A method for generating a vehicle active and passive integrated test scene considering damage

Abstract

The present application relates to the technical field of vehicle testing, and relates to a vehicle active and passive integrated test scene generation method considering damage. A plurality of traffic accident pre-collision data are acquired. A scene generation model is trained based on the traffic accident pre-collision data, wherein the damage degree of a weak road traffic participant in each traffic accident pre-collision data is a conditional input of the scene generation model, used to guide the scene generation model to learn and generate a pre-collision scene corresponding to the conditional input. In response to a generation request of the pre-collision scene, the generated condition corresponding to the generation request is input into the trained scene generation model, and pre-collision scene data corresponding to the generated condition is obtained, which can be used for vehicle active and passive safety testing. This makes it possible to verify in the vehicle active and passive safety testing whether the active and passive integration of the vehicle can effectively reduce the damage degree in the subsequent inevitable collision, thereby supporting the calibration and verification of the active and passive integrated control strategy.

Description

Technical Field

[0001] This specification relates to the field of vehicle testing technology, and in particular to a method for generating integrated active and passive vehicle test scenarios that take damage into account. Background Technology

[0002] With the road safety situation for vulnerable road users remaining severe, the number of injuries and fatalities caused by road traffic accidents worldwide remains high each year. With the rapid development of intelligent technologies, automotive safety systems are gradually shifting from passive protection to integrated active and passive protection. Intelligent driving systems, by integrating information from multiple sensors such as millimeter-wave radar, cameras, and lidar, achieve precise perception of the surrounding environment, enabling vehicles to identify potential hazards earlier than a collision. Simultaneously, the new generation of integrated safety controllers can coordinate and decide on multiple protective measures, such as active braking, seatbelt pretensioning, and seat position adjustment, thus forming a collaborative active and passive protection network.

[0003] However, existing technologies have significant shortcomings in evaluating this integrated active and passive protection. Due to the numerous combinations of dangerous scenarios in real roads and the highly proactive and random behavior of traffic participants, traditional methods based on limited test cases are difficult to comprehensively evaluate active and passive protection capabilities.

[0004] Therefore, this specification provides a method for generating integrated active and passive vehicle test scenarios that take into account damage. Summary of the Invention

[0005] This specification provides a method for generating integrated active and passive vehicle test scenarios that takes damage into account, in order to partially solve the aforementioned problems existing in the prior art.

[0006] The following technical solution is adopted in this specification: This specification provides a method for generating integrated active and passive vehicle test scenarios that take damage into account, including: S1. Acquire several traffic accident pre-collision data, each traffic accident pre-collision data includes the trajectory data of the accident vehicle within a preset time period before the collision, the trajectory data of the vulnerable road traffic participant within the preset time period before the collision, the age data of the vulnerable road traffic participant, the time label of the traffic accident pre-collision data, and the degree of injury of the vulnerable road traffic participant. S2. A scenario generation model is trained based on each traffic accident pre-collision data, wherein the degree of injury of the vulnerable road traffic participant in each traffic accident pre-collision data is the conditional input of the scenario generation model, which is used to guide the scenario generation model to learn and generate the pre-collision scenario corresponding to the conditional input. S3. In response to the request to generate a pre-collision scene, input the generation conditions corresponding to the generation request into the trained scene generation model to obtain the pre-collision scene data corresponding to the generation conditions.

[0007] Based on the aforementioned technical means, the pre-collision scenario data generated by this solution can be used for vehicle active and passive safety testing. This allows for the verification of whether the integrated active and passive protection of the vehicle effectively reduces the degree of damage in subsequent unavoidable collisions, thus supporting the calibration and verification of integrated active and passive control strategies. Using the degree of damage as a conditional guiding vector to train the scenario generation model, the model is forced to learn "what combination of relative trajectory, speed, and angle will lead to a specific level of damage." Therefore, when conducting vehicle active and passive safety testing under a desired degree of damage, the scenario generation model can generate pre-collision trajectories in a targeted manner. For example, when inputting the condition of "high degree of damage," it can generate extreme pre-collision trajectories that are highly likely to cause serious injury, significantly increasing the generation density of high-risk scenarios in safety testing. The scenario generation model can learn the heterogeneity of injury levels between the elderly and younger populations under the same kinematic collision conditions. This means that the generated scenarios are not only "severely impacted" but also "accurately targeted at specific vulnerable groups (such as the elderly)." This provides more refined test data for developing pedestrian protection systems for an aging society.

[0008] Furthermore, the scene generation model described in S2 is a variational autoencoder model that includes an encoding layer and a decoding layer; The scenario generation model in S2, trained based on pre-collision data from various traffic accidents, specifically includes: For each traffic accident pre-collision data, the traffic accident pre-collision data is input into the encoding layer of the scene generation model to be trained, and the latent variables under the conditional input in the traffic accident pre-collision data are output by the encoding layer. The conditional inputs and latent variables in the traffic accident pre-collision data are input into the decoding layer to obtain the pre-collision scene data output by the decoding layer; The pre-collision scene data output by the decoding layer is input into the pre-trained classification model to obtain the degree of damage output by the classification model. Based on the difference between the traffic accident pre-collision data and the pre-collision scene data output by the decoding layer, the reconstruction loss value is determined; based on the difference between the approximate posterior distribution and the standard Gaussian distribution of the latent variable, the divergence loss value is determined; and based on the difference between the damage level output by the classification model and the damage level of vulnerable road traffic participants in the traffic accident pre-collision data, the conditional loss value is determined. Based on the reconstruction loss value, the divergence loss value, and the conditional loss value, a comprehensive loss value is determined, and the scene generation model to be trained is trained with the minimum comprehensive loss value as the optimization objective.

[0009] Based on the aforementioned technical methods, by calculating the difference between the original trajectory and the generated trajectory (i.e., the pre-collision data of the traffic accident and the pre-collision scene data output by the decoding layer), it is ensured that the generated scene does not deviate from the distribution of real accident data in terms of kinematic morphology. This prevents the model from generating invalid trajectories that are "too fast to violate the laws of physics" or "drift instantaneously" simply to cater to the damage classifier. By feeding the pre-collision scene data generated by the decoder into the pre-trained classification model, the loss function calculates the difference in damage degree, meaning that the training process actively penalizes those generated results that "look like the input trajectory and the output trajectory, but do not hurt (or are too heavy) upon impact," ensuring that the generated trajectory, while maintaining a reasonable physical morphology, actually points to the specified damage degree in its implicit collision energy. The encoding layer not only compresses the trajectory into latent variables, but also requires that the latent variables contain damage degree features that can be "verified by the classifier" during training. By adding divergence loss to train the scene generation model, the latent space is constrained to be regular, ensuring that the latent variables learned under different conditions cluster around a standard normal distribution, facilitating sampling and generation. After training, the scene generation model not only learns how to "draw" trajectories, but also internalizes the physical intuition of what consequences those trajectories will lead to. This means that when generating massive amounts of test scenes in the future, it is not necessary to call the classification model for verification every time. The trajectories output by the decoder are already high-value samples that have been filtered for damage, which greatly reduces the computational cost of scene post-processing.

[0010] Furthermore, the method also includes S4: Based on the pre-collision scenario data corresponding to the generation conditions, the pre-collision scenario corresponding to the generation conditions is simulated in simulation software as an integrated active and passive test scenario for the vehicle.

[0011] Furthermore, S1 acquires several types of pre-collision data related to traffic accidents, specifically including: Raw traffic accident data involving vulnerable road traffic participants are obtained from a pre-designed database of future mobile traffic accident scenarios. The original traffic accident data is reconstructed using traffic accident reconstruction analysis software. Based on the reconstruction results, traffic accident collision data is determined. The traffic accident collision data includes dynamic data and static data. The dynamic data includes the trajectory data of the accident vehicle and the trajectory data of the vulnerable road traffic participant. The static data includes the age data of the vulnerable road traffic participant and the degree of injury of the vulnerable road traffic participant. Based on the traffic accident collision data, determine the traffic accident pre-collision data.

[0012] Furthermore, based on the traffic accident collision data, traffic accident pre-collision data is determined, specifically including: According to a preset time period, the dynamic data in the traffic accident collision data is sliced ​​to determine multiple dynamic sub-data; From multiple dynamic sub-data sets, filter out the dynamic sub-data sets before the collision occurs; For each dynamic sub-data point before the collision, the time label of the dynamic sub-data point is determined based on the time distance between the time period to which the dynamic sub-data point belongs and the time of the collision. Based on the dynamic sub-data, the static data, and the time tag of the dynamic sub-data, the traffic accident pre-collision data corresponding to the dynamic sub-data is determined.

[0013] Furthermore, the static data also includes an accident condition label; the accident condition label is one of the following: intersection ghost camera, intersection non-ghost camera, non-intersection ghost camera, and non-intersection non-ghost camera.

[0014] Furthermore, the degree of injury of vulnerable road traffic participants and the accident condition label in each traffic accident pre-collision data are used as conditional inputs to the scenario generation model.

[0015] This specification provides a vehicle active and passive integrated test scenario generation device that considers damage, including: The acquisition module is used to acquire several traffic accident pre-collision data. Each traffic accident pre-collision data includes the trajectory data of the accident vehicle within a preset time period before the collision, the trajectory data of the vulnerable road traffic participant within the preset time period before the collision, the age data of the vulnerable road traffic participant, the time label of the traffic accident pre-collision data, and the degree of injury of the vulnerable road traffic participant. The training module is used to train a scenario generation model based on various traffic accident pre-collision data. The degree of injury of vulnerable road traffic participants in each traffic accident pre-collision data is the conditional input of the scenario generation model, which is used to guide the scenario generation model to learn and generate the pre-collision scenario corresponding to the conditional input. The generation module is used to respond to the generation request of the pre-collision scene, input the generation conditions corresponding to the generation request into the trained scene generation model, and obtain the pre-collision scene data corresponding to the generation conditions.

[0016] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for generating an integrated active and passive vehicle test scenario that considers damage.

[0017] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a method for generating an integrated active and passive vehicle test scenario that takes damage into account.

[0018] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: The pre-collision scenario data generated by this solution can be used for vehicle active and passive safety testing. This allows for the verification of whether the integrated active and passive protection of the vehicle effectively reduces the degree of damage in subsequent unavoidable collisions, thus supporting the calibration and verification of integrated active and passive control strategies. The scenario generation model is trained using the degree of damage as a conditional guiding vector. The model is forced to learn "what combination of relative trajectory, speed, and angle will lead to a specific level of damage." Therefore, when conducting vehicle active and passive safety testing under a desired degree of damage, the scenario generation model can generate pre-collision trajectories in a targeted manner. For example, when inputting a "high degree of damage" condition, it can generate extreme pre-collision trajectories that are highly likely to cause serious injury, significantly increasing the generation density of high-risk scenarios in safety testing. The scenario generation model can learn the heterogeneity of injury levels between the elderly and younger populations under the same kinematic collision conditions. This means that the generated scenarios are not only "severely impacted" but also "accurately targeted at specific vulnerable groups (such as the elderly)." This provides more refined test data for developing pedestrian protection systems for an aging society. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart illustrating a method for generating integrated active and passive vehicle test scenarios that takes into account damage, as provided in an embodiment of this specification. Figure 2 This is a schematic diagram of a vehicle active and passive integrated test scenario generation device that takes damage into account, as provided in this specification. Figure 3 This specification provides a corresponding Figure 1 A schematic diagram of the structure of an electronic device. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0021] In embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0022] The technical solutions provided in the various embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0023] Figure 1 A flowchart illustrating a method for generating integrated active and passive vehicle test scenarios considering damage, as provided in the embodiments of this specification, includes the following steps: S1: Acquire several traffic accident pre-collision data, each traffic accident pre-collision data includes the trajectory data of the accident vehicle within a preset time period before the collision, the trajectory data of the vulnerable road traffic participant within the preset time period before the collision, the age data of the vulnerable road traffic participant, the time label of the traffic accident pre-collision data, and the degree of injury of the vulnerable road traffic participant.

[0024] This specification describes the process of generating integrated active and passive vehicle test scenarios that consider damage. In the embodiments described herein, this process can be executed by a server. However, this specification does not limit the type of device or platform used to generate these integrated active and passive vehicle test scenarios; for example, a personal computer, mobile terminal, or other such device or platform can also be used. For ease of description, the following description uses a server as the executing entity.

[0025] In one or more embodiments of this specification, the server may acquire several (examples) traffic accident pre-collision data. Each traffic accident pre-collision data includes trajectory data of the accident vehicle within a preset time period before the collision, trajectory data of a vulnerable road traffic participant within the preset time period before the collision, age data of the vulnerable road traffic participant, a timestamp of the traffic accident pre-collision data, and the degree of injury of the vulnerable road traffic participant. The timestamp indicates the proximity of the traffic accident pre-collision data to the moment of the collision.

[0026] Specifically, the server can retrieve raw traffic accident data involving vulnerable road users from the pre-defined Future Accident Scenario Study (FASS) database. This FASS database, built by the institution that proposed this solution, contains a large number of traffic accident cases involving vehicles and vulnerable road users. Each case includes key information such as road surveillance video, in-depth accident investigation data, and injury reports. All cases have been anonymized and do not involve the personal privacy information of accident participants, serving as a crucial data foundation for accident analysis and traffic safety research. The FASS database provides reliable data support for this research.

[0027] In the FASS database, the original traffic accident data involving vulnerable road traffic participants can be screened according to the following criteria: (1) involving a single vehicle and a single pedestrian. (2) the collision point is the front of the vehicle. (3) complete records of the individual information and injury status of the pedestrian or two-wheeled vehicle occupant. (4) on-site road monitoring video and in-depth investigation data can effectively support the extraction of road type and scene characteristics at the accident location, as well as the reconstruction of the complete interaction process between the vehicle and pedestrian before the accident. Cases with missing injury information, missing information of key time periods before the collision, or unstable trajectory reconstruction are all excluded. After the above screening, several accident cases are finally obtained for the research on pre-collision scene generation. Specifically, the information in the accident cases used in this study includes the following two categories: (1) Dynamic temporal characteristics: Complete trajectory data of the accident vehicle and vulnerable road traffic participants from a certain time point (e.g., 5 seconds) before the collision to the moment of the collision. The sampling frequency is 100Hz. The trajectory data is accompanied by temporal information such as speed and heading angle. This part of the temporal information can be reconstructed based on traffic accident reconstruction analysis software, such as PC Crash software.

[0028] (2) Static characteristics: pedestrian's age, gender, and Abbreviated Injury Scale (AIS) for each part of the body; road type at the accident location, whether there is visual obstruction, and whether the vulnerable road traffic participant is a ghost pedestrian.

[0029] Therefore, traffic accident reconstruction analysis software (such as PC Crash) is used to reconstruct the original traffic accident data. Based on the reconstruction results, traffic accident collision data can be determined. Traffic accident collision data includes dynamic data and static data. Dynamic data includes the trajectory data of the accident vehicles and the trajectory data of vulnerable road traffic participants. Static data includes the age data of vulnerable road traffic participants and the degree of injury of vulnerable road traffic participants (which can be determined by AIS).

[0030] This allows for the further determination of pre-collision data based on traffic accident collision data. Specifically, the server can slice the dynamic data in the traffic accident collision data according to preset time periods to determine multiple dynamic sub-data. From these multiple dynamic sub-data, the dynamic sub-data prior to the collision is selected. For example, for the trajectory data of the accident vehicle from 5 seconds before the collision to the moment of the collision, slicing it into preset time periods of three seconds can yield the trajectory data of the accident vehicle and vulnerable road users from 5 seconds to 3 seconds before the collision, from 4 seconds to 2 seconds before the collision, and from 3 seconds before the collision, respectively, as multiple dynamic sub-data.

[0031] Next, for each dynamic sub-data point before the collision, the server determines its timestamp based on the proximity of the time period to which the dynamic sub-data point belongs to the moment of the collision. Finally, based on the dynamic sub-data point, the static data, and the timestamp of the dynamic sub-data point, the server determines the corresponding pre-collision data for the traffic accident.

[0032] For each dynamic sub-data point, the time labels can be divided into three groups based on their distance from the collision time, with each group assigned a discrete time label 1, 2, or 3. Larger numbers indicate greater distance from the collision time. These time labels reflect the urgency of each dynamic sub-data point in the time dimension and serve as an additional static feature for each sub-data point.

[0033] S2: Train a scenario generation model based on pre-collision data of each traffic accident, wherein the degree of injury of vulnerable road traffic participants in each pre-collision data of a traffic accident is the conditional input of the scenario generation model, which is used to guide the scenario generation model to learn and generate the pre-collision scenario corresponding to the conditional input.

[0034] In one or more embodiments of this specification, the server can train a scenario generation model based on each traffic accident pre-collision data, wherein the degree of injury of vulnerable road traffic participants in each traffic accident pre-collision data is a conditional input of the scenario generation model, used to guide the scenario generation model to learn and generate a pre-collision scenario corresponding to the conditional input.

[0035] Specifically, the scene generation model is a variational autoencoder (VAE) model that includes an encoding layer and a decoding layer.

[0036] For each traffic accident pre-collision data point, the server can input the pre-collision data into the encoding layer of the scene generation model to be trained, obtaining the latent variables under the conditional input of the traffic accident pre-collision data output by the encoding layer. Then, the conditional input and latent variables from the traffic accident pre-collision data are input into the decoding layer to obtain the pre-collision scene data output by the decoding layer. During model training, conditional independence is assumed. , As a latent variable, For conditional input, for The unconditional prior distribution, for The conditional prior distribution, It follows a standard Gaussian distribution.

[0037] Then, the pre-collision scene data output from the decoding layer is input into the pre-trained classification model to obtain the damage level output by the classification model.

[0038] Therefore, the server can determine the reconstruction loss value based on the difference between the pre-collision data of the traffic accident and the pre-collision scene data output by the decoding layer; this difference is positively correlated with the reconstruction loss value. The divergence loss value is determined based on the difference between the approximate posterior distribution and the standard Gaussian distribution of the latent variables. The conditional loss value is also determined based on the difference between the damage level output by the classification model and the damage level of vulnerable road traffic participants in the pre-collision data of the traffic accident; this difference is positively correlated with the conditional loss value. Finally, based on the reconstruction loss value, divergence loss value, and conditional loss value, a comprehensive loss value is determined, and the scene generation model to be trained is trained with the minimum comprehensive loss value as the optimization objective.

[0039] In this model, the degree of injury of vulnerable road traffic participants in the pre-collision data of traffic accidents is used as a conditional input for the scene generation model. In this specification, the degree of injury of vulnerable road traffic participants can be divided into three types: no injury or mild injury, severe injury, and critical or fatal injury. The degree of injury can be vectorized and represented by two-dimensional ordinal cumulative encoding, which maps the above three degrees of injury to feature vectors [0,0], [1,0], and [1,1] respectively.

[0040] For the pre-collision data in traffic accidents, aside from the data used as conditional inputs, the age data of vulnerable road users and the time labels of the pre-collision data are considered static data and can be copied along the time dimension to facilitate concatenation with the dynamic sub-data, namely the trajectory data of the accident vehicle and the trajectory data of vulnerable road users within a preset time period before the collision. Similarly, the conditional input, namely the degree of injury of vulnerable road users, is also copied along the time dimension and concatenated with the dynamic sub-data before being input into the scene generation model.

[0041] By calculating the difference between the original trajectory and the generated trajectory (i.e., the pre-collision data of the traffic accident and the pre-collision scene data output by the decoding layer), it is ensured that the generated scene does not deviate from the distribution of real accident data in terms of kinematic morphology. This prevents the model from generating invalid trajectories that are "too fast to violate the laws of physics" or "drift instantaneously" simply to cater to the damage classifier. By feeding the pre-collision scene data generated by the decoder into the pre-trained classification model, the loss function calculates the difference in damage severity. This means that the training process actively penalizes those generated results where "the input trajectory and output trajectory look similar, but the impact is not painful (or too severe)," ensuring that the generated trajectory, while maintaining a reasonable physical morphology, actually points to the specified damage severity in its implicit collision energy. The encoding layer not only compresses the trajectory into latent variables, but also requires these latent variables to contain damage severity features that can be verified by the classifier during training. After training, the scene generation model not only learns how to "draw" trajectories, but also internalizes the physical intuition of what consequences the trajectory will lead to. This means that when generating a large number of test scenarios in the future, there is no need to call the classification model for verification every time. The trajectory output by the decoder is already a high-value sample that has been filtered for damage, which greatly reduces the computing power cost of post-processing the scenario.

[0042] S3. In response to the request to generate a pre-collision scene, input the generation conditions corresponding to the generation request into the trained scene generation model to obtain the pre-collision scene data corresponding to the generation conditions.

[0043] In one or more embodiments of this specification, after the scene generation model has been trained, the server can respond to a pre-collision scene generation request. Based on the generation request, generation conditions are determined, and the generation conditions corresponding to the generation request are input into the trained scene generation model to obtain pre-collision scene data corresponding to the generation conditions.

[0044] Subsequently, based on the pre-collision scenario data corresponding to the generation conditions, the pre-collision scenario corresponding to the generation conditions can be simulated in simulation software as a vehicle active and passive integrated test scenario. Thus, the pre-collision scenario data output by the scenario generation model is transformed into a repeatable and controllable vehicle active and passive integrated test scenario, constructing virtual test cases. This provides a unified and reliable test basis for evaluating the vehicle's pedestrian active collision avoidance and passive damage reduction performance, and can be used to comprehensively evaluate the realism of the generated pre-collision scenarios and their application potential in vehicle pedestrian protection performance evaluation.

[0045] based on Figure 1 This paper presents a method for generating integrated active and passive vehicle test scenarios that considers damage. The pre-collision scenario data generated by this method can be used for vehicle active and passive safety testing. This allows for the verification of whether the integrated active and passive protection of the vehicle effectively reduces the degree of damage in subsequent unavoidable collisions, thus supporting the calibration and verification of integrated active and passive control strategies. The scenario generation model is trained using the degree of damage as a conditional guiding vector. The model is forced to learn "what combination of relative trajectory, speed, and angle will lead to a specific level of damage." Therefore, when conducting vehicle active and passive safety tests under a desired degree of damage, the scenario generation model can generate pre-collision trajectories in a targeted manner. For example, when inputting the condition of "high degree of damage," it can generate extreme pre-collision trajectories that are highly likely to cause serious injury, significantly increasing the generation density of high-risk scenarios in safety testing. The scenario generation model can learn the heterogeneity of injury levels between the elderly and younger populations under the same kinematic collision conditions. This means that the generated scenarios are not only "severely impacted" but also "accurately targeted at specific vulnerable groups (such as the elderly)." This provides more refined test data for developing pedestrian protection systems for an aging society.

[0046] Furthermore, in one or more embodiments of this specification, the classification model can be trained using the following method: The server can acquire historical data of pre-collision scenarios and use the damage level of this historical data as a label. This historical data is then input into the classification model to be trained, and the output of the model is determined. Next, based on the difference between the output of the classification model and the labels, a loss value is determined, and the three-class classification model is trained with the minimum loss value as the optimization objective. The difference is positively correlated with the loss value.

[0047] In one or more embodiments of this specification, the static data constituting traffic accident pre-collision data may further include an accident condition label. The accident condition label is one of the following: a ghost camera at an intersection, a non-ghost camera at an intersection, a ghost camera at a non-intersection, and a non-ghost camera at a non-intersection. The accident condition label can be extracted from the static features of the raw traffic accident data obtained from the FASS database.

[0048] Furthermore, the degree of injury and accident condition labels of vulnerable road traffic participants in each traffic accident pre-collision data can be used as conditional inputs to the scene generation model, guiding the scene generation model to learn and generate pre-collision scenes corresponding to the degree of injury and accident condition of the conditional inputs.

[0049] Therefore, for the four accident condition labels, referencing the vectorized representation of damage severity using two-dimensional ordinal cumulative encoding, the four accident condition labels are represented using four-dimensional one-hot encoding. The damage severity encoding is then concatenated with the accident condition label encoding to construct a unified six-dimensional condition vector, which serves as the conditional input to the scene generation model. Building upon this, to further enhance the structured representation of damage features in the scene generation model, a learnable ordered damage embedding module can be introduced. This module applies a softplus function constraint to the embedding increments of adjacent damage severity levels, constructing a feature representation that increases progressively with damage severity. This ensures that the arrangement direction of damage severity in the latent space aligns with the actual risk increase direction, explicitly preserving the orderliness of risk levels at the representation level, thereby more efficiently guiding the generation model to learn the latent semantics that evolve smoothly with damage risk.

[0050] In one or more embodiments of this specification, due to significant differences in the acquisition coordinate system, initial positions of people and vehicles, and heading angles of different traffic accident raw data in the FASS database, directly using them for scene generation model training can easily lead to the model learning global pose deviations unrelated to the essence of the scene, making it difficult to stabilize the model training. Therefore, this study standardizes the trajectory time series of the dynamic sub-data obtained from each slice through rigid body coordinate transformation: First, the trajectories of vehicles and vulnerable road traffic participants are uniformly translated, so that the initial coordinates of vehicles are translated to the origin; further, the trajectories of vehicles and vulnerable road traffic participants are uniformly rotated, so that the initial heading angle of vehicles is rotated to 90 degrees (along the positive y-axis). Through the above processing, the trajectory time series of all dynamic sub-data are aligned to a unified reference coordinate system, enabling the model to focus more on learning the relative motion patterns and trajectory time series characteristics of vehicle-dynamic sub-data, and their relationship with the degree of injury of vulnerable road traffic participants.

[0051] Furthermore, to improve the generalization ability of the scene generation model for symmetrical scenes, this specification allows for further mirroring enhancement of the trajectory data in the standardized dynamic sub-data along the y-axis. Through the above slicing, standardization, and mirroring operations, the limited pre-collision data of traffic accidents can be expanded, thereby effectively supporting the training of the scene generation model. It should be noted that, to avoid information leakage caused by highly overlapping dynamic sub-data generated from the same original traffic accident data slices entering different datasets simultaneously, this paper uses the original traffic accident data as the smallest independent unit for dataset partitioning in subsequent model training.

[0052] Based on the damage-considered vehicle active and passive integrated test scenario generation method provided in one or more embodiments of this specification, following the same approach, this specification also provides a corresponding damage-considered vehicle active and passive integrated test scenario generation device, such as... Figure 2 As shown.

[0053] Figure 2 This specification provides a schematic diagram of a vehicle active and passive integrated test scenario generation device that considers damage, specifically including: The acquisition module 200 is used to acquire several traffic accident pre-collision data. Each traffic accident pre-collision data includes the trajectory data of the accident vehicle within a preset time period before the collision, the trajectory data of the vulnerable road traffic participant within the preset time period before the collision, the age data of the vulnerable road traffic participant, the time label of the traffic accident pre-collision data, and the degree of injury of the vulnerable road traffic participant. Training module 202 is used to train a scenario generation model based on each traffic accident pre-collision data, wherein the degree of injury of the vulnerable road traffic participant in each traffic accident pre-collision data is the conditional input of the scenario generation model, which is used to guide the scenario generation model to learn and generate the pre-collision scenario corresponding to the conditional input. The generation module 204 is used to respond to the generation request of the pre-collision scene by inputting the generation conditions corresponding to the generation request into the trained scene generation model to obtain the pre-collision scene data corresponding to the generation conditions.

[0054] Optionally, the scene generation model in the training module 202 is a variational autoencoder model including an encoding layer and a decoding layer; The training module 202 is further configured to, for each traffic accident pre-collision data, input the traffic accident pre-collision data into the encoding layer of the scene generation model to be trained, obtain the latent variables of the traffic accident pre-collision data output by the encoding layer, input the latent variables into the decoding layer, obtain the pre-collision scene data output by the decoding layer, input the pre-collision scene data output by the decoding layer into a pre-trained classification model, obtain the damage level output by the classification model, determine a first loss value based on the difference between the traffic accident pre-collision data and the pre-collision scene data output by the decoding layer; and determine a second loss value based on the difference between the damage level output by the classification model and the damage level of vulnerable road traffic participants in the traffic accident pre-collision data, determine a comprehensive loss value based on the first loss value and the second loss value, and train the scene generation model to be trained with the minimum comprehensive loss value as the optimization objective.

[0055] Optionally, the device further includes a simulation module 206; The simulation module 206 is used to simulate the pre-collision scenario corresponding to the generation conditions in the simulation software based on the pre-collision scenario data corresponding to the generation conditions, as a vehicle active and passive integrated test scenario.

[0056] Optionally, the acquisition module 200 is further configured to acquire original traffic accident data involving vulnerable road traffic participants from a preset future mobile traffic accident scenario research database, reconstruct the original traffic accident data using traffic accident reconstruction analysis software, determine traffic accident collision data based on the reconstruction results, the traffic accident collision data includes dynamic data and static data, the dynamic data includes trajectory data of the accident vehicle and trajectory data of the vulnerable road traffic participants, the static data includes age data of the vulnerable road traffic participants and the degree of injury of the vulnerable road traffic participants, and determine traffic accident pre-collision data based on the traffic accident collision data.

[0057] Optionally, the acquisition module 200 is further configured to slice the dynamic data in the traffic accident collision data according to a preset time period to determine multiple dynamic sub-data, filter out each dynamic sub-data before the collision from the multiple dynamic sub-data, determine the time tag of each dynamic sub-data before the collision based on the time period to which the dynamic sub-data belongs and the time of the collision, and determine the traffic accident pre-collision data corresponding to the dynamic sub-data based on the dynamic sub-data, the static data, and the time tag of the dynamic sub-data.

[0058] Optionally, the static data in the acquisition module 200 may further include an accident condition label; the accident condition label is one of the following: a ghost camera at an intersection, a non-ghost camera at an intersection, a ghost camera at a non-intersection, and a non-ghost camera at a non-intersection.

[0059] Optionally, the degree of injury and accident condition label of vulnerable road traffic participants in each traffic accident pre-collision data in the acquisition module 200 are used as conditional inputs to the scene generation model.

[0060] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 A method for generating integrated active and passive vehicle test scenarios that takes damage into account is provided.

[0061] This instruction manual also provides Figure 3 The diagram shows a schematic structural representation of the electronic device. Figure 3 As shown, at the hardware level, this electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above. Figure 1 A method for generating integrated active and passive vehicle test scenarios that takes damage into account is provided.

[0062] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software. In other words, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0063] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0064] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0065] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0066] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0067] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0069] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0071] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0072] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0073] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic or disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0074] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0075] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0076] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0077] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0078] The above description is merely an embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for generating integrated active and passive vehicle test scenarios considering damage, characterized in that, include: S1. Acquire several traffic accident pre-collision data, each traffic accident pre-collision data includes the trajectory data of the accident vehicle within a preset time period before the collision, the trajectory data of the vulnerable road traffic participant within the preset time period before the collision, the age data of the vulnerable road traffic participant, the time label of the traffic accident pre-collision data, and the degree of injury of the vulnerable road traffic participant. S2. A scenario generation model is trained based on each traffic accident pre-collision data, wherein the degree of injury of the vulnerable road traffic participant in each traffic accident pre-collision data is the conditional input of the scenario generation model, which is used to guide the scenario generation model to learn and generate the pre-collision scenario corresponding to the conditional input. S3. In response to the request to generate a pre-collision scene, input the generation conditions corresponding to the generation request into the trained scene generation model to obtain the pre-collision scene data corresponding to the generation conditions.

2. The method for generating a vehicle active and passive integrated test scenario considering damage as described in claim 1, characterized in that, The scene generation model described in S2 is a variational autoencoder model that includes an encoding layer and a decoding layer; S2 uses scenario generation models trained based on pre-collision data from various traffic accidents, specifically including: For each traffic accident pre-collision data, the traffic accident pre-collision data is input into the encoding layer of the scene generation model to be trained, and the latent variables under the conditional input in the traffic accident pre-collision data are output by the encoding layer. The conditional inputs and latent variables in the traffic accident pre-collision data are input into the decoding layer to obtain the pre-collision scene data output by the decoding layer; The pre-collision scene data output by the decoding layer is input into the pre-trained classification model to obtain the degree of damage output by the classification model. Based on the difference between the traffic accident pre-collision data and the pre-collision scene data output by the decoding layer, the reconstruction loss value is determined; based on the difference between the approximate posterior distribution and the standard Gaussian distribution of the latent variable, the divergence loss value is determined; and based on the difference between the damage level output by the classification model and the damage level of vulnerable road traffic participants in the traffic accident pre-collision data, the conditional loss value is determined. Based on the reconstruction loss value, the divergence loss value, and the conditional loss value, a comprehensive loss value is determined, and the scene generation model to be trained is trained with the minimum comprehensive loss value as the optimization objective.

3. The method for generating an integrated active and passive vehicle test scenario considering damage as described in claim 1, characterized in that, The method further includes S4: Based on the pre-collision scenario data corresponding to the generation conditions, the pre-collision scenario corresponding to the generation conditions is simulated in simulation software as an integrated active and passive test scenario for the vehicle.

4. The method for generating a vehicle active and passive integrated test scenario considering damage as described in claim 1, characterized in that, S1 acquires several types of pre-collision data related to traffic accidents, specifically including: Raw traffic accident data involving vulnerable road traffic participants are obtained from a pre-designed database of future mobile traffic accident scenarios. The original traffic accident data is reconstructed using traffic accident reconstruction analysis software. Based on the reconstruction results, traffic accident collision data is determined. The traffic accident collision data includes dynamic data and static data. The dynamic data includes the trajectory data of the accident vehicle and the trajectory data of the vulnerable road traffic participant. The static data includes the age data of the vulnerable road traffic participant and the degree of injury of the vulnerable road traffic participant. Based on the traffic accident collision data, determine the traffic accident pre-collision data.

5. The method for generating a vehicle active and passive integrated test scenario considering damage as described in claim 4, characterized in that, Based on the traffic accident collision data, the traffic accident pre-collision data is determined, specifically including: According to a preset time period, the dynamic data in the traffic accident collision data is sliced ​​to determine multiple dynamic sub-data; From multiple dynamic sub-data sets, filter out the dynamic sub-data sets before the collision occurs; For each dynamic sub-data point before the collision, the time label of the dynamic sub-data point is determined based on the time distance between the time period to which the dynamic sub-data point belongs and the time of the collision. Based on the dynamic sub-data, the static data, and the time tag of the dynamic sub-data, the traffic accident pre-collision data corresponding to the dynamic sub-data is determined.

6. The method for generating a vehicle active and passive integrated test scenario considering damage as described in claim 5, characterized in that, The static data also includes accident condition labels; the accident condition labels are one of the following: intersection ghost camera, intersection non-ghost camera, non-intersection ghost camera, and non-intersection non-ghost camera.

7. The method for generating an integrated active and passive vehicle test scenario considering damage as described in claim 6, characterized in that, The degree of injury and accident condition label of vulnerable road traffic participants in each traffic accident pre-collision data are used as conditional inputs to the scenario generation model.

8. A vehicle active and passive integrated test scenario generation device considering damage, characterized in that, include: The acquisition module is used to acquire several traffic accident pre-collision data. Each traffic accident pre-collision data includes the trajectory data of the accident vehicle within a preset time period before the collision, the trajectory data of the vulnerable road traffic participant within the preset time period before the collision, the age data of the vulnerable road traffic participant, the time label of the traffic accident pre-collision data, and the degree of injury of the vulnerable road traffic participant. The training module is used to train a scenario generation model based on various traffic accident pre-collision data. The degree of injury of vulnerable road traffic participants in each traffic accident pre-collision data is the conditional input of the scenario generation model, which is used to guide the scenario generation model to learn and generate the pre-collision scenario corresponding to the conditional input. The generation module is used to respond to the generation request of the pre-collision scene, input the generation conditions corresponding to the generation request into the trained scene generation model, and obtain the pre-collision scene data corresponding to the generation conditions.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 7.

Images