COMPUTER-IMPLEMENTED METHOD AND SYSTEM FOR CREATING A VIRTUAL VEHICLE ENVIRONMENT

Description

[0001] The present invention relates to a computer-implemented method for generating a virtual vehicle environment for testing highly automated driving functions of a motor vehicle using pre-captured video image data, radar data and / or a lidar point cloud of a real vehicle environment.

[0002] The present invention further relates to a system for generating a virtual vehicle environment for testing highly automated driving functions of a motor vehicle using pre-captured video image data, radar data and / or a lidar point cloud of a real vehicle environment.

[0003] The present invention further relates to a computer program and a computer-readable data carrier. State of the art

[0004] Graphical user interfaces for testing highly automated driving functions of a motor vehicle usually have a number of components that enable the management of parameter sets, the creation of a virtual vehicle environment and experiment management.

[0005] A scene setup of the virtual vehicle environment, i.e. a definition of static and dynamic objects of a scene, is carried out by manual configuration and import of objects stored in an object library.

[0006] CN 000205121971 U discloses a method for testing autonomous vehicles. According to the method, an autonomous vehicle is created in a simulation environment. According to pre-recorded state information of a virtual vehicle, a traffic environment is constructed in the simulation environment to enable the autonomous vehicle to drive in the traffic environment.

[0007] However, the above-mentioned methods have in common that there is a high level of effort required to create the virtual vehicle environment for testing the highly automated driving functions of the motor vehicle, which results in high personnel and cost expenditure.

[0008] US2019311487A1 discloses a method for providing a scene based on a lidar point cloud, which is augmented with objects represented by stored lidar point clouds. The augmentation can be selected by a user and / or by providing additional image data.

[0009] However, the procedure does not describe the creation of a new virtual vehicle environment, but only the extension of existing scenes.

[0010] US2019156558A1 describes a system and method for creating scenes for a virtual reality application.

[0011] However, no method is provided that suggests the most accurate representation of pre-captured video image data, radar data and / or a lidar point cloud in a virtual, synthetic vehicle environment.

[0012] Accordingly, there is a need to improve existing methods and systems for generating a virtual vehicle environment for testing highly automated driving functions of a motor vehicle in such a way that a simplified, more efficient and more cost-effective creation of the virtual vehicle environment is enabled.

[0013] It is therefore an object of the invention to provide a computer-implemented method, a system, a computer program and a computer-readable data carrier which enable a simplified, more efficient and more cost-effective generation of a virtual vehicle environment for testing highly automated driving functions of a motor vehicle. Disclosure of the invention

[0014] The object is achieved according to the invention by a computer-implemented method for generating a virtual vehicle environment for testing highly automated driving functions of a motor vehicle using pre-recorded video image data, radar data and / or a lidar point cloud of a real vehicle environment according to patent claim 1, a system for generating a virtual vehicle environment for testing highly automated driving functions of a motor vehicle using pre-recorded video image data, radar data and / or a lidar point cloud of a real vehicle environment according to patent claim 11, a computer program according to patent claim 12 and a computer-readable data carrier according to patent claim 13.

[0015] The invention relates to a computer-implemented method for generating a virtual vehicle environment for testing highly automated driving functions of a motor vehicle using pre-captured video image data, radar data and / or a lidar point cloud of a real vehicle environment.

[0016] The method comprises receiving the pre-captured video image data, radar data and / or the lidar point cloud of the real vehicle environment including a plurality of real objects pre-labeled or subsequently labeled by applying at least a first machine learning algorithm.

[0017] The real vehicle environment corresponds to the vehicle environment of the motor vehicle in road traffic, in particular in a variety of traffic situations. The real objects can be static objects such as traffic signs, buildings, vegetation, and / or parked vehicles. Furthermore, the real objects can be dynamic objects such as moving vehicles.

[0018] The method further comprises generating a first feature vector representing a respective real object by applying a second machine learning algorithm to the respective real object and storing the first feature vector.

[0019] The method further comprises providing a plurality of stored second feature vectors representing synthetically generated objects and identifying a second feature vector having a greatest similarity measure to the first feature vector.

[0020] The similarity measure can be defined based on predetermined properties of the first feature vector and the second feature vector.

[0021] The synthetically generated objects are divided into different object categories and represent the real objects identified in the video image data, radar data and / or the lidar point cloud of the real vehicle environment.

[0022] The method further comprises selecting the identified second feature vector and calling up a stored synthetic object which is assigned to the second feature vector and corresponds to the real object, or procedurally generating the synthetic object corresponding to the real object depending on the identified similarity measure, and integrating the synthetic object into a predetermined virtual vehicle environment, wherein the virtual vehicle environment is a computer-generated representation of the sensor-based recorded real vehicle environment, wherein if the similarity measure (MA) of the identified second feature vector (M2) to the first feature vector (M1) is greater than or equal to a predetermined threshold value (T), the identified second feature vector (M2) is selected and the stored synthetic object (14a, 14b, 14c, 14d, 14e) assigned to the second feature vector (M2), which corresponds to the real object (12a, 12b, 12c, 12d,12e) is called (S6A), and if the similarity measure (MA) of the identified second feature vector (M2) to the first feature vector (M1) is less than a predetermined threshold value (T), the synthetic object (14a, 14b, 14c, 14d, 14e) is procedurally generated (S6B).,

[0023] The virtual vehicle environment is a computer-generated representation of the sensor-based real vehicle environment.

[0024] Procedural generation refers to a method for creating 3D objects in real time during the execution of the computer program.

[0025] The 3D objects are not generated randomly, but rather the generation follows deterministic algorithms in order to be able to generate the same content again and again under the same initial conditions.

[0026] The invention further relates to a system for generating a virtual vehicle environment for testing highly automated driving functions of a motor vehicle using pre-acquired video image data, radar data, and / or a lidar point cloud of a real vehicle environment. The system comprises means for receiving the pre-acquired video image data, radar data, and / or the lidar point cloud of the real vehicle environment, including a plurality of real objects that have been pre-identified or subsequently identified by applying at least one first machine learning algorithm.

[0027] The system further comprises means for applying at least a first machine learning algorithm to the video image data, the radar data and / or the lidar point cloud for detecting and classifying the plurality of real objects, as well as means for generating a first feature vector representing a respective real object, which means are configured to apply a second machine learning algorithm to the respective real object, and means for storing the first feature vector.

[0028] The system further comprises means for providing a plurality of stored second feature vectors representing synthetically generated objects and means for identifying a second feature vector having a greatest degree of similarity to the first feature vector.

[0029] The system further comprises means for selecting the identified second feature vector and calling a stored synthetic object associated with the second feature vector, which corresponds to the real object, and means for procedurally generating the synthetic object corresponding to the real object, wherein the means for selecting the identified second feature vector and calling a stored synthetic object associated with the second feature vector or the means for procedurally generating the synthetic object corresponding to the real object are used depending on the identified similarity measure.

[0030] Furthermore, the system has means for integrating the synthetic object into a predetermined virtual vehicle environment, wherein the virtual vehicle environment is a computer-generated representation of the sensor-based detected real vehicle environment, wherein if the similarity measure (MA) of the identified second feature vector (M2) to the first feature vector (M1) is greater than or equal to a predetermined threshold value (T), the identified second feature vector (M2) is selectable and the stored synthetic object (14a, 14b, 14c, 14d, 14e) associated with the second feature vector (M2), which corresponds to the real object (12a, 12b, 12c, 12d, 12e), is callable, and wherein if the similarity measure (MA) of the identified second feature vector (M2) to the first feature vector (M1) is less than a predetermined threshold value (T), the synthetic object (14a, 14b, 14c, 14d, 14e) can be generated procedurally.

[0031] The invention further relates to a computer program with program code for carrying out the method according to the invention when the computer program is executed on a computer.

[0032] The invention further relates to a computer-readable data carrier with program code of a computer program in order to carry out the method according to the invention when the computer program is executed on a computer.

[0033] One idea of the present invention is to automatically construct a synthetic 3D scene from real video image data of a test drive of a motor vehicle.

[0034] Due to the generation of a first feature vector representing a respective real object by applying a machine learning algorithm to the respective real object, it is advantageously not necessary to perform direct object recognition and assignment to a synthetic object.

[0035] Direct object recognition would require the respective objects to be recognized by an artificial neural network. This would require extensive training of the artificial neural network using a large set of training data.

[0036] Instead, within the scope of the present invention, existing data of typical objects in a traffic scene, which are stored in a database, can be used. This allows for a simple assignment of the second feature vector to a corresponding synthetic object.

[0037] By comparing the respective first feature vectors representing real objects with already existing synthetically generated objects representing second feature vectors, it is thus advantageously possible to automatically assign corresponding real and synthetically generated objects, so that the synthetically generated objects thus identified can then be integrated into the virtual vehicle environment.

[0038] By automating the scene setup of the virtual vehicle environment using the data obtained from the real vehicle environment, a significant simplification of the scene setup of the virtual vehicle environment can be achieved in an advantageous manner, combined with a considerable increase in efficiency and a concomitant reduction in costs.

[0039] Further embodiments of the present invention are the subject of the further subclaims and the following description with reference to the figures.

[0040] According to one aspect of the invention, the method further comprises that the characterization, in particular detection and classification, of the plurality of real objects is carried out by a supervised learning algorithm or by an unsupervised learning algorithm, in particular an artificial neural network.

[0041] Thus, an algorithm suitable for a data type and / or other systemic specifications can be used advantageously.

[0042] This advantageously makes it possible to call up a synthetic object that has a high degree of similarity to the selected real object.

[0043] Even if there is no sufficient match between the first feature vector corresponding to the real object and one of the plurality of synthetically generated objects corresponding to the second feature vectors, a synthetic object corresponding to the real object can nevertheless be advantageously generated by the procedural generation of the synthetic object.

[0044] According to a further aspect of the invention, the method further comprises the step of performing the procedural generation of the synthetic object using the first feature vector generated by the second machine learning algorithm.

[0045] This advantageously ensures that the synthetic object generated in this way corresponds to the underlying real object to a high degree.

[0046] According to a further aspect of the invention, the method further comprises the second machine learning algorithm for generating, in particular for encoding, the first feature vector being formed by a first variational autoencoder. The variational autoencoder has the advantage that it can generate a dimensionally reduced or complexity-reduced representation of the underlying real data.

[0047] According to a further aspect of the invention, the method further comprises generating the second feature vector representing a respective synthetic object by applying a third machine learning algorithm to the respective synthetic object, and storing the second feature vector in a data store.

[0048] Thus, an assignment or a feature comparison of the first feature vectors and the second feature vectors can be carried out and thus an assignment of the real objects to synthetic objects can be carried out.

[0049] According to a further aspect of the invention, the method further comprises classifying the real objects contained in the video image data, the radar data and / or the lidar point cloud into main classes by the first machine learning algorithm, wherein the real objects assigned to a main class are classified into subclasses by applying a fourth machine learning algorithm.

[0050] This advantageously allows for a step-by-step classification of the objects contained in the real data. In a first step, the real objects are roughly classified into main classes. In a second step, a detailed classification of the real objects assigned to the main classes into corresponding subclasses is then performed.

[0051] According to a further aspect of the invention, the method further comprises that, in the case of similarity, in particular a similar data structure, of the first feature vector and the second feature vector, a direct comparison is carried out between the first feature vector and the plurality of second feature vectors in order to identify the similarity measure.

[0052] Thus, an efficient identification of a suitable second feature vector or a comparison of the first feature vectors with the second feature vectors can advantageously be made possible without any other intermediate steps.

[0053] According to a further aspect of the invention, the method further comprises that, in the case of dissimilarity, in particular a dissimilar data structure, of the first feature vector and the second feature vector, a classifier establishing a correlation between the first feature vector and the second feature vector is used to identify the similarity measure of the second feature vector to the first feature vector.

[0054] Thus, when using different types of feature vectors with, for example, different types of coding, it is advantageously possible to establish a correlation between the different feature vectors and thus to assign corresponding first feature vectors to corresponding second feature vectors.

[0055] According to a further aspect of the invention, the method further comprises that the plurality of stored first feature vectors and / or second feature vectors represent 2-D image data, 3-D image data, radar data and / or a lidar point cloud.

[0056] The source data underlying the second feature vectors can thus advantageously be of different types, so that a large number of different source data obtained in reality can be used and correlated with or assigned to synthetically generated objects or object data.

[0057] According to a further aspect of the invention, the method further comprises that if the similarity measure of the identified second feature vector to the first feature vector is less than the predetermined threshold value, a message is generated that the stored plurality of second feature vectors do not have a sufficient similarity measure to the identified first feature vector.

[0058] The user is thus advantageously informed about this fact and can, for example, expand a data store of the synthetically generated objects underlying the second feature vectors.

[0059] The features of the method described herein are also applicable to other virtual environments, such as testing other types of vehicles in different environments. Short description of the drawings

[0060] For a better understanding of the present invention and its advantages, reference is now made to the following description in conjunction with the accompanying drawings.

[0061] The invention will be explained in more detail below using exemplary embodiments shown in the schematic illustrations of the drawings. Fig. 1 is a flowchart of a method for generating a virtual vehicle environment for testing highly automated driving functions of a motor vehicle according to a preferred embodiment of the invention; Fig. 2 is a schematic diagram of the application of the method to a real vehicle environment for generating the virtual vehicle environment according to the preferred embodiment of the invention; and Fig. 3 is a diagram of a system for generating the virtual vehicle environment for testing highly automated driving functions of the motor vehicle according to the preferred embodiment of the invention.

[0062] Unless otherwise indicated, like reference numerals refer to like elements in the drawings. Detailed description of the embodiments

[0063] Fig. 1 shows a flowchart of a method for generating a virtual vehicle environment for testing highly automated driving functions of a motor vehicle according to a preferred embodiment of the invention.

[0064] The method comprises receiving S1A, S1B the pre-captured video image data 10a, radar data and / or the lidar point cloud of the real vehicle environment U1 including a plurality of pre-marked S1a or subsequently marked S1W by applying at least one first machine learning algorithm A1, real objects 12a, 12b, 12c, 12d, 12e.

[0065] The method further comprises generating S2 a first feature vector M1 representing a respective real object 12a, 12b, 12c, 12d, 12e by applying a second machine learning algorithm A2 to the respective real object 12a, 12b, 12c, 12d, 12e and storing S3 the first feature vector M1.

[0066] The method further comprises providing S4 a plurality of stored second feature vectors M2, which represent synthetically generated objects 14a, 14b, 14c, 14d, 14e, and identifying S5 a second feature vector M2 having a greatest similarity measure MA to the first feature vector M1.

[0067] Furthermore, the method comprises selecting S6A the identified second feature vector M2 to generate a synthetic object 14a, 14b, 14c, 14d, 14e corresponding to the real object 12a, 12b, 12c, 12d, 12e.

[0068] Alternatively, a procedural generation S6B of the synthetic object 14a, 14b, 14c, 14d corresponding to the real object 12a, 12b, 12c, 12d takes place.

[0069] The selection of step S6A or S6B depends on the identified similarity measure MA. Subsequently, the synthetic object 14a, 14b, 14c, 14d, 14e is integrated into a predefined virtual vehicle environment U2 (step S7).

[0070] Fig. 2 shows a schematic diagram of the application of the method to a real vehicle environment for generating the virtual vehicle environment according to the preferred embodiment of the invention.

[0071] In the exemplary embodiment presented here, video image data are used in both the virtual vehicle environment U2 and the real vehicle environment U1.

[0072] Alternatively or additionally, radar data and / or a lidar point cloud of a real vehicle environment can be used to generate the virtual vehicle environment.

[0073] The pre-captured video image data 10a are first characterized by applying a first machine learning algorithm A1.

[0074] The labeling comprises detecting and classifying the plurality of real objects 12a, 12b, 12c, 12d, 12e by the first machine learning algorithm A1.

[0075] In the present embodiment, the first machine learning algorithm A1 is formed by a supervised learning algorithm, in particular an artificial neural network.

[0076] Alternatively, the first machine learning algorithm A1 can be formed, for example, by an unsupervised learning algorithm, in particular an artificial neural network. The first machine learning algorithm A1 classifies the real objects 12a, 12b, 12c, 12d contained in the video image data 10a into main classes C1, C2, C3, C4.

[0077] The real objects 12a, 12b, 12c, 12d assigned to a main class C1, C2, C3, C4 are further classified into subclasses 12d1, 12d2 by applying a fourth machine learning algorithm A4.

[0078] In the present embodiment, the object 12a is a tree, the object 12b is a pedestrian, the object 12c is a street lamp, the plurality of objects 12d are different vehicles and the object 12e is a traffic sign.

[0079] The objects mentioned above thus correspond to their respective main classes. Subclasses are defined by subtypes of the objects, such as different tree species and / or different types of pedestrians, selected by gender, age, and other classifiable characteristics.

[0080] The traffic sign 12e represents a special feature here, since in the present exemplary embodiment it is neither generated by assigning a second feature vector to a first feature vector nor is it generated procedurally, but is generated from other data sources or by other methods, since an exact, detailed reproduction of the traffic sign information is important here.

[0081] Alternatively, traffic signs can be generated, for example, by a synthetic object stored in a data memory if there is a sufficiently high match between a first feature vector and a corresponding second feature vector.

[0082] The real objects 12a, 12b, 12c, 12d contained therein are then selected from the real vehicle environment U1 and, by applying the second machine learning algorithm A2 to the respective real object 12a, 12b, 12c, 12d, a first feature vector M1 representing the respective real object 12a, 12b, 12c, 12d is generated.

[0083] This procedure is carried out for all of the objects 12a, 12b, 12c, 12d contained in the real vehicle environment U1.

[0084] Furthermore, for each of the first feature vectors M1 thus generated, which are stored in a corresponding data memory, a comparison with or an assignment to a second feature vector M2 is carried out.

[0085] For this purpose, a plurality of stored second feature vectors M2 are provided, which represent synthetically generated objects 14a, 14b, 14c, 14d. For this purpose, a second feature vector M2 having the greatest similarity measure MA to the first feature vector M1 is identified.

[0086] If the similarity measure MA of the identified second feature vector M2 to the first feature vector M1 is greater than or equal to a predetermined threshold value T, the identified second feature vector M2 is selected.

[0087] Subsequently or as a result, a stored synthetic object 14a, 14b, 14c, 14d, 14e assigned to the second feature vector M2, which corresponds to the real object 12a, 12b, 12c, 12d, 12e, is retrieved from a data memory.

[0088] Previously or in advance, the second feature vector M2 representing a respective synthetic object 14a, 14b, 14c, 14d, 14e was generated by applying a third machine learning algorithm A3 to the respective synthetic object 14a, 14b, 14c, 14d, 14e and stored in the data memory S3.

[0089] If the similarity measure MA of the identified second feature vector M2 to the first feature vector M1 is smaller than the predetermined threshold value T, the synthetic object 14a, 14b, 14c, 14d is procedurally generated.

[0090] The procedural generation can be performed, for example, by a software application suitable for this purpose, such as the Unreal Engine. The procedural generation of the synthetic object 14a, 14b, 14c, 14d is performed using the first feature vector M1 generated by the second machine learning algorithm A2.

[0091] The second machine learning algorithm A2 for generating, in particular for encoding, the first feature vector M1 is formed in the present embodiment by a first variational autoencoder.

[0092] The object 14e, which is formed by a traffic sign, is generated in the present embodiment by another suitable method from the traffic sign 12e contained in the real vehicle environment U1.

[0093] If the first feature vector M1 and the second feature vector M2 are similar, in particular if the data structure is similar, in order to identify the similarity measure MA, a direct comparison is carried out between the first feature vector M1 and the plurality of second feature vectors M2, as shown in the present embodiment.

[0094] In the case of dissimilarity, in particular a different data structure, of the first feature vector M1 and the second feature vector M2, a classifier establishing a correlation between the first feature vector M1 and the second feature vector M2 can alternatively be used to identify the similarity measure MA of the second feature vector M2 to the first feature vector M1.

[0095] In the present embodiment, the plurality of stored first feature vectors M1 and second feature vectors M2 are formed by 3-D image data.

[0096] Alternatively, the first feature vectors M1 and the second feature vectors M2 can be represented, for example, by 2-D image data, radar data and / or a lidar point cloud.

[0097] If the similarity measure MA of the identified second feature vector M2 to the first feature vector M1 is less than the predetermined threshold value T, a message ME is further generated according to the present embodiment that the stored plurality of second feature vectors M2 do not have a sufficient similarity measure MA to the identified first feature vector M1.

[0098] Fig. 3 shows a diagram of a system for generating the virtual vehicle environment for testing highly automated driving functions of the motor vehicle according to the preferred embodiment of the invention.

[0099] The system 2 comprises means 20 for receiving the pre-captured video image data 10a, radar data and / or the lidar point cloud of the real vehicle environment U1 including a plurality of real objects 12a, 12b, 12c, 12d, 12e characterized by applying at least one first machine learning algorithm A1.

[0100] Alternatively, the real objects 12a, 12b, 12c, 12d, 12e can, for example, be marked in advance.

[0101] The system 2 further comprises means 22 for applying at least a first machine learning algorithm A1 to the video image data 10a, the radar data and / or the lidar point cloud for detecting and classifying the plurality of real objects 12a, 12b, 12c, 12d, 12e.

[0102] The system 2 further comprises means 24 for generating a first feature vector M1 representing a respective real object 12a, 12b, 12c, 12d, 12e, which means 24 are configured to apply a second machine learning algorithm A2 to the respective real object 12a, 12b, 12c, 12d, 12e and means 26 for storing the first feature vector M1.

[0103] The system 2 further comprises means 28 for providing a plurality of stored second feature vectors M2, which represent synthetically generated objects 14a, 14b, 14c, 14d, 14e.

[0104] Furthermore, the system 2 comprises means 30 for identifying a second feature vector M2 having a greatest similarity measure MA to the first feature vector M1.

[0105] Furthermore, the system 2 comprises means 32 for selecting S6A the identified second feature vector M2 and calling a stored synthetic object 14a, 14b, 14c, 14d, 14e associated with the second feature vector M2, which corresponds to the real object 12a, 12b, 12c, 12d, 12e and means 34 for procedurally generating the synthetic object 14a, 14b, 14c, 14d, 14e corresponding to the real object 12a, 12b, 12c, 12d, 12e.

[0106] The means 32 for selecting S6A the identified second feature vector M2 and calling a stored synthetic object 14a, 14b, 14c, 14d, 14e associated with the second feature vector M2 or the means 34 for procedurally generating the synthetic object 14a, 14b, 14c, 14d, 14e corresponding to the real object 12a, 12b, 12c, 12d, 12e are used depending on the identified similarity measure MA.

[0107] Furthermore, the system 2 comprises means 36 for integrating the synthetic object 14a, 14b, 14c, 14d, 14e into a predetermined virtual vehicle environment U2. The means 30 for identifying a second feature vector M2 having a greatest similarity measure MA to the first feature vector M1 have a connection to both the means 26 for storing the first feature vector M1 and the means 28 for providing a plurality of stored second feature vectors M2 representing synthetically generated objects 14a, 14b, 14c, 14d, 14e.

[0108] Although specific embodiments have been illustrated and described herein, it will be understood by those skilled in the art that numerous alternative and / or equivalent implementations exist. It should be noted that the exemplary embodiment or exemplary embodiments are merely examples and are not intended to limit the scope, applicability, or configuration in any way.

[0109] Rather, the foregoing summary and detailed description will provide one skilled in the art with a convenient road map for implementing at least one exemplary embodiment, it being understood that various changes in functionality and arrangement of elements may be made without departing from the scope of the appended claims and their legal equivalents.

[0110] In general, this application is intended to cover modifications, adaptations, or variations of the embodiments presented herein.

Claims

1. A computer-implemented method for generating a virtual vehicle environment (U2) for testing highly automated driving functions of a motor vehicle (1) using pre-captured video image data (10a), radar data, and / or a lidar point cloud of a real vehicle environment (U1), comprising the steps of: receiving (S1A, S1B) the pre-captured video image data (10a), radar data, and / or the lidar point cloud of the real vehicle environment (U1) including a plurality of real objects (12a, 12b, 12c, 12d, 12e) identified in advance (S1A) or subsequently (S1B) by applying at least one first machine learning algorithm (A1); generating (S2) a first feature vector (M1) representing a respective real object (12a, 12b, 12c, 12d, 12e) by applying a second machine learning algorithm (A2) to the respective real object (12a, 12b, 12c, 12d, 12e) and storing (S3) the first feature vector (M1); providing (S4) a plurality of stored second feature vectors (M2) representing synthetically generated objects (14a, 14b, 14c, 14d, 14e); identifying (S5) a second feature vector (M2) having a greatest similarity measure (MA) to the first feature vector (M1); selecting (S6A) the identified second feature vector (M2) and calling a stored synthetic object (14a, 14b, 14c, 14d, 14e) associated with the second feature vector (M2) and corresponding to the real object (12a, 12b, 12c, 12d, 12e) or procedurally generating (S6B) the synthetic object (14a, 14b, 14c, 14d, 14e) corresponding to the real object (12a, 12b, 12c, 12d, 12e) as a function of the identified similarity measure (MA); and integrating (S7) the synthetic object (14a, 14b, 14c, 14d, 14e) into a specified virtual vehicle environment (U2), the virtual vehicle environment being a computer-generated representation of the real vehicle environment captured by sensors, wherein if the similarity measure (MA) of the identified second feature vector (M2) to the first feature vector (M1) is greater than or equal to a specified threshold value (T), then the identified second feature vector (M2) is selected and the stored synthetic object (14a, 14b, 14c, 14d, 14e) associated with the second feature vector (M2) and corresponding to the real object (12a, 12b, 12c, 12d, 12e) is called (S6A), and wherein if the similarity measure (MA) of the identified second feature vector (M2) to the first feature vector (M1) is less than a specified threshold value (T), then the synthetic object (14a, 14b, 14c, 14d, 14e) is procedurally generated (S6B).

2. The computer-implemented method according to claim 1, characterized in that the identifying (S1A, S1B), in particular detecting and classifying, of the plurality of real objects (12a, 12b, 12c, 12d, 12e) is carried out by a supervised learning algorithm or by an unsupervised learning algorithm, in particular an artificial neural network.

3. The computer-implemented method according to claim 2, characterized in that the procedural generation (S6B) of the synthetic object (14a, 14b, 14c, 14d, 14e) is carried out using the first feature vector (M1) generated by the second machine learning algorithm (A2).

4. The computer-implemented method according to any one of the preceding claims, characterized in that the second machine learning algorithm (A2) for generating, in particular for encoding, the first feature vector (M1) is formed by a first variational autoencoder.

5. The computer-implemented method according to any one of the preceding claims, characterized in that the second feature vector (M2) representing a respective synthetic object (14a, 14b, 14c, 14d, 14e) is generated by applying a third machine learning algorithm (A3) to the respective synthetic object (14a, 14b, 14c, 14d, 14e), and wherein the second feature vector (M2) is stored (S3) in a data memory.

6. The computer-implemented method according to any one of the preceding claims, characterized in that the first machine learning algorithm (A1) classifies the real objects (12a, 12b, 12c, 12d) present in the video image data (10a), the radar data, and / or the lidar point cloud into main classes (C1, C2, C3, C4), wherein the real objects (12a, 12b, 12c, 12d) associated with a main class (C1, C2, C3, C4) are classified into subclasses (12d1, 12d2) by applying a fourth machine learning algorithm (A4).

7. The computer-implemented method according to any one of the preceding claims, characterized in that if the first feature vector (M1) and the second feature vector (M2) are of the same type, in particular have the same data structure, then a direct comparison is carried out between the first feature vector (M1) and the plurality of second feature vectors (M2) in order to identify the similarity measure (MA).

8. The computer-implemented method according to any one of the preceding claims, characterized in that if the first feature vector (M1) and the second feature vector (M2) are of different types, in particular have a different type of data structure, then a classifier establishing a correlation between the first feature vector (M1) and the second feature vector (M2) is used to identify the similarity measure (MA) of the second feature vector (M2) to the first feature vector (M1).

9. The computer-implemented method according to any one of the preceding claims, characterized in that the plurality of stored first feature vectors (M1) and / or second feature vectors (M2) represent 2-D image data, 3-D image data, radar data, and / or a lidar point cloud.

10. The computer-implemented method according to any one of claims 3 to 9, characterized in that if the similarity measure (MA) of the identified second feature vector (M2) to the first feature vector (M1) is less than the specified threshold value (T), a message (ME) is generated that the stored plurality of second feature vectors (M2) do not have a sufficient similarity measure (MA) to the identified first feature vector (M1).

11. A system (2) for generating a virtual vehicle environment (U2) for testing highly automated driving functions of a motor vehicle using pre-captured video image data (10a), radar data, and / or a lidar point cloud of a real vehicle environment (U1), comprising: means (20) for receiving the pre-captured video image data (10a), radar data, and / or the lidar point cloud of the real vehicle environment (U1) including a plurality of real objects (12a, 12b, 12c, 12d, 12e) identified in advance or subsequently by applying at least one first machine learning algorithm (A1); means (22) for applying at least one first machine learning algorithm (A1) to the video image data (10a), the radar data, and / or the lidar point cloud for detecting and classifying the plurality of real objects (12a, 12b, 12c, 12d, 12e); means (24) for generating a first feature vector (M1) representing a respective real object (12a, 12b, 12c, 12d, 12e), said means (24) being configured to apply a second machine learning algorithm (A2) to the respective real object (12a, 12b, 12c, 12d, 12e), and means (26) for storing the first feature vector (M1); means (28) for providing a plurality of stored second feature vectors (M2) representing synthetically generated objects (14a, 14b, 14c, 14d, 14e); means (30) for identifying a second feature vector (M2) having a greatest similarity measure (MA) to the first feature vector (M1); means (32) for selecting (S6A) the identified second feature vector (M2) and for calling a stored synthetic object (14a, 14b, 14c, 14d, 14e) associated with the second feature vector (M2) and corresponding to the real object (12a, 12b, 12c, 12d, 12e), and means (34) for procedurally generating the synthetic object (14a, 14b, 14c, 14d, 14e) corresponding to the real object (12a, 12b, 12c, 12d, 12e), wherein the means (32) for selecting (S6A) the identified second feature vector (M2) and for calling a stored synthetic object (14a, 14b, 14c, 14d, 14e) associated with the second feature vector (M2) or the means (34) for procedurally generating the synthetic object (14a, 14b, 14c, 14d, 14e) corresponding to the real object (12a, 12b, 12c, 12d, 12e) are used as a function of the identified similarity measure (MA); and means (36) for integrating the synthetic object (14a, 14b, 14c, 14d, 14e) into a specified virtual vehicle environment (U2), wherein the virtual vehicle environment is a computer-generated representation of the real vehicle environment captured by sensors, wherein if the similarity measure (MA) of the identified second feature vector (M2) to the first feature vector (M1) is greater than or equal to a specified threshold value (T), then the identified second feature vector (M2) is selected and the stored synthetic object (14a, 14b, 14c, 14d, 14e) associated with the second feature vector (M2) and corresponding to the real object (12a, 12b, 12c, 12d, 12e) is called (S6A), and wherein if the similarity measure (MA) of the identified second feature vector (M2) to the first feature vector (M1) is less than a specified threshold value (T), then the synthetic object (14a, 14b, 14c, 14d, 14e) is procedurally generated (S6B).

12. A computer program having program code for performing the method according to any one of the claims 1 through 10 when the computer program is executed on a computer.

13. A computer-readable data storage medium having program code of a computer program for performing the method according to any one of the claims 1 through 10 when the computer program is executed on a computer.