Testing the vehicle's surrounding environment sensor system and / or surrounding environment perception

Abstract

A method (100) for testing an environment sensor system and / or environment perception of a vehicle (1) traveling on land or water, comprising the steps of: obtaining (110) model data (3) of at least a part of the environment of the vehicle (1), the model data (3) being generated with reference to data characterizing the environment of the vehicle (1) from at least one bird's-eye view; transforming the model data (3) into a reference coordinate system (1a) of the environment sensor system and / or environment perception; a step (120) of converting the transformed model data (3') into a corresponding model data (3') based on the transformed model data; a step (130) of comparing the transformed model data (3') with sensor data (2) provided by a surrounding environment sensor system or surrounding environment perception and / or with processing results (2a) of this sensor data; and a step (140) of evaluating from the result (130a) of this comparison (130) how well (4) the sensor data (2) and / or the processing results (2a) match the real situation in the surrounding environment of the vehicle (1).

Description

[Technical Field]

[0001] The present invention relates to testing a vehicle's surrounding environment sensor system and / or surrounding environment perception, in particular for driving assistance systems and at least partially automated driving systems. [Background technology]

[0002] Driving assistance systems and at least partially automated driving systems make decisions about interventions in the vehicle's driving dynamics based on surrounding environment perception, which is typically obtained by evaluating measurement data recorded by the vehicle's surrounding environment sensor systems. The surrounding environment perception can indicate, for example, which objects are contained in the vehicle's surrounding environment and which areas are freely passable.

[0003] The more accurate the surrounding environment perception, the better the system tends to be able to make decisions about the vehicle's subsequent driving dynamics. However, as the demand for accuracy increases, the costs of the surrounding environment sensor system and the subsequent connected evaluation increase disproportionately. Therefore, for the mass production start of vehicles, relatively inexpensive hardware for the surrounding environment sensor system and evaluation is usually used, and the obtained surrounding environment perception is likely checked, at least on a spot check basis, against a reference standard with known accuracy. In this regard, if the surrounding environment perception is consistently found to be plausible, it is assumed that this surrounding environment perception corresponds sufficiently accurately to the actual situation in the vehicle's surrounding environment. Summary of the Invention

[0004] Within the scope of the present invention, a method for testing an environment sensor system and / or environment perception of a vehicle traveling on land or water has been developed. An ambient environment sensor system is any assembly that includes one or more sensors and is configured to record sensor data related to one or more physical measurands from the vehicle's ambient environment. The sensor data represent values ​​of the measurands, and may, among other things, represent a spatial distribution of the measurand values. The sensor data may include, for example, images, video images, ultrasound images, thermal images, radar data, and / or lidar data. That is, the sensor data does not necessarily have to be present in a two-dimensional or three-dimensional mesh, but may also be present, for example, as a point cloud.

[0005] Surroundings awareness refers to any processing of sensor data from surroundings sensor systems that provides a semantic description of the situation in the vehicle's surroundings and therefore provides a better basis for making decisions about subsequent driving dynamics than the raw sensor data. Surroundings awareness may, for example, indicate which objects are located where in the vehicle's surroundings.

[0006] In this method, model data of at least a portion of the vehicle's surroundings is obtained. This model data is generated with reference to data characterizing the vehicle's surroundings from at least one bird's-eye view. As data, for example, an image showing the vehicle's surroundings from at least one bird's-eye view may be referenced. However, data from all other measurement methods, such as those mentioned above in connection with sensor data, may also be used. The model data may, for example, imply an evaluation of an image with respect to at least one view point that is also evaluated by the surroundings perception from sensor data.

[0007] The model data may imply, for example, a representation of the vehicle's environment in the form of one or more images and / or one or more point clouds, but the model data may also be the same or similar data types as implied by, for example, processing results provided by environment perception and semantic segmentation or other representations of objects in the vehicle's environment.

[0008] The bird's-eye view may be the view of any vehicle flying in the sky or space. Data from the bird's-eye view may be recorded, for example, by at least one drone, by at least one airplane, by at least one airship, or even by at least one satellite.

[0009] The model data is transformed into the reference frame of the surroundings sensor system and / or surroundings perception. If this model accurately corresponds to the real situation in the vehicle's surroundings, the transformed model data may, among other things, represent the sensor data and / or surroundings perception as they would ideally occur from the vehicle's point of view.

[0010] The transformed model data is compared with sensor data and / or processing results of the sensor data provided by the surrounding environment sensor system or surrounding environment perception, and the results of this comparison are used to evaluate how well the sensor data and / or processing results correspond to the real situation in the vehicle's surrounding environment.

[0011] In particular, using model data characterizing the vehicle's environment from at least one bird's-eye view to test an environment sensor system and / or environment perception operating from a completely different view may at first glance appear to be disadvantageous, since it requires a transformation into the reference frame of the additional environment sensor or environment perception. However, this seeming disadvantage is more than compensated for by two points.

[0012] For one thing, the bird's-eye view provides a more complete view of the vehicle's surroundings than can be achieved from the vehicle's own viewpoint. In most situations, the vehicle's surroundings are completely visible from one or more bird's-eye views. In particular, road users and other objects do not obscure one another. Once a situation is completely captured, it can be computationally transformed into many vehicle viewpoints. In contrast, a model created directly from the vehicle's viewpoint is only available for this viewpoint. While present, information about obscured objects from this viewpoint is not physically captured and therefore is not available even through more expensive transformations to other viewpoints. This means that once recorded, the model data can be used in a variety of ways.

[0013] Of course, bird's-eye view capture cannot be perfect in all situations. For example, tunnels cannot be seen from a bird's-eye view, and at highway interchanges, the roadways partially overlap in multiple levels. However, the model does not need to provide comparison data for testing the environment sensor system or the environment perception at every location where the vehicle is present at any time. Rather, it is sufficient to perform spot checks with the model data at specific time and / or spatial intervals. If, during this spot check, the sensor data and / or processing results determined in the vehicle match the converted model data, it can be assumed that the environment sensor system or the environment perception is working correctly even in situations that have not been specifically tested.

[0014] Secondly, model data from a bird's-eye view can be captured faster and cheaper than test drives. For example, aerial photography can capture the entire traffic event at a traffic node at once. To capture the same traffic event from each vehicle's perspective, many test drives would have to be performed in all possible directions and traffic relationships at this traffic node.

[0015] This method can be applied, among other things, during the validation of the environment sensor system and / or the environment perception of a vehicle under development. For example, a test drive of a test vehicle equipped with the environment sensor system or the environment perception to be tested can be observed by one or more drones. In this way, model data that is particularly relevant to the test drive and therefore best comparable to the sensor data or processing results can be obtained.

[0016] During normal driving operation of the vehicle, model data relevant to this particular trip is often unavailable, but at least one limited comparison with this model data is still possible: the vehicle should, among other things, perceive, for example, road markings, structural lane lines, buildings, and / or trees that change very infrequently or not at all, in the locations where the model predicts them.

[0017] Thus, in one particularly advantageous embodiment, the number of objects of the same type present at the same location is determined within the context of the comparison based on the transformed model data, on the one hand, and the surrounding environment recognition, on the other. Information about which objects are present at which locations is the most important basis for determining the vehicle's subsequent driving dynamics. Likelihood checks against the model data can identify a wide range of malfunctions in the surrounding environment sensor system or surrounding environment recognition. For example, if an object is recognized in the wrong location, this may indicate, for example, that the vehicle is not locating itself accurately enough or that the surrounding environment sensor system is not properly aligned or calibrated. Misclassification of an object or the complete absence of an object in the processing results provided by the surrounding environment recognition may indicate, for example, that the quality of the sensor data recorded by the surrounding environment sensor system is too poor or that the image classifier used in the surrounding environment sensor system is not up to date. Intentional manipulation using so-called "adversarial samples" can also be detected. In such manipulations, specific patterns are intentionally introduced into the vehicle's surrounding environment, which causes misclassification when processing sensor data in the surrounding environment recognition. In other words, in tests, for example, by attaching a semi-permeable film with a particular dot pattern over the camera lens, it has already been possible to disrupt the processing of recorded images in the Surrounding Area Perception in such a way that all pedestrians disappear within the semantic segmentation of the vehicle's surroundings provided by this Surrounding Area Perception.

[0018] In a further particularly advantageous embodiment, the accuracy of the temporal synchronization between the transformed model data on the one hand and the processing results resulting from the environmental perception on the other hand is determined. This determination may involve passive measurement of the temporal synchronization. However, it is particularly advantageous if the transformed model data on the one hand and the processing results resulting from the environmental perception on the other hand are actively synchronized with each other. For this purpose, for example, synchronization signals simultaneously recorded by the vehicle and by the drone used to create the model can be used. After active synchronization, the accuracy of the temporal synchronization can be determined based on the difference from zero (relative to perfect synchronization) or based on the residual error that remains after synchronization.

[0019] The comparison of which object is located where, based on the transformed model data on the one hand and the processing results from the surrounding environment perception on the other hand, is made to depend on the probability that the respective object will not change its position and / or orientation within a time span corresponding to this precision. In other words, the temporal precision established by passive measurements and / or active synchronization determines which type of object should be referred to in the subsequent spatial comparison between the transformed model data on the one hand and the sensor data or processing results on the other hand.

[0020] If the model data is related to the same time point as the sensor data or the processing results determined by the surrounding environment recognition from the sensor data, matching of the recognized objects and their positions is possible without any restrictions. If the model data on the one hand and the sensor data or the processing results on the other hand are not perfectly synchronized in time, the position of a particular object may change within a time span corresponding to this time difference. That is, in particular, for example, a vehicle or other road user may move further within this time span. In contrast, static objects will still remain in the same place. That is, for these objects, comparison is still meaningful even in the case of a time difference.

[0021] Thus, for example, road markings, lane lines, buildings, and / or trees, among others, which do not depend on the accuracy of the time synchronization, may be incorporated into the comparison. As explained above, the situation in the vehicle's surroundings tends to be more completely captured from a bird's-eye view than from the vehicle's perspective. Therefore, a comparison of the transformed model data with the sensor data or processing results determined by the vehicle's on-board computer may involve, among other things, examining which objects and to what extent can be captured by the vehicle's surroundings sensor system based on the transformed model data. If a particular object is not visible from the vehicle's perspective, the absence of this object in the sensor data or processing results of the surroundings sensor system or surroundings perception cannot be "blamed."

[0022] In a further advantageous embodiment, in response to determining that the sensor data and / or processing results are inconsistent with the actual situation in the vehicle's environment, at least one additional sensor and / or at least one additional ambient perception of the ambient environment sensor system is activated. Alternatively or in combination therewith, at least one driving assistance system or at least partially automated driving system may have its functional capabilities limited or deactivated.

[0023] For example, a camera-based ambient sensor system may be temporarily disrupted by precipitation or sunlight shining directly onto the image sensor. In this case, an additional sensor modality, such as radar, that is less vulnerable to such disruption may be utilized. The use of an alternative sensor modality may also negate the effects of tampering with the aforementioned "adversarial samples."

[0024] Such manipulation can also be countered, for example, by using additional environmental awareness, for example by using an additional neural network that is constructed and / or trained differently from the original environmental awareness neural network.

[0025] Alternatively or in combination, parameters characterizing the behavior of the surrounding environment sensor system and / or the surrounding environment perception can be optimized with the goal of better matching the sensor data and / or processing results with the real situation in the vehicle's surroundings. This is particularly advantageous when applying this method in the development process of the surrounding environment sensor system or the surrounding environment perception, where validation based on test drives observed from a bird's-eye view can provide feedback on which aspects of the surrounding environment sensor system and / or the surrounding environment perception still need to be optimized.

[0026] Parameters characterizing the behavior of an ambient sensor system can be, among other things, the operating parameters of a camera or other sensor, i.e., an optimal compromise between high resolution on the one hand and a high frame rate per second on the other hand, for example when the available bandwidth for the data stream is limited.

[0027] Parameters characterizing the behavior of the environment perception may include, among others, parameters (eg weights) of a neural network used in evaluating the sensor data, for example. The present invention also provides a method for creating a model of a vehicle's surroundings, the model being formed for use in the aforementioned surroundings sensor system and / or surroundings perception testing method.

[0028] In this method, data characterizing the vehicle's surroundings from the bird's-eye view of multiple drones is obtained. For example, an image showing the vehicle's surroundings from at least one bird's-eye view may be used as the data. However, data from any of the other measurement methods mentioned above in connection with the sensor data may also be used. With reference to this data and known information about the vehicle's appearance and / or geometry, the distance and / or orientation of at least one of the drones relative to the vehicle is determined. Based on this distance and / or orientation, the data and / or information derived from this data is integrated into a model. Alternatively or in combination, the distance between the vehicle and each drone may be determined with reference to further information, such as GPS signals, further electromagnetic signals, ultrasound, the distance of a predetermined immobile object, or triangulation results. For triangulation, multiple drones with different viewpoints relative to the vehicle may cooperate, for example.

[0029] Therefore, this method only requires information about the vehicle as known information. The drone's position is also required for integration. This position is typically acquired during flight. This method can be further supported by providing the vehicle with markings that have a particularly high contrast and are easily recognizable from the air.

[0030] Combining data, e.g., images, recorded by multiple drones can, on the one hand, improve the quality of the model. On the other hand, multiple drones can work together to track the same vehicle. Depending on the road category on which the vehicle is traveling, it may travel faster than the drone can fly. In this case, the vehicle can be "handed over" from the spatial coverage area of ​​one drone to that of another, somewhat like a moving mobile phone user being handed over from one radio cell to the next.

[0031] The determined information about the distance and / or orientation of each drone relative to the vehicle can be used not only to form a model, but also to transform this model into the reference frame of the vehicle's ambient sensor system or ambient perception.

[0032] In a particularly advantageous embodiment, the model implies a spatial distribution of at least one quantity of interest in the vehicle's environment, for example, for which a parameterization method can be performed, the parameters of which can be determined using the recorded data or information derived from this data.

[0033] Integrating data or information into a model can involve, inter alia, determining equations for quantities of interest from the data and solving a system of equations composed of these equations. Alternatively or in combination, these equations can also contain parameters of a parameterized distribution of the quantities as unknowns. In this case, the more information there is overall about the quantities sought at each location of interest, the more equations the system of equations contains. In this case, for example, conflicts between information provided by multiple drones can be automatically resolved so that the resulting errors are minimized. The density of available information can vary greatly over space. That is, there can be locations where a lot of information is available, but there can also be locations where very little information is available.

[0034] In a particularly advantageous embodiment, the model implies a semantic segmentation of the vehicle's surroundings, in the sense of which locations are occupied by which types of objects, as a spatial distribution of a quantity of interest, i.e., the occupancy by a certain type of object is the quantity of interest, and the spatial distribution of this quantity of interest is determined. As explained above, the occupancy of locations by a certain type of object is the most important basis for decisions about interventions in the vehicle's driving dynamics.

[0035] In a further advantageous embodiment, the integration into the model implies the reconstruction of the geometry of at least one object in the vehicle's environment by photogrammetry from the images as data. In this way, the location occupied by the object can be determined not only in two dimensions but also in three dimensions, for example, so that the height of the object can also be determined. The magnitude of the object's shadow cast can also be used, for example, to determine the height of the object.

[0036] In a further advantageous embodiment, the integration into the model implies that data recorded by different drones at different times and / or information derived from these data are correlated with each other based on these times, for example, if one and the same vehicle first travels through the capture range of a first drone and then travels through the capture range of a second drone, the data from both observations of this vehicle by the respective drones or the information derived from this data can be computed with each other.

[0037] This method may, inter alia, be implemented entirely or partly on a computer. The invention therefore also relates to a computer program having machine-readable instructions which, when executed on one or more computers, cause the one or more computers to carry out one of the aforementioned methods. In this sense, vehicle control devices and embedded systems for technical equipment, which are also capable of executing machine-readable instructions, may also be considered computers.

[0038] The invention also relates to a machine-readable data storage medium and / or a download product comprising a computer program, which is a digital product that can be transmitted over a data network, i.e. downloaded by a user of the data network, and that can be offered for immediate download, for example in an online shop.

[0039] Additionally, the computer may have a computer program, a machine-readable data storage medium, or a downloadable product. Further measures for improving the invention are detailed below on the basis of the figures together with a description of preferred exemplary embodiments of the invention. [Brief explanation of the drawings]

[0040] [Figure 1] 1 illustrates an exemplary embodiment of a method 100 for testing the surrounding environment sensor system and / or surrounding environment awareness of a vehicle 1. FIG. [Figure 2] FIG. 2 illustrates an exemplary embodiment of a method 200 for creating a model 3. [Figure 3] FIG. 1 shows a traffic situation 10 as an example of application of the method 100. DETAILED DESCRIPTION OF THE INVENTION

[0041] FIG. 1 is a schematic flow diagram of one exemplary embodiment of a method 100 for testing an environment sensor system and / or environment awareness of a vehicle 1 . In step 110, model data 3 of at least a part of the environment surrounding the vehicle 1 is obtained. This model data 3 has been created with reference to data characterizing the environment surrounding the vehicle 1 from at least one bird's-eye view.

[0042] In step 120, the model data 3 is transformed into the reference frame 1a of the vehicle's 1 environment sensor system and / or environment perception. In step 130, the transformed model data 3' is compared with the sensor data 2 provided by the ambient sensor system or ambient perception and / or the processing results 2a of this sensor data.

[0043] From the result 130a of the comparison 130, it is evaluated in step 140 how well the sensor data 2 and / or the processed results 2a match the real situation in the surrounding environment of the vehicle 1. The degree of match is represented by the reference numeral 4.

[0044] In step 150, a binary determination is made as to whether the sensor data 2 and / or the processing results 2a are inconsistent with the actual situation in the vehicle's environment based on the degree 4 and any criteria. If they are inconsistent (boolean value 0), in step 160, at least one additional sensor of the ambient environment sensor system and / or at least one additional ambient environment perception may be activated. Alternatively or in combination, in step 170, at least one driving assistance system or at least a partially automated driving system may have its functional capabilities limited or deactivated. Furthermore, in step 180, alternatively or in combination, parameters characterizing the behavior of the ambient environment sensor system and / or ambient environment perception may be optimized with the goal of better matching the sensor data 2 and / or the processing results 2a with the actual situation in the vehicle's environment.

[0045] Within the framework of the comparison 130, it can be determined in block 131 how many objects of the same type are present at the same location, on the one hand based on the transformed model data (3') and on the other hand based on the surrounding environment perception.

[0046] In this regard, according to block 131a, the accuracy of the temporal synchronization between the transformed model data 3' on the one hand and the processing result 2a resulting from the surrounding environment perception on the other hand can be determined. According to block 131b, the comparison can be made dependent on the probability that each object will not change its position and / or orientation within a time span corresponding to this accuracy. According to block 131c, road markings, lane lines, buildings, and / or trees that are not dependent on the accuracy of the temporal synchronization can be incorporated into the comparison 130. As explained above, the accuracy of the temporal synchronization can be passively measured, but can also be actively generated.

[0047] According to block 132, the comparison 130 may involve checking which objects and in what proportions based on the transformed model data 3' can be captured by the vehicle's 1 surroundings sensor system.

[0048] 2 is a schematic flow diagram of one exemplary embodiment of a method 200 for creating a model 3 of the environment of a vehicle 1. This model 3 is prepared for use in the method 100 described above.

[0049] In step 210, data 5a-5c, eg, images, characterizing the environment surrounding the vehicle 1 from the bird's-eye view of a plurality of drones 6a-6c are obtained. In step 220, based on these data 5a-5c and known information 1* about the appearance and / or geometry of the vehicle 1, at least one distance 7a-7c and / or orientation 8a-8c of each drone 6a-6c relative to the vehicle 1 is determined.

[0050] In step 230, the data 5a-5c and / or information derived from the data 5a-5c are integrated into the model 3 based on these distances 7a-7c and / or orientations 8a-8c.

[0051] Based on block 231, the model 3 may imply a spatial distribution of at least one quantity of interest within the surrounding environment of the vehicle 1. According to block 231a, an equation for the quantity of interest can be determined from the data 5a to 5c, which can then be solved according to block 231b.

[0052] Based on block 231c, model 3 may imply a semantic segmentation of the surrounding environment of vehicle 1, in the sense of which locations are occupied by which types of objects, as a spatial distribution of quantities of interest.

[0053] According to block 232, integration 230 into the model 3 may imply reconstructing the geometry of at least one object in the vehicle's environment by photogrammetry from the images as data 5a-5c.

[0054] Based on block 233, integrating 230 into the model 3 may imply correlating data 5a-5c recorded by different drones 6a-6c at different times and / or information derived from these data 5a-5c with each other based on these times.

[0055] 3 illustrates an exemplary traffic situation 10 that may be observed by three drones 6a-6c to create a model 3 according to method 200. The traffic situation 10 may include: - Roadway 14 with crosswalk 14a, a vehicle 1 traveling on a roadway 14, the surrounding environment sensor system and / or surrounding environment perception of which will later be inspected by a model 3 to be created; two further vehicles 11 and 12 travelling on carriageway 14; A pedestrian 13 crossing a crosswalk 14a on a roadway 14, - Pillar 15 at the edge of roadway 14 and Includes:

[0056] 3a shows a side view, and FIG. 3b shows a top view. This comparison shows that the traffic situation 10 can be seen much better from a top view than from a side view, and the traffic situation can be captured much more clearly. For example, the pedestrian crossing 14a is not visible in the side view, and it is difficult to distinguish which direction the vehicles 1, 11, and 12 are traveling on the roadway 14.

[0057] Figure 3c shows the traffic situation 10 from the point of view of the driver of vehicle 11. Here, the limited field of view from this point of view already provides clearly less information: the left edge of the pedestrian crossing 14a and the pillar 15 are missing, so the driver does not see, for example, whether another person is entering the crosswalk 14a from the left. The vehicle 12 in front is also missing. Only the front of the oncoming vehicle 11 is visible, so it is difficult to judge, for example, its length.

Claims

1. A method (100) for testing an ambient environment sensor system and / or ambient environment awareness of a land or water vehicle (1), comprising: a step (110) in which model data (3) of at least a part of the surrounding environment of the vehicle (1) is obtained, the model data (3) being created with reference to data characterizing the surrounding environment of the vehicle (1) from at least one bird's-eye view; a step (120) in which the model data (3) is transformed into the reference coordinate system (1a) of the environment sensor system and / or environment perception; a step (130) in which the transformed model data (3') is compared with the sensor data (2) provided by the ambient environment sensor system or ambient environment perception and / or with the results of processing the sensor data (2a), a step (140) in which, from the result (130a) of the comparison (130), it is evaluated how well (4) the sensor data (2) and / or the processed result (2a) correspond to the actual situation in the surrounding environment of the vehicle (1), The method (100) includes creating the model data (3) using data (5a-5c) characterizing the surrounding environment of the vehicle (1) from the bird's-eye view of each drone recorded at different times by different first and second drones (6a-6c), and / or information derived from the data (5a-5c), wherein the data (5a-5c) and / or information derived from the data (5a-5c) are data characterizing the surrounding environment of the vehicle (1) from the bird's-eye view of the first drone and / or information derived from the data (5a-5c) when the vehicle (1) moves from the spatial capture range of the first drone to the spatial capture range of the second drone.

2. The method (100) according to claim 1, wherein within the framework of the comparison (130), it is determined (131) how many objects of the same type are present at the same location based on the transformed model data (3') on the one hand and on the surrounding environment perception on the other hand.

3. The accuracy of the temporal synchronization between the transformed model data (3') on the one hand and the processing results (2a) resulting from the environmental perception on the other hand is determined (131a), and The method (100) of claim 2, wherein the comparison (130) is made dependent on the probability (131b) that each of the objects will not change its position and / or orientation within a time span corresponding to the accuracy.

4. 4. The method (100) of claim 3, wherein road markings, lane lines, buildings, and / or trees that are not dependent on the accuracy of the time synchronization are incorporated (131c) into the comparison (130).

5. The method (100) of claim 1, wherein the comparison (130) implies checking (132) which objects and in what proportions can be captured by the surrounding environment sensor system of the vehicle (1) based on the transformed model data (3').

6. In response to determining (150) that the sensor data (2) and / or the processing results (2a) are inconsistent with a real-world situation in the vehicle's (1) surrounding environment, At least one additional sensor and / or at least one additional environmental perception of the environmental sensor system is activated (160); At least one driving assistance system or at least one partially automated driving system has its functional capabilities limited or deactivated (170); and / or The method (100) of claim 1, wherein parameters characterizing the behavior of the surrounding environment sensor system and / or surrounding environment perception are optimized (180) with the goal of better matching the sensor data (2) and / or processing results (2a) with the actual situation in the surrounding environment of the vehicle (1).

7. 2. A method (200) for creating a model (3) of the environment of a vehicle (1) for use in an environment sensor system and / or a method (100) for testing environment perception according to claim 1, comprising: - a step (210) in which data (5a-5c) characterizing the surrounding environment of the vehicle (1) from the bird's-eye view of the first and second drones (6a-6c) are obtained; a step (220) in which the distance (7a-7c) and / or the orientation (8a-8c) of at least one of the first and second drones (6a-6c) relative to the vehicle (1) is determined with reference to the data (5a-5c) and known information (1*) about the appearance and / or geometry of the vehicle (1); A method (200) for creating said model (3), comprising a step (230) in which said data (5a-5c) and / or information derived from said data (5a-5c) are integrated into said model (3) based on said distances (7a-7c) and / or orientations (8a-8c).

8. 8. The method (200) of claim 7, wherein the model (3) implies (231) a spatial distribution of at least one quantity of interest in the surrounding environment of the vehicle (1).

9. 9. The method (200) of claim 8, wherein said integration (230) into said model implies determining (231a) an equation for said quantity of interest from said data (5a-5c) and solving (231b) a system of equations constructed from said equations.

10. 9. The method (200) of claim 8, wherein the model (3) implies a semantic segmentation (231c) of the surrounding environment of the vehicle (1) in terms of which locations are occupied by which types of objects as a spatial distribution of quantities of interest.

11. 8. The method (200) of claim 7, wherein the integration (230) into the model (3) implies reconstructing (232) the geometry of at least one object in the vehicle's environment by photogrammetry from images as data (5a-5c).

12. 8. The method (200) of claim 7, wherein the integration (230) into the model (3) implies correlating (233) the data (5a-5c) recorded by different first and second drones (6a-6c) at different times and / or information derived from the data (5a-5c) with each other based on the time points.

13. A computer program comprising machine-readable instructions that, when executed on one or more computers, cause said one or more computers to perform the method (100, 200) of any one of claims 1 to 12.

14. 14. A machine-readable data storage medium having a computer program according to claim 13.

15. One or more computers having the computer program of claim 13.

16. 15. One or more computers having the machine-readable data storage medium of claim 14.

Images