Detection of a trailer by means of an artificial neural network

Abstract

The invention relates to a method and a system for detecting a trailer (40) which is towed by a towing vehicle (38), with the aid of a radar system (10, 12) of the towing vehicle (38), comprising: obtaining radar data (20) captured by the radar system from the radar system (10, 12), and inputting the radar data (20) into an artificial neural network (18), wherein the artificial neural network (18) receives and processes the inputted radar data (20) in order to generate therefrom at least one output signal (26) which indicates the detection of the presence of a trailer (40) of the towing vehicle (38) which is towed by the towing vehicle (38) and / or indicates trailer data which characterize a trailer (40) of the towing vehicle (38) which is towed by the towing vehicle (38). The invention also relates to a method for training the artificial neural network (18), comprising: capturing (56) radar data (20) with the aid of a radar system (10, 12) of a towing vehicle (38) during at least one training journey, and training (72) the artificial neural network (18) on the basis of at least the radar data (20) and reference data assigned to the radar data.

Description

[0001] Description

[0002] title

[0003] Detection of a trailer using an artificial neural network

[0004] The invention relates to a method for detecting a trailer of a towing vehicle by means of a radar system of the towing vehicle.

[0005] State of the art

[0006] Motor vehicles are often equipped with a radar system that provides location data about the traffic environment for a driver assistance system, for example for cruise control and distance control, blind spot monitoring, lane change assist, emergency braking assist or, particularly in the case of a truck, trailer maneuvering assist. If the vehicle is towing a trailer, this requires certain adjustments to the functionality of the driver assistance system. For example, a speed limit must be observed when driving with trailers. Blind spot monitoring can result in false warnings if the radar system interprets the trailer as another vehicle in the blind spot (e.g. when cornering). With a lane change assist, the warning zone must be adjusted to the greater length of the towing vehicle and trailer combination.

[0007] US 2020 / 0282910 A1 describes a method for training an image-based system for detecting a trailer angle using images of trailer drawbars at different trailer angles.

[0008] In one example, the system includes a neural network.

[0009] Disclosure of the invention

[0010] The object of the invention is to provide a novel method for detecting the presence of a trailer and / or trailer data that characterize the trailer in more detail.

[0011] This object is achieved according to the invention by a method for detecting a trailer of the towing vehicle pulled by a towing vehicle using a radar system of the towing vehicle, comprising the following steps:

[0012] Obtaining radar data acquired by the radar system from the radar system; and

[0013] Inputting the radar data into an artificial neural network, wherein the artificial neural network receives and processes the input radar data to generate therefrom at least one output signal indicating the detection of a presence of a trailer of the towing vehicle pulled by the towing vehicle and / or trailer data characterizing a trailer of the towing vehicle pulled by the towing vehicle.

[0014] Thus, with the help of the neural network, the presence of a trailer or radar data that further characterizes the trailer can be extracted directly from the radar data. This particularly simplifies the training of the neural network. An end-to-end learning method can be used for training. This can reduce the effort required to adapt the method to, for example, new sensors in a radar system to creating a new training dataset. Direct access to the radar data can also enable robust detection or characterization of the trailer.

[0015] The radar system can, in particular, be an environmental radar system of the towing vehicle. In the step of obtaining radar data, current radar data acquired by the radar system is obtained from the radar system. In particular, the radar data acquired by the radar system can be obtained from the radar system while the towing vehicle is traveling and / or input into the artificial neural network during the travel.

[0016] The method may comprise: receiving the at least one output signal from the artificial neural network.

[0017] Generally, a trailer has a lighting system that must be connected to the vehicle's electrical system by plugging in a connector. When this electrical connection is established, a detection signal can be generated simultaneously, indicating the presence of a trailer. If the artificial neural network generates at least one output signal indicating the detection of a trailer, the presence of a trailer can be detected in a way other than by detecting the electrical connection, thus achieving a higher degree of redundancy. This is useful, for example, if the electrical coupling is defective or if the lighting system has been forgotten to be connected.

[0018] If the artificial neural network generates at least one output signal indicating trailer data characterizing a trailer of the towing vehicle pulled by the towing vehicle, this can serve to additionally verify the presence of the trailer. Further advantageous embodiments of the invention are specified below and in the subclaims. In embodiments, the radar data input to the artificial neural network comprise point detections of reflection points. The point detections can each comprise a radial distance and / or a radial velocity. The point detections can each comprise at least one parameter from a radial distance, a radial velocity, an azimuth angle, an elevation angle, and a radar cross section (RCS). The point detections can be transmitted as a point cloud or input to the artificial neural network.The input vector of the neural network can be padded to a maximum number of points (point detections).

[0019] In embodiments, the radar data input to the artificial neural network comprises at least one spectrum comprising a definition range corresponding to a distance range. The spectrum may further comprise a definition range corresponding to a Doppler range or speed range. The values ​​of the spectrum may, in particular, comprise amplitudes, intensities, and / or phases of the received radar signals. For example, the radar data input to the artificial neural network may comprise at least one distance-Doppler spectrum. For example, multiple spectra from respective channels of the radar system may be input. The spectrum may also comprise a definition range corresponding to an angular range.

[0020] The radar data contains detections of the towing vehicle's surroundings. The radar data can be provided through prior signal processing. For example, radar signals received by the radar system can be sampled and converted into digital radar data.

[0021] In embodiments, the artificial neural network comprises a feature extension

[0022] 5 reaction network into which the radar data is input; at least one object detection head connected to the feature extraction network for outputting object data that characterizes objects in the vicinity of the towing vehicle detected on the basis of the radar data; and a trailer detection head for outputting the at least one output signal, i.e., the at least one output signal that indicates the detection of a presence of a trailer of the towing vehicle towed by the towing vehicle and / or indicates trailer data that characterizes a trailer of the towing vehicle towed by the towing vehicle. In particular, the object data can be generated by the artificial neural network from processing the input radar data. In other words, the artificial neural network receives the input radar data and processes it in order to also generate the object data therefrom. The artificial neural network thus generates the at least one output signal and the object data.The trailer detection head can be another object detection head. The artificial neural network can thus comprise a plurality of object detection heads connected to the feature extraction network for outputting object data that characterize objects detected in the vicinity of the towing vehicle based on the radar data, wherein one of the object detection heads is a trailer detection head for outputting the at least one output signal. ^.

[0023] This makes it possible, when using an artificial neural network for object detection, to also use or "reuse" the feature extraction network (in particular, an encoder and a backbone of the neural network) for trailer detection. This is a further advantage for the end-to-end learning approach when training the artificial neural network. By jointly using the feature extraction network for the at least one object detection head and for the trailer detection head, trailer characterization, for example, can be performed as an additional task alongside object detection in a separate head.

[0024] In embodiments, the trailer data comprise at least one of an articulation angle, a number of axles, a wheelbase, a trailer type classification, a length, a width, a height, and a coupling position (position of a coupling point). The trailer data can thus comprise dynamic trailer data, i.e., data that changes during the journey, such as an articulation angle of the trailer and, if a lift axle is present, a number of axles and / or a wheelbase of the trailer and, if applicable, a trailer type classification. However, the trailer data can also comprise static trailer data, i.e., data that does not change during the journey, such as a number of axles of the trailer, a wheelbase of the trailer, a trailer type classification, a length of the trailer, a width of the trailer, a height of the trailer, and a coupling position, in particular a coupling position on the towing vehicle.In this sense, the output signal indicating the detection of a trailer presence represents static or dynamic trailer data. The output signal, i.e. the output of the trailer characterization, can thus include information as a static or dynamic variable, indicating whether a trailer is currently attached to the towing vehicle. One or more articulation angles (trailer articulation angles) can be output. As mentioned, the number of axles and thus the wheelbase can change during operation if a lift axle is present. The coupling point can vary greatly from model to model in commercial vehicles and, less frequently, in passenger cars. In general, this is required, alongside the other variables, to more precisely describe the position of the trailer relative to the towing vehicle. A trailer type can, for example, be understood as a characterization based on the number of axles and / or the number of articulation points per trailer.A trailer type, for example, can be classified from trailer types. These trailer types can include: center-axle trailers, turntable trailers, and possibly also dollies. A dolly is a very short trailer with a fifth wheel coupling.

[0025] In embodiments, the method further comprises: receiving movement data (in particular vehicle movement data) recorded by the towing vehicle's own sensor system from the towing vehicle's own sensor system; and also inputting the movement data into the artificial neural network, wherein the artificial neural network receives and processes the input radar data and the input movement data in order to generate the at least one output signal therefrom. Since the trailer is connected to the towing vehicle via a coupling and thus the movement of the towing vehicle directly influences the movement of the trailer, current movement data also contains information regarding the trailer movement. The movement data can in particular comprise one or more of the following data: the current speed, the yaw rate, the steering angle. In particular, the steering angle can comprise a steering angle on a front axle and a steering angle on a rear axle of the towing vehicle.

[0026] The object is further achieved by a system for a towing vehicle for detecting a trailer of the towing vehicle pulled by the towing vehicle, the system comprising a radar system and an artificial neural network, the system comprising an evaluation and control unit configured to input radar data acquired by the radar system into the artificial neural network, the artificial neural network configured to receive and process the input radar data in order to generate at least one output signal therefrom which indicates the detection of a presence of a trailer of the towing vehicle pulled by the towing vehicle and / or indicates trailer data which characterizes a trailer of the towing vehicle pulled by the towing vehicle. The evaluation and control unit can in particular be configured to carry out the described method.

[0027] The artificial neural network of the described method or system may have been trained according to the method described below.

[0028] According to a further aspect of the invention, a method is provided for training an artificial neural network for use in the method described above or for use as the artificial neural network of the described system. The method for training the artificial neural network comprises the following steps:

[0029] Acquiring radar data using a radar system of a towing vehicle (and, if necessary, also acquiring vehicle movement data of the towing vehicle) during at least one training run with the towing vehicle with a trailer of the towing vehicle pulled by the towing vehicle; assigning reference data, wherein the reference data corresponds to the radar data

[0030] (and, where applicable, also the vehicle movement data), wherein the reference data indicates the presence of a trailer of the towing vehicle pulled by the towing vehicle and / or indicates reference trailer data characterising the trailer of the towing vehicle pulled by the towing vehicle; and

[0031] Training the artificial neural network based at least on the radar data and the reference data assigned to it.

[0032] When training the artificial neural network, the radar data can be used as

[0033] The inputs of the artificial neural network can be used, and the output signal can be expected or received as the output of the artificial neural network. The output signal can be compared with the reference data. The artificial neural network can be adjusted based on the comparison. The artificial neural network can thus be trained using an end-to-end learning method.

[0034] The method for training the artificial neural network may include one or more of the features or steps described below.

[0035] In particular, the method may further comprise: acquiring vehicle movement data of the towing vehicle during the at least one training run, wherein (in the step of assigning reference data) the reference data is assigned to the radar data and the vehicle movement data. The vehicle movement data may, in particular, include vehicle movement data.

[0036] The method (the training method) can further comprise: acquiring further radar data using the radar system of the towing vehicle (and optionally also acquiring further vehicle movement data of the towing vehicle) during at least one training run with the towing vehicle without a trailer towed by the towing vehicle; and assigning reference data which are assigned to the further radar data (and optionally also to the further vehicle movement data), wherein the reference data indicate that no trailer of the towing vehicle towed by the towing vehicle is present. The radar data and the reference data with which the neural network is trained can thus comprise situations with a trailer and situations without a trailer. The described method for detecting a trailer of the towing vehicle towed by a towing vehicle can in particular comprise the method for training the artificial neural network.The method may include entering or recording fixed parameters of the trailer in question. The fixed parameters may, in particular, be part of the reference data.

[0037] The method may include recording the radar data (and, if applicable, the vehicle movement data) during the respective training run. In particular, the actual training of the artificial neural network may take place after the completion of one or more training runs.

[0038] The method may include: preprocessing the radar data. Preprocessing may include filtering the radar data by distance, speed, and / or solid angle. Such filtering simplifies processing because these variables are easy to narrow down for the particular trailer used. The method may include: varying the acquired radar data (and any associated vehicle movement data), in particular by augmentation. The augmentation or variation may, for example, include removing and / or adding individual detections (e.g., point detections), rotating all or individual detections (e.g., point detections), and / or mirroring radar data or detections.

[0039] The reference data can, for example, include a reference angle (reference articulation angle) and / or a reference number of axles. The reference data can, for example, also include the above-mentioned trailer data that characterizes the trailer of the towing vehicle pulled by the towing vehicle. The trailer data can, for example, include at least one of a trailer articulation angle, a trailer type, a trailer width, a trailer length, a trailer height, a trailer wheelbase, the number and / or positions of trailer axles, and a trailer coupling point at which a coupling of the trailer is coupled to the towing vehicle.

[0040] The method can comprise: determining the reference data by evaluating the radar data. The evaluation of the radar data can, for example, comprise a principal component analysis (PCR) in order to, for example, obtain an orientation of a point cloud (set of point detections) as the trailer orientation. Furthermore, the evaluation of the radar data can comprise: evaluating the radar data for the detection of micro-Doppler detections in the area of ​​the axles or wheels of the trailer in order to determine the position and / or number of rotating axles of the trailer. The evaluation of the radar data can comprise: evaluating a distribution of detected speeds on parts of the trailer (in particular rigid parts of the trailer, for example a trailer body), in order to, for example, determine a mass, a size and / or other parameters of the trailer.

[0041] The method may include: obtaining reference data from a steering angle sensor, for example a steering angle sensor of the towing vehicle or the trailer. ^

[0042] The method may include: Obtaining reference data via a bus system of the towing vehicle. For example, reference data can be supplemented or enhanced by evaluating bus data, particularly trailer bus data. In the commercial vehicle sector, a vehicle bus, via which bus data 19 such as trailer bus data can be obtained, is often available as a standardized protocol.

[0043] The method may include checking the acquired radar data and the respective assigned reference data for sufficient completeness with respect to the trailer data values ​​and / or vehicle situations to be trained. For example, the existing data set (radar data and reference data) can be checked for completeness with respect to dynamic variables.

[0044] To achieve sufficient system performance of the detection method, it is desirable that both a validation dataset and the training dataset contain a minimum number of measurement-label pairs (pairs of radar data and associated reference data). These pairs should cover a wide range of possible states.

[0045] The method may include: starting to train the artificial neural network if the result of the sufficient completeness check is positive.

[0046] The method may include generating additional radar data and associated reference data, particularly through augmentation, particularly as described above, if the result of the test is that sufficient completeness is not yet achieved. In this way, if desirable data is not yet included in the training set, it can be additionally generated through the variation / augmentation methods and the assignment of appropriate reference data.

[0047] The method may include: obtaining movement data recorded by the towing vehicle's own sensor system from the towing vehicle's own sensor system, wherein the artificial neural network is trained based at least on the radar data, the reference data assigned to it, and associated movement data. The movement data may be used as input to the artificial neural network and / or as part of the reference data.

[0048] The method may include: acquiring vehicle movement data of the towing vehicle during the at least one training run, wherein the reference data is assigned to the radar data and the vehicle movement data. Acquiring vehicle movement data may include: receiving vehicle movement data acquired by the towing vehicle's own sensor system from the towing vehicle's own sensor system.

[0049] The described detection method can also be used to link (in particular to fuse) the evaluations obtained from the artificial neural network (the at least one output signal and, if applicable, the object data) with detection data from further environmental sensors of the towing vehicle, for example with environmental sensors that comprise at least one of a further radar sensor, a lidar sensor and a video-based sensor.

[0050] The following examples are explained in more detail using the drawings. They show:

[0051] Fig. 1 is a block diagram of a trailer detection system according to the invention;

[0052] Fig. 2 is a block diagram of an artificial neural network of the system; Fig. 3 is a sketch of a towing vehicle and a trailer in a top view;

[0053] Fig. 4 Examples of radar data in the form of a point cloud or in the form of a distance Doppler spectrum; and

[0054] Fig. 5 is a flowchart of a method for training the artificial neural network.

[0055] The system shown in Fig. 1 for detecting a trailer pulled by a towing vehicle has a radar system with two radar sensors 10, 12 and an electronic evaluation and control unit 14, which includes a trailer characterization module 16. The radar sensors 10, 12 are installed at the left and right rear corners of a towing vehicle (motor vehicle) and each monitor an area behind and to the left or right of the vehicle. If necessary, the detection ranges of the two radar sensors can overlap so that the area immediately behind the vehicle can also be detected. The system can include additional radar sensors (not shown), in particular front sensors, as well as additional environmental sensors, in particular a lidar sensor and a camera-based

[0056] Sensor.

[0057] The trailer characterization module 16 contains an artificial neural network 18, which is a feedforward neural network. The neural

[0058] Network 18 therefore has no feedback connections. Alternatively, the neural network 18 can be a recursive neural network or a convolutional neural network. The evaluation and control unit 14 receives radar data 20 acquired by the radar sensors 10, 12 of the radar system. The radar data 20 includes detections of radar reflections in the vehicle's environment. The radar data can be transmitted in the form of a point cloud, the points of which are characterized, for example, by a radial distance d, radial velocity v, azimuth and elevation angles, and an estimate of the radar cross-section. Alternatively, the radar data can also be transmitted in the form of, for example, a distance Doppler spectrum.

[0059] Furthermore, the evaluation and control unit 14 receives movement data 22 of the towing vehicle from a sensor system 24 of the towing vehicle.

[0060] The evaluation and control unit 14 inputs the received radar data 20 and movement data 22 as an input vector into the artificial neural network 18. This network processes the specified data and generates an output signal 26 from it. The output signal 26 is temporally smoothed (in the time domain) by a filter 27 to compensate for fluctuations. The filtered output signal 26' is then output by the trailer characterization module 16 as the result of trailer detection and trailer characterization. The output signal 26 or 26' includes, in particular, a trailer presence signal and trailer data that further characterize the detected trailer of the towing vehicle pulled by the towing vehicle.The trailer data includes, for example, a trailer's articulation angle, a trailer type, a trailer width, a trailer length, a trailer height, a trailer wheelbase, the number and positions of trailer axles, and a trailer coupling point at which a trailer coupling is coupled to the towing vehicle. The trailer articulation angle is the current angle of the trailer drawbar relative to the coupling point. The estimated trailer data and the trailer presence signal of the output signal 26' can be made available, for example, to driver assistance functions 15 implemented in the evaluation and control unit 14.

[0061] Fig. 2 schematically shows the structure of an embodiment of a neural network 28 implemented in the evaluation and control unit, which includes the neural network 18 of the trailer characterization module 16. The neural network 28 is used for object detection and includes a feature extraction network 30, also referred to as the backbone, as well as several object detection heads connected to an output of the feature extraction network 30, which include a trailer detection head 32 and further object detection heads 34 for detecting other objects. The feature extraction network 30 receives the radar data 20 and the movement data 22 as input. The trailer detection head 32 outputs the output signal 26, which includes the trailer presence signal and the trailer data. The further object detection heads 34 output object data 36 of detected objects. The object data characterizes 20 detected objects in the vicinity of the search vehicle based on the radar data.

[0062] Fig. 3 shows a schematic plan view of a towing vehicle 38 and a trailer 40. The radar sensors 10, 12 are schematically shown at the rear corners of the towing vehicle 38. Schematically shown are reflections from reflection points 42 at corners of the body of the trailer 40, detected by the radar sensor 12 as radar reflections. Additional reflection points 44 can arise and be detected, for example, on wheel arches 46 of the trailer 40. Furthermore, reflection points 48 can be detected with micro-Doppler effects due to rotating wheels. In addition to the speed and yaw rate (or yaw velocity) of the towing vehicle 38, the intrinsic sensor system 24 can also detect an articulation angle between the longitudinal axis f of the towing vehicle 38 and a longitudinal axis a of the trailer 40. The position of the coupling point K is also schematically shown.

[0063] In the left-hand part, Fig. 4 shows, by way of example and schematically, radar data 20 in the form of a point cloud of detected radar reflections, wherein, in addition to the radial distance d and the radial velocity v, further features characterizing the radar reflection, such as azimuth and / or elevation angles and a radar cross-section, can be assigned to the individual radar reflections of the point cloud. In the right-hand part, Fig. 4 schematically shows, as an alternative example of radar data 20, a distance-velocity spectrum or distance-Doppler spectrum in which a definition range in a first dimension comprises a distance range d0 to d1 and in a second dimension comprises a velocity range v0 to v1. The values ​​of the spectrum can in particular be amplitudes or intensities of the received, sampled radar signals of a respective channel of the relevant radar sensor 10, 12.

[0064] A method for training the neural network 18 or the corresponding components of the neural network 28 is explained below with reference to Fig. 5.

[0065] In a first branch 50 of building the training dataset, static training data (unchangeable parameters) are collected for the respective trailer in step 52. In a parallel branch 54 of building the training dataset, dynamic data (data on dynamic variables) are compiled.

[0066] In step 56, radar data 20 and self-motion data 22 from at least one test drive of a towing vehicle 38 with a trailer 40 are recorded. In step 58, the radar data 20 is preprocessed, which may, for example, include filtering the radar data 20 for radar data that could potentially relate to the trailer 40. In step 60, the radar data or detections acquired during the test drive can be varied or supplemented with additional artificially generated radar data (augmentation).

[0067] In step 64, reference data is assigned to the radar data, for example, a reference angle for the trailer's articulation angle and a reference axle number. Reference data for the trailer data to be estimated later by the neural network are compiled here and assigned to the radar data. Data can be accessed in step 62, for example, via a vehicle bus, from the vehicle's own sensor system 24, and / or from at least one external sensor.

[0068] In step 66, the training data set is checked with regard to predefined characteristics of the training data set (for example, it is checked whether the radar data set corresponds to a desired radar data set and thus fulfills a desired completeness criterion), and in step 68, a decision is made as to whether a desired completeness criterion of the training data set has been met. If not, additional training data is generated in step 70, for example, using the method as in step 60. If the completeness criterion is met, the training data set, which includes the static data and the dynamic data, is used in step 72 to train the neural network.

Claims

Patent claims 1 . Method for detecting a trailer (40) of the towing vehicle (38) towed by a towing vehicle (38) using a radar system (10, 12) of the towing vehicle (38), comprising the following steps: Obtaining radar data (20) acquired by the radar system from the radar system (10, 12); and Inputting the radar data (20) into an artificial neural network (18), wherein the artificial neural network (18) receives and processes the input radar data (20) to generate at least one output signal (26) indicating the detection of a presence of a trailer (40) of the towing vehicle (38) pulled by the towing vehicle (38) and / or indicating trailer data characterizing a trailer (40) of the towing vehicle (38) pulled by the towing vehicle (38).

2. The method according to claim 1, wherein the radar data (20) input into the artificial neural network (18) comprise point detections of reflection points.

3. The method according to claim 1, wherein the radar data (20) input into the artificial neural network (18) input radar data (20) comprise at least one spectrum which comprises a definition range corresponding to a distance range.

4. Method according to one of the preceding claims, wherein the artificial neural network (18) comprises: a feature extraction network (30) into which the radar data are input; at least one object detection head (34) connected to the feature extraction network (30) for outputting object data (36) which are Radar data (20) characterize detected objects in the environment of the towing vehicle (38); and a trailer detection head (32) for outputting the at least one output signal (26).

5. The method according to any one of the preceding claims, wherein the trailer data comprises at least one of an articulation angle, a number of axles, a wheelbase, a classification of a trailer type, a length, a width, a height and a coupling position (K).

6. A method according to any one of the preceding claims, wherein the method further comprises: Obtaining movement data (22) recorded by a self-sensor system (24) of the towing vehicle (38) from the self-sensor system (24); and Inputting the movement data (22) into the artificial neural network (18), wherein the artificial neural network (18) receives and processes the input radar data (20) and the input movement data (22) to generate the at least one output signal (26) therefrom.

7. Method according to one of the preceding claims, wherein the method further comprises: Outputting a trailer presence signal or the trailer data to at least one driver assistance function (15) implemented in an evaluation and control unit (14) of the towing vehicle, based on the output signal (26) of the artificial neural network (18).

8. System for a towing vehicle, for detecting a trailer (40) of the towing vehicle (38) towed by the towing vehicle (38), the system comprising a radar system (10, 12) and an artificial neural network (18), the system comprising an evaluation and control unit (14) which is set up 9 -1 is to input radar data (20) acquired by the radar system (10, 12) into the artificial neural network (18), wherein the artificial neural network (18) is configured to receive and process the input radar data (20) in order to generate therefrom at least one output signal (26) which indicates the detection of a presence of a trailer pulled by the towing vehicle (38) (40) of the towing vehicle (38) and / or trailer data characterizing a trailer (40) of the towing vehicle (38) pulled by the towing vehicle (38).

9. A method for training an artificial neural network for use in the method according to any one of claims 1 to 7 or for use as the artificial neural network of the system according to claim 8, comprising the following steps: Acquiring (56) radar data (20) with the aid of a radar system (10, 12) of a towing vehicle (38) during at least one training run with the towing vehicle (38) with a trailer (40) of the towing vehicle (38) pulled by the towing vehicle (38); Assigning (64) reference data, wherein the reference data is assigned to the radar data (20), wherein the reference data indicates the presence of a trailer (40) of the towing vehicle (38) pulled by the towing vehicle (38) and / or indicates reference trailer data characterizing the trailer (40) of the towing vehicle (38) pulled by the towing vehicle (38); and Training (72) the artificial neural network (18) based at least on the radar data (20) and the reference data assigned thereto.

10. The method of claim 9, wherein the method further comprises: acquiring (62) vehicle movement data (22) of the towing vehicle (38) during the at least one training run, wherein the reference data is assigned to the radar data (20) and the vehicle movement data (22).

Images