories of background vehicles directly affect the performance and experimental results of the autonomous vehicles. However, the collected actual trajectory data, limited by sample size and diversity, may not include the key attribute combinations crucial for AV testing.

[0003] It is evident that there is an urgent need for a method to generate high-risk test scenarios for autonomous vehicles with multiple backgrounds, which can improve the safety and reliability of autonomous vehicles. Summary of the Invention

[0004] In view of this, the present disclosure provides a method for generating high-risk test scenarios for autonomous vehicles with multiple backgrounds, which at least partially solves the problem of poor safety and reliability in the prior art.

[0005] This disclosure provides a method for generating high-risk test scenarios for autonomous vehicles with multiple backgrounds, including:

[0006] Step 1: Based on the testing requirements of intelligent vehicles, perform in-depth accident analysis on real traffic accidents, collect and organize relevant accident data and reconstruct it, and extract the spatiotemporal features of all vehicles during the accident. Among them, all vehicles include colliding vehicles and non-colliding vehicles. The spatiotemporal features include the pre-collision timing features of colliding vehicles, the timing interaction features between non-colliding vehicles and colliding vehicles, and static features.

[0007] Step 2: Combine the pre-collision timing features of the colliding vehicles with the timing interaction features between the non-collision vehicles and the accident vehicles, and fuse static features to create a scene spatiotemporal coupling matrix.

[0008] Step 3: Sample the scene spatiotemporal coupling matrix according to the preset sampling interval to obtain multiple scene cross-sections;

[0009] Step 4: Input all scene cross-sections into the GAN network to generalize and generate spatiotemporal coupled cross-sections of the scene;

[0010] Step 5: Based on the scene spatiotemporal coupling section, and according to the distribution of pre-collision timing features, timing interaction features and static features in the scene spatiotemporal coupling matrix, expand to obtain a complete spatiotemporal coupling test scene.

[0011] According to a specific implementation of an embodiment of this disclosure, step 1 specifically includes:

[0012] Step 1.1: Extract the state information of the colliding vehicle at each moment within a preset time period before the collision to form a pre-collision time sequence feature. The state information includes the longitudinal coordinates, lateral coordinates, velocity, acceleration, orientation, steering and relative position difference of the colliding vehicle at each coordinate.

[0013] Step 1.2: Extract the conflict patterns between the non-collision vehicle and the collision vehicle within a preset time period before the collision to form temporal interaction features. The conflict patterns include longitudinal same direction, longitudinal opposite direction, left side and right side.

[0014] Step 1.3: Extract information on vehicle type, lighting conditions, weather, road surface water, road type, number of lanes, and line of sight obstruction from the accident scene to form static features.

[0015] According to a specific implementation of this disclosure, the calculation process for generating the spatiotemporal coupling section of the scene is as follows:

[0016]

[0017] in, This represents a noise vector obtained from a Gaussian or uniform distribution. Represents real samples Data distribution This represents a sample generated by the generator. Data distribution Expressing expectations, G represents the discriminator, and G represents the generator. This represents the loss function.

[0018] According to a specific implementation of this disclosure, the collision vehicles include a host vehicle and a target vehicle used for testing during the simulation process, and step 5 specifically includes:

[0019] Step 5.1: Extract static features from the scene. ( These elements constitute the spatial elements of the scene.

[0020] Step 5.2: Define the target vehicle and each non-collision vehicle as background vehicles, and extract the dynamic information of each background vehicle. ( These constitute the scene time elements, where the state description of each background vehicle is... ;

[0021] Step 5.3: Expand the state description of each background vehicle according to the preset step size and update rules until the simulation duration is reached;

[0022] Step 5.4, fuse static features in the scene ( ), to construct a complete spatiotemporal coupled test scenario.

[0023] According to a specific implementation of an embodiment of this disclosure, the update rule is as follows:

[0024]

[0025]

[0026]

[0027]

[0028]

[0029] ( )

[0030] in, This represents the dynamic state of the target vehicle at time t. These represent the target vehicle's x-coordinate, y-coordinate, velocity, acceleration, and orientation at time t, respectively. Indicates the time step. Indicates a moment in time. Rate of change of orientation Indicates vehicles under different conflict types The distribution This represents the reciprocal of the acceleration at time t. Indicating different forms of conflict The distribution of .

[0031] The autonomous driving multi-background high-risk test scenario generation scheme in this embodiment includes: Step 1, according to the testing requirements of intelligent vehicles, performing in-depth accident deconstruction on real traffic accidents, collecting and organizing relevant accident data and reconstructing it, extracting the spatiotemporal features of all vehicles during the accident, wherein all vehicles include colliding vehicles and non-colliding vehicles, and the spatiotemporal features include the pre-collision timing features of colliding vehicles, the timing interaction features between non-colliding vehicles and colliding vehicles, and static features; Step 2, combining the pre-collision timing features of colliding vehicles and the timing interaction features between non-colliding vehicles and accident vehicles, and fusing static features to create a scene spatiotemporal coupling matrix; Step 3, sampling the scene spatiotemporal coupling matrix according to a preset sampling interval to obtain multiple scene cross-sections; Step 4, inputting all scene cross-sections into a GAN network to generalize and generate scene spatiotemporal coupling cross-sections; Step 5, based on the scene spatiotemporal coupling cross-sections, expanding to obtain a complete spatiotemporal coupling test scenario according to the distribution of pre-collision timing features, timing interaction features and static features in the scene spatiotemporal coupling matrix.

[0032] The beneficial effects of this disclosure are as follows: The scheme utilizes generative adversarial networks to generate expanded accident scene cross-sections. Then, based on the distribution of dynamic, static, and interactive features within the accident scene cross-sections, the cross-sections are expanded into complete spatiotemporally coupled test scenarios. This enrichment-enhancing method can improve the performance evaluation of autonomous vehicles in high-risk situations involving multi-context vehicle interactions. By comprehensively testing autonomous vehicles in the generated test scenarios, their performance under various complex multi-context vehicle interaction conditions can be better evaluated. This contributes to improving the safety and reliability of autonomous vehicles and promotes their practical application. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart illustrating a method for generating high-risk test scenarios for autonomous vehicles with multiple backgrounds, provided in an embodiment of this disclosure. Detailed Implementation

[0035] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0036] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0037] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0038] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0039] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0040] This disclosure provides a method for generating high-risk test scenarios for autonomous vehicles with multiple backgrounds. This method can be applied to the testing process of autonomous vehicles in big data scenarios.

[0041] See Figure 1 This is a flowchart illustrating a method for generating high-risk test scenarios for autonomous vehicles with multiple backgrounds, provided in an embodiment of this disclosure. Figure 1 As shown, the method mainly includes the following steps:

[0042] Step 1: Based on the testing requirements of intelligent vehicles, perform in-depth accident analysis on real traffic accidents, collect and organize relevant accident data and reconstruct it, and extract the spatiotemporal features of all vehicles during the accident. Among them, all vehicles include colliding vehicles and non-colliding vehicles. The spatiotemporal features include the pre-collision timing features of colliding vehicles, the timing interaction features between non-colliding vehicles and colliding vehicles, and static features.

[0043] Furthermore, step 1 specifically includes:

[0044] Step 1.1: Extract the state information of the colliding vehicle at each moment within a preset time period before the collision to form a pre-collision time sequence feature. The state information includes the longitudinal coordinates, lateral coordinates, velocity, acceleration, orientation, steering and relative position difference of the colliding vehicle at each coordinate.

[0045] Step 1.2: Extract the conflict patterns between the non-collision vehicle and the collision vehicle within a preset time period before the collision to form temporal interaction features. The conflict patterns include longitudinal same direction, longitudinal opposite direction, left side and right side.

[0046] Step 1.3: Extract information on vehicle type, lighting conditions, weather, road surface water, road type, number of lanes, and line of sight obstruction from the accident scene to form static features.

[0047] In practice, based on the testing requirements of intelligent vehicles, in-depth accident analysis of real traffic accidents can be performed to collect and organize relevant accident data. The collected accident data can then be reconstructed, and the spatiotemporal characteristics of all vehicles during the accident can be extracted.

[0048] For example, the pre-collision trajectory matrix contains the necessary elements for building the simulation test, divided into dynamic features, namely pre-collision temporal features, and temporal interaction features. Pre-collision temporal features include: the state of each vehicle in the accident scenario at every moment within 10 seconds before the collision; specifically, with a step size of 0.1 seconds, the longitudinal coordinates, lateral coordinates, velocity, acceleration, orientation, steering, and relative position difference of each vehicle within each step. Temporal interaction features include: the interaction features between each non-collision vehicle and the colliding vehicle in the accident scenario within 10 seconds before the collision; specifically, the position, behavior, and conflict type relative to the colliding vehicle. Conflict types include four conflict types: longitudinally in the same direction, longitudinally in opposite directions, left-side, and right-side. Static features include various vehicle types, lighting conditions, weather, road surface water conditions, road type, number of lanes, and line-of-sight obstruction in the scene.

[0049] Step 2: Combine the pre-collision timing features of the colliding vehicles with the timing interaction features between the non-collision vehicles and the accident vehicles, and fuse static features to create a scene spatiotemporal coupling matrix.

[0050] In practice, after obtaining the pre-collision timing features and static features of the colliding vehicles, as well as the timing interaction features between the non-collision vehicles and the accident vehicles, these three types of feature data can be spliced ​​together to form a scene spatiotemporal coupling matrix.

[0051] Step 3: Sample the scene spatiotemporal coupling matrix according to the preset sampling interval to obtain multiple scene cross-sections;

[0052] In practice, the scene is extracted into several scene sections, specifically from 10.0 seconds before the accident to 1.0 second before the accident. The scene sections are sampled in 0.5-second intervals, which contain all information in the scene at the selected time. Specifically, this includes the status of each vehicle, including longitudinal coordinates, lateral coordinates, speed, acceleration, orientation, steering, relative position difference, position of non-collision vehicles relative to collision vehicles, behavior, conflict type, lighting conditions, weather, road surface water conditions, road type, first collision point, number of lanes, and line of sight obstruction, in order to obtain multiple scene sections.

[0053] Step 4: Input all scene cross-sections into the GAN network to generalize and generate spatiotemporal coupled cross-sections of the scene;

[0054] Based on the above embodiments, the calculation process for generating the spatiotemporal coupling section of the scene is as follows:

[0055]

[0056] in, This represents a noise vector obtained from a Gaussian or uniform distribution. Represents real samples Data distribution This represents a sample generated by the generator. Data distribution Expressing expectations, G represents the discriminator, and G represents the generator. This represents the loss function.

[0057] In practice, a GAN network is used to generalize and generate scene cross-sections. Specifically, each scene cross-section is used as a sample as input to the GAN, where the basic mathematical form of the GAN is:

[0058]

[0059] in It is a noise vector obtained from a Gaussian or uniform distribution. It is a real sample The data distribution, and Samples generated by the generator The data distribution. When the discriminator is optimal, the generator's loss function is equivalent to minimizing the true data distribution. With the distribution of generated data The Jensen-Shannon divergence between them. Specifically, This represents the various dynamic and static variables in S3.

[0060] Step 5: Based on the scene spatiotemporal coupling section, and according to the distribution of pre-collision timing features, timing interaction features and static features in the scene spatiotemporal coupling matrix, expand to obtain a complete spatiotemporal coupling test scene.

[0061] Based on the above embodiments, the collision vehicles include the host vehicle and the target vehicle used for testing during the simulation process, and step 5 specifically includes:

[0062] Step 5.1: Extract static features from the scene. ( These elements constitute the spatial elements of the scene.

[0063] Step 5.2: Define the target vehicle and each non-collision vehicle as background vehicles, and extract the dynamic information of each background vehicle. ( These constitute the scene time elements, where the state description of each background vehicle is... ;

[0064] Step 5.3: Expand the state description of each background vehicle according to the preset step size and update rules until the simulation duration is reached;

[0065] Step 5.4, fuse static features in the scene ( ), to construct a complete spatiotemporal coupled test scenario.

[0066] Furthermore, the update rule is as follows:

[0067]

[0068]

[0069]

[0070]

[0071]

[0072] ( )

[0073] in, This represents the dynamic state of the target vehicle at time t. These represent the target vehicle's x-coordinate, y-coordinate, velocity, acceleration, and orientation at time t, respectively. Indicates the time step. Indicates a moment in time. Rate of change of orientation Indicates vehicles under different conflict types The distribution This represents the reciprocal of the acceleration at time t. Indicating different forms of conflict The distribution of .

[0074] In specific implementation, in step 5, based on the scene cross-section generated by the generator in step 4, spatiotemporal coupling expansion is performed according to the distribution of dynamic features, static features, and interactive features. The specific steps can be as follows:

[0075] 1) Extract static environment information from the scene ( These elements constitute the spatial elements of the scene.

[0076] 2) Considering that there are only two vehicles involved in the collision at the defined accident scene, the target vehicle and each non-collision vehicle are defined as background vehicles, and the dynamic information of each background vehicle is extracted, without considering the main vehicle when creating the test scenario. ( These constitute the scene time elements, where the state description of each background vehicle is... ;

[0077] 3) Expand the time sequence state of each background vehicle in 0.1s increments until the simulation duration is reached. Update the parameters at each time step as follows:

[0078]

[0079]

[0080]

[0081]

[0082]

[0083] ( )

[0084] 4) Integrate static environmental information from the scene ( ), to build a complete scenario.

[0085] in Indicates the first A static information variable. This represents the dynamic state of the target vehicle at time t. These represent the target vehicle's x-coordinate, y-coordinate, velocity, acceleration, and orientation at time t, respectively. The time step is set to 0.1s. Indicates a moment in time. Rate of change of orientation Vehicles under different conflict types The distribution Let be the reciprocal of the acceleration at time t. Indicating different forms of conflict The distribution of .

[0086] The method for generating high-risk test scenarios for autonomous vehicles with multiple backgrounds provided in this embodiment utilizes generative adversarial networks to expand the cross-section of an accident scene. Then, based on the distribution of dynamic, static, and interactive features within the accident scene cross-section, the scene cross-section is expanded into a complete spatiotemporally coupled test scenario. This method allows for full utilization of accident data to generate more diverse high-risk test scenarios for vehicles with multiple backgrounds, thereby improving the performance evaluation capabilities of autonomous vehicles under various dangerous multi-vehicle interaction conditions. This technological innovation brings new possibilities to autonomous driving testing and helps promote the development and application of autonomous driving technology.

[0087] It should be understood that the various parts of this disclosure can be implemented in hardware, software, firmware, or a combination thereof.

[0088] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. An automatic driving multi-background vehicle high-risk test scene generation method, characterized in that, include: Step 1: Based on the testing requirements of intelligent vehicles, perform in-depth accident analysis on real traffic accidents, collect and organize relevant accident data and reconstruct it, and extract the spatiotemporal features of all vehicles during the accident. Among them, all vehicles include colliding vehicles and non-colliding vehicles. The spatiotemporal features include the pre-collision timing features of colliding vehicles, the timing interaction features between non-colliding vehicles and colliding vehicles, and static features. Step 1 specifically includes: Step 1.1: Extract the state information of the colliding vehicle at each moment within a preset time period before the collision to form a pre-collision time sequence feature. The state information includes the longitudinal coordinates, lateral coordinates, velocity, acceleration, orientation, steering and relative position difference of the colliding vehicle at each coordinate. Step 1.2: Extract the conflict patterns between the non-collision vehicle and the collision vehicle within a preset time period before the collision to form temporal interaction features. The conflict patterns include longitudinal same direction, longitudinal opposite direction, left side and right side. Step 1.3: Extract information on vehicle type, lighting conditions, weather, road surface water, road type, number of lanes, and line-of-sight obstruction at the accident scene to form static features; Step 2: Combine the pre-collision timing features of the colliding vehicles with the timing interaction features between the non-collision vehicles and the accident vehicles, and fuse static features to create a scene spatiotemporal coupling matrix. Step 3: Sample the scene spatiotemporal coupling matrix according to the preset sampling interval to obtain multiple scene cross-sections; Step 4: Input all scene cross-sections into the GAN network to generalize and generate spatiotemporal coupled scene cross-sections. The calculation process for generating these spatiotemporal coupled scene cross-sections is as follows: in, This represents a noise vector obtained from a Gaussian or uniform distribution. Represents real samples Data distribution This represents a sample generated by the generator. Data distribution Expressing expectations, G represents the discriminator, and G represents the generator. Represents the loss function; Step 5: Based on the scene spatiotemporal coupling section, according to the distribution of pre-collision timing features, timing interaction features and static features in the scene spatiotemporal coupling matrix, expand to obtain a complete spatiotemporal coupling test scene; The collision vehicles include the main vehicle and the target vehicle used for testing during the simulation. Step 5 specifically includes: Step 5.1, extracting static features in the scene constitute scene space elements;​ Step 5.2, define the target vehicle and each non-collision vehicle as background vehicles, and extract the dynamic information of each background vehicle , constitute the scene time element, wherein the state description of each background vehicle is , wherein respectively represent the x-coordinate, y-coordinate, speed, acceleration, and orientation of the target vehicle at time t.​ Step 5.3: Expand the state description of each background vehicle according to the preset step size and update rules until the simulation duration is reached; Step 5.4, fuse static features in the scene ( ), to construct a complete spatiotemporal coupled test scenario.

2. The method according to claim 1, characterized in that... The update rule is ( ) in, This represents the dynamic state of the target vehicle at time t. Indicates the time step. Indicates a moment in time. Rate of change of orientation Indicates vehicles under different conflict types The distribution This represents the reciprocal of the acceleration at time t. Indicating different forms of conflict The distribution of .

Images