Dynamic determination of deviations of a relative arrangement of radar sensors

Abstract

A computer-implemented method for dynamically determining deviations (140) of a relative arrangement (160) of at least one first radar sensor (162) and one second radar sensor (164) from each other is disclosed. The method comprises receiving first reflection data (130) of a first reflection signal determined using the first radar sensor (162), second reflection data (132) of a second reflection signal determined using the second radar sensor (164), crosstalk data (134) of a crosstalk signal determined using the first radar sensor (162), first noise data (136) of a first noise signal of ambient noise of the first sensor determined using the first radar sensor (162), and second noise data (138) of a second noise signal of ambient noise of the second sensor determined using the second radar sensor (164).Using the first reflection data (130), the second reflection data (132), the crosstalk data (134), the first noise data (136) and the second noise data (138), a deviation (140) of the relative arrangement (160) of the first and second radar sensors (162, 164) is determined.

Description

FIELD OF TECHNOLOGY

[0001] The invention relates to a computer-implemented method, a computer program and a computer unit for dynamically determining deviations of a relative arrangement of at least a first radar sensor and a second radar sensor to each other. STATE OF THE ART

[0002] The demands placed on modern radar systems, particularly due to the increasing automation of vehicles, are constantly rising. To meet these increased demands and enhance the performance of radar systems, the complexity of the systems is being increased in terms of both hardware and software. However, increasing software complexity, such as by employing super-resolution approaches, can have the disadvantage of significantly increasing the computing power required to run the software, and even then, the resulting improvements are often limited. Another approach is to increase hardware complexity, for example, by increasing the number of transmit and / or receive channels. Such hardware expansions, however, require more complex hardware, which is generally more expensive.

[0003] Therefore, there is a need for an alternative approach to improving the performance of radar systems, especially radar systems for vehicles, such as automobiles.

[0004] It is an object of the invention to provide an improved method for dynamically determining deviations of a relative arrangement of at least one first radar sensor and one second radar sensor from each other. The objects underlying the invention are achieved by the features of the independent claims.

[0005] In one aspect, a computer-implemented method for dynamically determining deviations of a relative arrangement of at least a first radar sensor and a second radar sensor is disclosed, which includes receiving first reflection data of a first reflection signal determined using the first radar sensor. The first reflection signal results from reflections of a first radar signal emitted by the first radar sensor from one or more environmental objects and includes distance information about one or more distances between the first radar sensor and the one or more environmental objects. Second reflection data of a second reflection signal determined using the second radar sensor are also received.The second reflection signal results from reflections of a second radar signal emitted by the second radar sensor from one or more surrounding objects and includes distance information about one or more distances between the second radar sensor and these objects. At least crosstalk data from a crosstalk signal determined using the first radar sensor is received. This crosstalk signal results from crosstalk between the second radar signal of the second radar sensor and the first radar sensor. Initial noise data from an initial noise signal of the first sensor's ambient noise, determined using the first radar sensor, is received. Furthermore, second noise data from a second noise signal of the second sensor's ambient noise, determined using the second radar sensor, is received.A deviation in the relative arrangement of the first and second radar sensors is determined using the first reflection data, the second reflection data, the crosstalk data, the first noise data, and the second noise data.

[0006] This allows the use of a combination of multiple radar sensors, for example, two radar sensors, to improve the performance of a radar system. This approach eliminates the need for complex modifications to either the radar system's software or hardware. The two or more radar sensors are spaced apart and independent of each other, meaning they are not, or only roughly, synchronized. Hereinafter, such a combination of at least two independent radar sensors, whose detection ranges at least partially overlap, will be referred to as a stereo radar system.For example, in the case of N radar sensors with N ≥ 2, reflection data from N are received using reflection signals determined by the N radar sensors, noise data from N are received using noise signals determined by the N radar sensors, and crosstalk data from one to N are received using crosstalk signals determined by one to N of the N radar sensors. Using the aforementioned data, deviations in the relative arrangement of the N radar sensors are determined, for example.

[0007] A stereo radar system is, for example, a stereo radar system of a vehicle, in particular a motor vehicle. The radar sensors of the stereo radar system are, for example, installed in a vehicle, in particular a motor vehicle, and configured to detect the vehicle's surroundings or objects in the vehicle's surroundings. The stereo radar system comprises, for example, a radar sensor of a front radar, corner radar, side radar, and / or rear radar of a vehicle, in particular a motor vehicle. For example, the stereo radar system comprises a combination of a radar sensor of a front radar and a radar sensor installed in a corner of the vehicle.

[0008] Such a stereo radar system can be used in a motor vehicle, for example, to implement assisted, automated, and / or autonomous driving functions. For instance, such a stereo radar system can be used in a motor vehicle for adaptive cruise control (ACC). ACC is a speed control system in motor vehicles that incorporates the distance to a vehicle ahead as an additional feedback and control variable. ACC can, for example, be part of a radar-based emergency braking system. Furthermore, such a stereo radar system can be used, for example, for a cross-traffic alert system to detect cross traffic behind the vehicle when reversing.

[0009] A stereo radar system, similar to a stereo camera system in optical image processing, improves the radar system's performance by combining radar data from at least two radar sensors at different levels. However, for optimal combined evaluation of the radar signals acquired by the two or more sensors, their relative arrangement—that is, their relative positions and / or orientations to one another—must be precisely known. Even deviations on the order of the radar signal wavelength, for example, about 4 mm at a frequency of 77 GHz, can drastically reduce the performance of a stereo setup with two or more radar sensors. Typical radar frequencies used, for example, are in the range of 76 to 77 GHz.

[0010] The radar sensors can be, for example, radar sensors of a frequency-modulated continuous wave radar (FMCW radar). In an FMCW radar, the transmitted radar signal is frequency-modulated. This frequency modulation enables distance measurement using an indirect time-of-flight measurement by comparing the frequency of the received radar signal to a reference, such as the transmitted radar signal. Radial velocities of detected objects can be determined, for example, using the Doppler shift. Thus, radar signals received by the radar sensors include information about the distance to detected objects, the radial velocity of the detected objects, and the distance of the objects.For example, when MIMO (Multiple Input Multiple Output) radar sensors are used, the position of objects can be determined not only in terms of their distance, but also, for example, in terms of azimuth and elevation angles. The radar signal can be analyzed using suitable signal processing, such as fast Fourier transforms (FFT), and the positions of detected objects, including distance, azimuth, and / or elevation angles, as well as radial velocity, can be determined. The presence of an object within the detection range of the radar sensors is indicated, for example, by an intensity peak in the received radar signal.

[0011] For example, determining the deviation in the relative arrangement of the first and second radar sensors enables online calibration, i.e., calibration during operation of the radar sensors and / or a vehicle incorporating the radar sensors. Such online calibration, for instance, of the relative positioning and / or alignment of the at least two radar sensors to each other, ensures optimal performance in all situations from the stereo approach, i.e., from the combined use of the at least two radar sensors or the combined evaluation of radar signals received using the at least two radar sensors. During online calibration, the absolute position of the at least two radar sensors relative to each other is not determined; rather, deviations in their relative arrangement are determined, i.e., deviations from the original calibration.Determining such deviations makes it possible to dynamically compensate for these deviations that may occur during the operation of radar sensors, for example in a vehicle.

[0012] Various effects are combined for calibration or deviation determination. For example, a machine learning module, perhaps with at least one neural network, can be used.

[0013] First, for example, reflection data from reflection signals determined using radar sensors are used to determine the deviation. This reflection data describes reflection signals resulting from reflections of radar signals from the radar sensors off one or more environmental objects. This reflection data includes, for example, distance information about one or more distances from the individual radar sensor to the one or more reflecting environmental objects, determined using the reflection signals. Unlike visual sensors, such as cameras, which do not natively provide distance information, radar sensors allow distances to be determined directly from the reflection signals to the corresponding one or more reflecting objects. For example, distance information to one or more significant objects is determined, e.g.,The radar sensors are selected for objects whose reflection signals each exhibit an intensity exceeding a predefined threshold. For example, the radar sensors can be positioned relatively close to each other, approximately 0.5 m to 2 m apart. In this case, it can be assumed that the reflecting objects, particularly the significant ones, should be largely identical.

[0014] From this distance information, deviations in the relative arrangement of the radar sensors can be determined, for example, using triangulation. Alternatively, distances with corresponding significant objects can be selected, and the intensity or power spectra of the reflection signals for these selected distances can be determined. The corresponding intensity or power spectra describe, for example, the intensity or power distribution of the reflection signal for a specific distance or distance range as a function of an azimuth angle and a polar or elevation angle. Based on the positional differences of intensity peaks or maxima within the area defined by the azimuth and elevation angles, deviations in the relative arrangement of the radar sensors can also be determined.In this case, intensity I and power P are proportional to each other with an area A as the constant of proportionality, i.e., I = P / A.

[0015] For example, a distance Doppler profile can be created, which describes the distribution of intensity or power as a function of distance and radial velocity. For this purpose, the parameter space for distance and radial velocity can be discretized by dividing it into intervals or bins. Based on intensity peaks, bins can be identified within which one or more objects are present or have been detected. For each of these bins, an intensity or power spectrum can then be determined as a function of azimuth and elevation angles, corresponding to a two-dimensional image. This image allows the position of the corresponding object(s) within the area defined by the azimuth and elevation angles to be determined.

[0016] For example, a machine learning module with one or more neural networks is used to determine the deviation. For instance, the reflection data represents an initial input for the machine learning module.

[0017] Secondly, for example, crosstalk data from a crosstalk signal between the second and first radar sensors is used to determine the deviation. The corresponding crosstalk signal results from crosstalk between the radar signal emitted by the second and the first radar sensors and includes crosstalk information related to that crosstalk. When using a machine learning module, the crosstalk data, for instance, constitutes a second input for the module.

[0018] Thus, crosstalk between radar sensors can be used to determine the deviation. If the radar sensors are arranged appropriately, there can be a direct signal path between the at least two radar sensors. This crosstalk signal, i.e., a radar signal emitted by one of the two radar sensors and received by the other, can be used as a further basis for determining the deviation in the form of crosstalk data. Specifically, the crosstalk signal transmitted between the at least two radar sensors is a radar signal transmitted via a direct transmission path between them.

[0019] For example, further crosstalk data can be received from a crosstalk signal determined using the second radar sensor, which results from crosstalk of the first radar signal from the first radar sensor to the second radar sensor.

[0020] Thirdly, noise data from ambient noise signals determined using at least two radar sensors is used to determine the deviation. This ambient noise is, for example, in the immediate vicinity of each radar sensor. The detected ambient noise depends, for example, on the installation angle of the respective radar sensor. Thus, noise data from the at least two radar sensors can be used, in particular, to determine a deviation in the relative alignment of the at least two radar sensors. When using a machine learning module, the noise data from the at least two radar sensors constitutes, for example, a third input for the module.

[0021] To create the most accurate possible calibration of the relative arrangement of the at least two radar sensors of the stereo radar, the temporal relationship of several measurements is used, i.e., a temporal profile of reflection data, crosstalk data and noise data is used.

[0022] The specific deviation is provided, for example, for a dynamic adjustment of a combined evaluation of radar signals received using at least two radar sensors for a dynamic compensation of the specific deviation of the relative arrangement.

[0023] Examples illustrate the calibration, particularly automatic calibration, of a stereo radar system comprising at least two radar sensors. Vibration, thermal changes, and mechanical alterations can significantly change the initial orientation and / or position of the radar sensors relative to each other. Such vibration, thermal changes, and mechanical alterations can occur, especially in the case of a stereo radar system integrated into a motor vehicle. Due to these dynamic changes in the relative arrangement of the radar sensors, an initial static calibration, for example, at the end of production of the stereo radar system and / or a device such as a motor vehicle into which the stereo radar system is installed, may not be sufficient to guarantee precise and / or error-free operation of the stereo radar system in the long term.

[0024] Effective dynamic auto-calibration enables the reliable operation of a stereo radar system. Unlike a single radar sensor, a stereo approach with multiple radar sensors provides significantly more data for calibration. To effectively combine this wealth of data, acquired by the multiple radar sensors of the stereo radar system, a machine learning module can be employed. This module comprises one or more neural networks, particularly a recurrent neural network, which is well-suited for such data combination. This approach can achieve high calibration accuracy and, moreover, ensure sufficient robustness.For example, such dynamic auto-calibration can provide a basis for a series-production-ready version of stereo radar systems, which are particularly suitable for use in motor vehicles.

[0025] For example, the deviation of the relative arrangement includes a deviation in the relative positioning of the at least two radar sensors relative to each other. For example, the deviation of the relative arrangement includes a deviation in the relative orientation of the at least two radar sensors relative to each other. Thus, deviations resulting from both translations and rotations can be determined.

[0026] For example, the method further includes dynamically adjusting a combined evaluation of third radar signals received using the at least two radar sensors. The dynamic adjustment of the combined evaluation includes dynamically adjusting one or more parameter values ​​used for the combined evaluation, which describe the relative arrangement of the at least two radar sensors.

[0027] Through appropriate dynamic adjustment, the effects of deviations can be compensated for in the combined evaluation of radar signals, which combines radar signals from at least two radar sensors. For example, parameter values ​​used in the evaluation calculation, which are based on the relative arrangement of the at least two radar sensors, are adjusted in such a way that the deviations are taken into account and errors or inaccuracies due to unaccounted-for deviations can be avoided. For instance, the radar signals received by each radar sensor individually are used to determine the deviation and then used for a combined evaluation, with the results of the deviation determination forming the basis for the combined evaluation.

[0028] For example, the one or more parameter values ​​of the relative arrangement of the at least two radar sensors include a distance between the at least two radar sensors. For example, the one or more parameter values ​​of the relative arrangement of the at least two radar sensors include a relative inclination of the at least two radar sensors to each other. Thus, deviations in the form of translations as well as rotations can be taken into account during compensation.

[0029] For example, the deviation of the relative arrangement is determined using a machine learning module that is trained to receive the first reflection data, the second reflection data, the crosstalk data, the first noise data and the second noise data as inputs and to output the deviation of the relative arrangement of the first and second radar sensors.

[0030] For example, the procedure can further include training a machine learning module to provide or train a trained machine learning module to determine the deviation of the relative arrangement. For example, the trained machine learning module can be trained to receive the first reflection data, the second reflection data, the crosstalk data, the first noise data, and the second noise data as inputs and to output the deviation of the relative arrangement of the first and second radar sensors.

[0031] To train the machine learning module, a set of training datasets can be provided. For example, each training dataset can include first training reflection data, second training reflection data, training crosstalk data, first training noise data, second training noise data, and a training deviation. The machine learning module can be trained to receive the combination of first training reflection data, second training reflection data, training crosstalk data, first training noise data, and second training noise data from the training datasets as input and, in response, provide the training deviation of the respective training dataset as output.

[0032] For example, the machine learning module includes at least one neural network. For example, the machine learning module includes at least one recurrent neural network.

[0033] To effectively combine the majority of data acquired using the multiple radar sensors of the stereo radar system, a recurrent neural network, for example, can be used, as it is well-suited for such data combination. This allows for high calibration accuracy and also ensures sufficient robustness.

[0034] For example, each radar sensor provides different data, which can be acquired using one or more individual measurements. For instance, the first radar sensor receives reflection data from objects with high reflectivity, crosstalk data from a direct radar signal received from the second radar sensor, and noise data from a noise signal in the immediate vicinity of the first radar sensor. Similarly, the second radar sensor receives reflection data from objects with high reflectivity and noise data from a noise signal in the immediate vicinity of the second radar sensor. If the radar sensors are positioned close to each other, for example, approximately 0.5 m to 2 m apart, and the distance to the detected objects is sufficiently large, for example, 50 m or more, it can be assumed that the objects are largely identical.

[0035] To jointly evaluate the received data, a machine learning module with a multi-stage configuration can be used. The first stage involves preprocessing the various received data points. This stage can utilize classical signal processing algorithms or even include one or more neural networks. The goal of preprocessing is to transform all the data into a similar and interpretable form for the network. One approach is to transform all the data into a 2D representation or to selectively extract specific properties from the received data.

[0036] The preprocessed data is then used, for example, as input for a neural network, such as a recurrent neural network. This neural network performs a regression, for instance, to determine a deviation in the relative arrangement of the radar sensors, such as in their position and / or orientation relative to each other, and outputs this deviation as a response to the input. In a further stage, the output of the neural network is used, for example, during post-processing, to compensate for this specific deviation. For instance, one or more parameter values ​​used for a combined evaluation of radar signals received from at least two radar sensors are adjusted using the specific deviation to compensate for it.The adjusted parameter values ​​describe, for example, the relative arrangement of the at least two radar sensors to each other. For example, the parameter values ​​for the relative arrangement of the at least two radar sensors include a distance between the at least two radar sensors. For example, the parameter values ​​for the relative arrangement of the at least two radar sensors include a relative inclination of the at least two radar sensors to each other.

[0037] A recurrent neural network can advantageously be used for evaluating a temporal relationship of the measurements, i.e., a temporal progression of reflection data, crosstalk data, and noise data.

[0038] Using a recurrent neural network, the temporal relationship of individual measurements can be taken into account and weighted accordingly. This allows calibration to be performed over extended periods. For example, if the radar sensors are those of a vehicle, effects that occur during driving, such as those caused by potholes, can be efficiently suppressed during long-term calibration. Furthermore, by considering the temporal profile, for example using a recurrent neural network, effects caused by periodic vibrations of vehicle components can be estimated, determined, and compensated for.

[0039] A recurrent neural network is a feedback-based neural network which, unlike a feedforward network, includes connections from model neurons of one layer to neurons of the same or a previous layer. Such recurrent connections of model neurons can be used to determine temporally encoded information in data. A recurrent neural network has a memory because it uses information from previous inputs to influence the current input and output. Another characteristic of recurrent networks is that they share parameters across all layers of the network. While feedforward networks have different weights across each node, recurrent neural networks share, for example, the same weight parameter within each layer.However, these weights are still adjusted by the processes of backpropagation and gradient descent to facilitate reinforcement learning.

[0040] Examples of such a recurrent neural network are an Elman network, a Jordan network, a Hopfield network, or a fully interconnected neural network.

[0041] Training a recurrent neural network involves, for example, the following procedure: A training dataset is fed into the recurrent neural network. The recurrent neural network receives this training dataset and performs some complex calculations with randomly initialized variables. A predicted result is output by the recurrent neural network in response to receiving the training dataset. This result is compared to an expected training result from the training dataset, which generally yields an error. If the corresponding error propagates back along the same path, the variables are adjusted. The aforementioned steps are repeated until the variables are well-defined and the result matches the expected training result.The difference to a feedforward network arises from the fact that the previous inputs are known before the result is evaluated. Recurrent neural networks can therefore be viewed as multiple neural feedforward networks, with information being passed from one network to another.

[0042] For example, the machine learning module includes one or more convolutional neural networks (CNNs). For instance, the evaluation of the temporal relationship of the measurements—that is, the temporal evolution of reflection data, crosstalk data, and noise data—can be processed not recursively, but using one or more convolutional neural networks.

[0043] A convolutional neural network comprises three main layer types: convolutional layers, pooling layers, and fully connected layers. For example, a convolutional neural network includes a structure with one or more convolutional layers followed by a pooling layer. This structure can be repeated one or more times. Convolutional layers are configured to perform discrete convolutions and detect and extract features from the input data. Pooling layers, also known as downsampling layers, reduce the dimensions and thus the number of input parameters. A fully connected layer completes the convolutional neural network. In such a fully connected layer, every node of the output layer is directly connected to a node of the previous layer.Such a fully interconnected layer enables classification based on the features extracted by the previous layers and the various associated filters.

[0044] For example, Fast-ABC can be used as a neural network, as described in Xiaoru Xie et al., “Fast-ABC: A Fast Architecture for Bottleneck-Like Based Convolutional Neural Networks”, 2019 IEEE Computer Society Annual Symposium on VLSI (ISVLSI), pages 1 to 6. Fast-ABC provides an accelerator for BLO (Bottleneck-Like Operations)-based convolutional networks.

[0045] For example, a UNet can be used as a neural network, as described in Olaf Ronneberger, “U-Net: Convolutional Networks for Biomedical Image Segmentation” (https: / / arxiv.org / abs / 1505.04597). The architecture of the U-Net is based on a so-called fully convolutional network. The basic idea is to augment an ordinary contraction network with successive layers in which pooling operations are replaced by upsampling operators. These layers thus increase the resolution of the output. A subsequent convolutional layer can then learn to assemble a precise output based on this information. The upsampling part of the U-Net contains a large number of feature channels, which allow the network to pass contextual information to higher-resolution layers.As a result, the expansive path is more or less symmetrical to the contracting part and leads to a U-shaped architecture of the network.

[0046] For example, a ResNet, or Residual Neural Network, can be used as the neural network, as described in Kaiming He et al., “Deep Residual Learning for Image Recognition” (https: / / arxiv.org / abs / 1512.03385). A ResNet is a deep learning model in which weighted layers learn residual functions with respect to the layer inputs. The corresponding network includes skip connections, also called residual connections, which perform identity mappings and are combined with the layer outputs by addition.

[0047] For example, a time course of the deviation of the relative arrangement of the first and second radar sensors is determined.

[0048] For this purpose, a temporal relationship between several measurements is used, i.e., a time course of reflection data, crosstalk data, and noise data. For example, the time course of the crosstalk behavior can be taken into account. In particular, periodic deviations over time can also be determined in this way.

[0049] For example, a periodically repeating pattern of the deviation over time is determined. This determined deviation, as part of the periodically repeating pattern, is used for a periodically repeating dynamic adjustment of the combined evaluation of the third radar signals received using at least two radar sensors. Thus, for example, periodically occurring vibrations, such as those caused by the characteristics of a vehicle containing the radar sensors or by the road surface on which the vehicle is traveling, can be effectively determined and compensated for.

[0050] In another aspect, a computer program for dynamically determining deviations of a relative arrangement of at least one first radar sensor and one second radar sensor is disclosed. The computer program comprises machine-readable program instructions. Execution of the machine-readable program instructions by a processor unit of a computer unit causes the processor unit to control the computer unit to receive a first reflection signal determined using the first radar sensor. The first reflection signal results from reflections of a first radar signal emitted by the first radar sensor from one or more surrounding objects and includes distance information about one or more distances between the first radar sensor and the one or more surrounding objects. Second reflection data of a second reflection signal determined using the second radar sensor is also received.The second reflection signal results from reflections of a second radar signal emitted by the second radar sensor from one or more surrounding objects and includes distance information about one or more distances between the second radar sensor and these objects. At least crosstalk data from a crosstalk signal determined using the first radar sensor is received. This crosstalk signal results from crosstalk between the second radar signal of the second radar sensor and the first radar sensor. Initial noise data from an initial noise signal of the first sensor's ambient noise, determined using the first radar sensor, is received. Furthermore, second noise data from a second noise signal of the second sensor's ambient noise, determined using the second radar sensor, is received.A deviation in the relative arrangement of the first and second radar sensors is determined using the first reflection data, the second reflection data, the crosstalk data, the first noise data, and the second noise data.

[0051] For example, a computer program product for dynamically determining deviations of a relative arrangement of at least one first radar sensor and a second radar sensor comprises a computer-readable storage medium containing machine-readable program instructions. Execution of these machine-readable program instructions by a processor unit of a computer unit causes the processor unit to control the computer unit to receive a first reflection signal determined using the first radar sensor. The first reflection signal results from reflections of a first radar signal emitted by the first radar sensor from one or more surrounding objects and includes distance information about one or more distances between the first radar sensor and these objects. Second reflection data of a second reflection signal determined using the second radar sensor is also received.The second reflection signal results from reflections of a second radar signal emitted by the second radar sensor from one or more surrounding objects and includes distance information about one or more distances between the second radar sensor and these objects. At least crosstalk data from a crosstalk signal determined using the first radar sensor is received. This crosstalk signal results from crosstalk between the second radar signal of the second radar sensor and the first radar sensor. Initial noise data from an initial noise signal of the first sensor's ambient noise, determined using the first radar sensor, is received. Furthermore, second noise data from a second noise signal of the second sensor's ambient noise, determined using the second radar sensor, is received.A deviation in the relative arrangement of the first and second radar sensors is determined using the first reflection data, the second reflection data, the crosstalk data, the first noise data, and the second noise data.

[0052] In another aspect, a computer unit for dynamically determining deviations of a relative arrangement of at least one first radar sensor and one second radar sensor is disclosed. The computer unit comprises a processor unit and a memory unit with machine-readable program instructions. Execution of the machine-readable program instructions by the processor unit causes the processor unit to control the computer unit to receive a first reflection signal determined using the first radar sensor. The first reflection signal results from reflections of a first radar signal emitted by the first radar sensor from one or more surrounding objects and includes distance information about one or more distances between the first radar sensor and the one or more surrounding objects.Second reflection data of a second reflection signal determined using the second radar sensor are received. The second reflection signal results from reflections of a second radar signal emitted by the second radar sensor from one or more environmental objects and includes distance information about one or more distances between the second radar sensor and the one or more environmental objects. At least crosstalk data of a crosstalk signal determined using the first radar sensor are received. The crosstalk signal results from crosstalk between the second radar signal of the second radar sensor and the first radar sensor. First noise data of a first noise signal of the first sensor's ambient noise, determined using the first radar sensor, are received.Furthermore, second noise data from a second noise signal of the second radar sensor, determined using the second radar sensor's ambient noise, are received. A deviation in the relative arrangement of the first and second radar sensors is determined using the first reflection data, the second reflection data, the crosstalk data, the first noise data, and the second noise data.

[0053] It is understood that one or more of the aforementioned embodiments can be combined with each other, as long as the embodiments do not exclude each other. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The following examples are explained in more detail using the drawings. They show: Fig. 1. A flowchart of an exemplary procedure for dynamically determining deviations of a relative arrangement of radar sensors, Fig. 2. A block diagram of an exemplary computer unit for dynamically determining deviations of a relative arrangement of radar sensors. Fig. 3 a block diagram of an exemplary arrangement of radar sensors, and Fig. 4 a flowchart of an exemplary procedure for dynamically determining deviations of a relative arrangement of radar sensors. DETAILED DESCRIPTION

[0055] In the following, similar elements are marked with the same reference symbols.

[0056] Fig. Figure 1 shows an exemplary computer-implemented method for dynamically determining deviations of a relative arrangement of at least one first radar sensor and one second radar sensor from each other. In block 200, initial reflection data of a first reflection signal determined using the first radar sensor are received. The first reflection signal results from reflections of a first radar signal emitted by the first radar sensor from one or more surrounding objects and includes distance information about one or more distances between the first radar sensor and the one or more surrounding objects.

[0057] Block 202 receives second reflection data of a second reflection signal determined using the second radar sensor. This second reflection signal results from reflections of a second radar signal emitted by the second radar sensor off one or more surrounding objects and includes distance information about one or more distances between the second radar sensor and these objects.

[0058] Block 204 receives at least crosstalk data from a crosstalk signal determined using the first radar sensor. The crosstalk signal results from crosstalk between the second radar signal from the second radar sensor and the first radar sensor.

[0059] In block 206, the first noise data of an initial noise signal of the first sensor, determined using the first radar sensor, is received.

[0060] In block 208, second noise data of a second noise signal of ambient noise from the second sensor, determined using the second radar sensor, are received.

[0061] Block 210 determines the deviation in the relative arrangement of the first and second radar sensors using the first reflection data, the second reflection data, the crosstalk data, the first noise data, and the second noise data. The determined deviation in the relative arrangement includes, for example, a deviation in the relative positioning of the at least two radar sensors relative to each other. For example, the deviation in the relative arrangement includes a deviation in the relative orientation of the at least two radar sensors relative to each other. To determine the deviation in the relative arrangement of the two radar sensors, a machine learning module is used, which may include one or more neural networks. The neural networks may include, for example, one or more recurrent neural networks.

[0062] The specific deviation is provided, for example, for a dynamic adjustment of a combined evaluation of radar signals received using at least two radar sensors for a dynamic compensation of the specific deviation of the relative arrangement.

[0063] In Block 212, for example, a dynamic adjustment of a combined evaluation of radar signals received using at least two radar sensors takes place. This dynamic adjustment of the combined evaluation includes a dynamic adjustment of one or more parameter values ​​used for the combined evaluation, which describe the relative arrangement of the at least two radar sensors.

[0064] Fig. Figure 2 shows an exemplary computer unit or computer device 102 for dynamically determining deviations of a relative arrangement of at least one first radar sensor and one second radar sensor from each other. The computer unit 102 can be integrated into various types of systems. For example, the computer unit 102 is integrated into a motor vehicle. This computer unit 102 integrated into the motor vehicle can also be implemented as a distributed system. The computer unit 102 shown comprises a processor unit or computing unit 104. The processor unit 104 can be, for example, an integrated circuit in the form of a microprocessor or a microcontroller in an embedded system. The processor unit 104 shown represents one or more processor units.The computer unit 102 shown also includes an optional hardware interface 106. The hardware interface enables the processor unit 104 to control other components, such as radar sensors. The processor unit 104 shown can also be connected to an optional user interface 108. The user interface 108 could, for example, be a user interface 108 of an on-board computer of a motor vehicle. The user interface 108 could also include, for example, a display device. This could include, for example, a two-dimensional computer display, a touchscreen, a virtual reality system, and / or an augmented reality system.

[0065] The depicted processor unit 104 is also connected to a memory unit 110. The memory unit 110 represents various types of storage that the processor unit 104 can access. In one example, the memory unit 110 is a non-transferable storage medium.

[0066] The storage unit 110 is depicted as containing machine-readable and machine-executable program instructions 120. These machine-readable program instructions 120 enable the processor unit 104 to perform various numerical, image-processing, and computational tasks. The machine-readable program instructions 120 also allow the processor unit 104 to control and operate other components via the hardware interface 106, such as two or more radar sensors. The execution of the machine-readable program instructions 120 by the processor unit 104 can cause the processor unit 104 to control the computer unit 102 in order to execute the procedure for dynamically determining deviations of a relative arrangement of at least one first radar sensor and a second radar sensor from each other.

[0067] The depicted storage unit 110 further comprises, for example, initial reflection data 130 of an initial reflection signal determined using the first radar sensor. The initial reflection signal results from reflections of an initial radar signal emitted by the first radar sensor from one or more surrounding objects and includes distance information about one or more distances between the first radar sensor and the one or more surrounding objects. The initial reflection data 130 are received by the computer unit 110, for example, via the hardware interface 106 from the first radar sensor.

[0068] The depicted storage unit 110 further comprises, for example, second reflection data 132 of a second reflection signal determined using the second radar sensor. The second reflection signal results from reflections of a second radar signal emitted by the second radar sensor from one or more environmental objects and includes distance information about one or more distances between the second radar sensor and the one or more environmental objects. The second reflection data 132 are received by the computer unit 110, for example, via the hardware interface 106 from the second radar sensor.

[0069] The depicted storage unit 110 further comprises, for example, at least crosstalk data 134 of a crosstalk signal determined using the first radar sensor. The crosstalk signal results from crosstalk between the second radar signal of the second radar sensor and the first radar sensor. The crosstalk data 134 are received by the computer unit 110, for example, via the hardware interface 106 from the first radar sensor.

[0070] The depicted storage unit 110 further comprises, for example, initial noise data 136 of an initial noise signal of ambient noise from the first radar sensor, determined using the first radar sensor. The initial noise data 136 are received by the computer unit 110, for example, via the hardware interface 106 from the first radar sensor.

[0071] The depicted storage unit 110 further comprises, for example, second noise data 138 of a second noise signal of ambient noise from the second radar sensor, determined using the second radar sensor. The second noise data 138 are received by the computer unit 110, for example, via the hardware interface 106 from the second radar sensor.

[0072] The first reflection data 130, the second reflection data 132, the crosstalk data 134, the first noise data 136, and the second noise data 138 are used by the processor unit 104 to determine a deviation 140 of the relative arrangement of the first and second radar sensors. For example, the determined deviation 140 of the relative arrangement of the radar sensors includes a deviation in the relative positioning of the at least two radar sensors relative to each other and / or in the relative orientation of the at least two radar sensors relative to each other.

[0073] Furthermore, the computer unit 102 can be configured to compensate for the specific deviation 140. For example, one or more parameter values ​​142 used for a combined evaluation of radar signals received using the at least two radar sensors are adjusted using the specific deviation 140 in order to compensate for this specific deviation 140. The adjusted parameter values ​​142 describe, for example, the relative arrangement of the at least two radar sensors to each other. For example, the parameter values ​​142 for the relative arrangement of the at least two radar sensors include a distance between the at least two radar sensors. For example, the parameter values ​​142 for the relative arrangement of the at least two radar sensors include a relative inclination of the at least two radar sensors to each other.

[0074] For example, determining the deviation 140 of the relative arrangement is performed using a machine learning module 122, which is trained to receive the first reflection data 130, the second reflection data 132, the crosstalk data 134, the first noise data 136, and the second noise data 138 as inputs and to output the deviation 140 of the relative arrangement of the first and second radar sensors. For example, the machine learning module 122 includes at least one neural network 124, which is, for example, a recurrent neural network. Furthermore, the machine learning module 122 can, for example, include a submodule 126, which is configured for preprocessing the received data 130, 132, 134, 136, and 138. For example, the preprocessed data is used as input for the neural network 124.Furthermore, the machine learning module 122 can, for example, include a submodule 128, which is configured to compensate for the deviation 140 by adjusting the parameter values ​​142. For example, the output of the neural network 124, perhaps in the form of the determined deviation 140, is used for post-processing.

[0075] Fig. Figure 3 shows an exemplary arrangement 160 of two radar sensors 162, 164 of a stereo radar system. The arrangement 160 comprises a first radar sensor 162 and a second radar sensor 164, which are arranged at a distance 166 from each other. The two radar sensors can, for example, be aligned parallel to each other with their detection fields. For example, the radar sensors 162, 164, as in the example shown, can be tilted or inclined relative to each other. In the example shown, the second radar sensor 164 is inclined by an inclination 168 relative to the first radar sensor 162. Deviations in the relative positioning, in particular the distance between the radar sensors 162, 164, and / or the relative orientation of the radar sensors 162, 164, can occur, for example, as a result of vibrations, thermal changes, but also mechanical changes.This can significantly alter the initial orientation and / or position of the radar sensors 162, 164 relative to each other. Such vibrations, thermal changes, and mechanical changes can occur, particularly in the case of a stereo radar system integrated into a motor vehicle. Precisely defined deviations can be used to compensate for parameter values ​​that are applied to a combined evaluation of radar signals received using the two radar sensors 162, 164. The adjusted parameter values ​​describe, for example, the change in the distance 166 between the two radar sensors 162, 164 resulting from the deviation and / or the change in the inclination 168 of the two radar sensors 162, 164 relative to each other resulting from the deviation.

[0076] Fig.Figure 4 shows an exemplary computer-implemented method for dynamically determining deviations of a relative arrangement of at least a first radar sensor and a second radar sensor using a machine learning module.

[0077] Each of the at least two radar sensors provides various data acquired using one or more individual measurements. For example, the first radar sensor receives initial reflection data 130 from objects with high reflectivity, crosstalk data 134 of a direct radar signal received by the first radar sensor from the second radar sensor, and initial noise data 136 of a noise signal in the immediate vicinity of the first radar sensor. The initial reflection signal 130 results from reflections of an initial radar signal emitted by the first radar sensor from one or more surrounding objects and includes distance information about one or more distances between the first radar sensor and these surrounding objects. The crosstalk data includes crosstalk information about crosstalk between the second radar signal from the second radar sensor and the first radar sensor.

[0078] For example, the second radar sensor receives second reflection data 132 from objects with high reflectivity as well as second noise data 138 from a noise signal in the vicinity of the second radar sensor. The second reflection signal 132 results from reflections of a second radar signal emitted by the second radar sensor from one or more surrounding objects and includes distance information about one or more distances between the second radar sensor and the one or more surrounding objects.

[0079] If the at least two radar sensors are arranged at a small distance of approximately 0.5 m to 2 m relative to each other, it can be assumed that the objects are largely identical.

[0080] To jointly evaluate the data 130, 132, 134, 136, 138 received from at least two radar sensors, a machine learning module 122 can be used, which may have a multi-stage configuration. In a first stage, for example, the various received data 130, 132, 134, 136, 138 are preprocessed by a specially configured sub-module or preprocessing module 126. At this stage, classical signal processing algorithms can be used, but it can also include one or more neural networks. The goal of the preprocessing is to convert all data 130, 132, 134, 136, 138 into a similar form that is interpretable by the network. One possibility, for example, is to transform all data 130, 132, 134, 136, 138 into a 2D representation or to specifically extract properties from the received data.Furthermore, preprocessing can include, for example, outlier detection, averaging and / or filtering.

[0081] The preprocessed data is then used, for example, as input for a neural network, such as a recurrent neural network 124, at the next stage. This recurrent neural network 124 performs, for example, a regression to determine a deviation in the relative arrangement of the radar sensors, such as in the position and / or orientation of the radar sensors relative to each other, and outputs this as a response to the received input.

[0082] At a further stage, the output of the recurrent neural network 124 is used, for example, during post-processing by a configured sub-module or post-processing module 128 to compensate for the deviation determined by the recurrent neural network 124. For example, one or more parameter values ​​used for a combined evaluation of radar signals received using the at least two radar sensors are adjusted using the determined deviation in order to compensate for this deviation. The adjusted parameter values ​​describe, for example, the relative arrangement of the at least two radar sensors to each other. For example, the parameter values ​​for the relative arrangement of the at least two radar sensors include a distance between the at least two radar sensors.For example, the parameter values ​​of the relative arrangement of the at least two radar sensors include a relative inclination of the at least two radar sensors to each other. Furthermore, post-processing can include, for example, averaging and / or filtering.

[0083] Although the invention is illustrated and described in detail in the drawings and the preceding description, this illustration and description is to be regarded as exemplary and not limiting; the invention is not limited to the disclosed examples.

[0084] Other variations of the disclosed examples can be understood and carried out by those skilled in the art when carrying out the claimed invention with reference to the drawings, the description, and the accompanying claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" does not exclude multiple elements. The mere fact that certain features are mentioned in differing dependent claims does not mean that a combination of these features cannot be advantageous. Any reference numerals in the claims should not be interpreted as limiting the scope of protection.

[0085] A single processor or other unit can perform the functions of several elements mentioned in the claims. A computer program can be stored / distributed on a suitable medium, for example, on an optical storage medium or a solid-state medium supplied with or as part of other hardware, but it can also be distributed in other ways, for example, via the Internet or other wired or wireless telecommunications systems.

[0086] As those skilled in the art will understand, aspects of the present invention can be embodied in the form of a device, a method, or a computer program product. Accordingly, aspects of the present invention can take the form of a purely hardware variant, a purely software variant (including firmware, resident software, microcode, etc.), or a variant that combines software and hardware aspects, which may be generally referred to here as a "circuit," "module," or "system." Furthermore, aspects of the present invention can take the form of a computer program product embodied in one or more computer-readable media containing computer-executable code.

[0087] Any combination of one or more computer-readable media can be used. The computer-readable medium can be a computer-readable signaling medium or a computer-readable storage medium. A "computer-readable storage medium," as used here, includes any tangible storage medium capable of storing instructions executable by a processor or computing system of a computer unit. The computer-readable storage medium may be referred to as a computer-readable non-transitory storage medium. The computer-readable storage medium may also be referred to as a tangible computer-readable medium. In some embodiments, a computer-readable storage medium may also be capable of storing data accessible to the processor or computing system of the computer unit.Examples of computer-readable storage media include: a floppy disk, a magnetic hard disk drive, a solid-state drive, flash memory, a USB flash drive, random access memory (RAM), read-only memory (ROM), an optical disc, a magneto-optical disc, and the register file of the processor or computer system. Examples of optical discs are compact discs (CDs) and digital versatile discs (DVDs), such as CD-ROM, CD-RW, CD-R, DVD-ROM, DVD-RW, or DVD-R discs. The term "computer-readable storage medium" also refers to various types of recording media that the computer can access via a network or communication connection. For example, data can be retrieved via a modem, the internet, or a local network.Computer-executable code embodied on a computer-readable medium may be transmitted via any suitable medium, including but not limited to wireless transmission, wired transmission, fiber optic cable, radio frequency transmission, etc., or via a suitable combination of the aforementioned media.

[0088] A computer-readable signaling medium can contain a propagating data signal with computer-executable code embodied therein, for example, in a baseband or as part of a carrier wave. Such a transmitted signal can take any form, including, but not limited to, electromagnetic or optical signals, or a suitable combination thereof. A computer-readable signaling medium can be any computer-readable medium that is not a computer-readable storage medium and that can transmit, propagate, or transport a program for use by or in conjunction with a command execution system, apparatus, or device.

[0089] A "computer memory," "storage unit," or "memory" is an example of a computer-readable storage medium. Computer memory is any memory that a processor or computing system can directly access.

[0090] A "processor system," "processor unit," "computer system," or "computer unit," as used herein, comprises an electronic component capable of executing a program, a machine-executable instruction, or computer-executable code. References to the processor system or computer system that include an example "a processor system" or "computer system" are to be understood as meaning that the example may include more than one processor system, processor unit, computer system, computer unit, or processor core. For example, the processor system or computer system may be a multi-core processor. A processor system, processor unit, computer system, or computer unit may also refer to a collection of processor units or computer units within a single computer system or distributed across multiple computer systems.The terms "processor system," "processor unit," "computer system," or "computing unit" should also be interpreted as potentially referring to a collection or network of computing devices, each comprising a processor or computing system. The machine-executable code or instructions may be executed by multiple computing systems or processors located within the same computing device or even distributed across multiple computing devices.

[0091] Machine-readable or machine-executable instructions, or computer-readable or computer-executable code, may comprise instructions or a program that causes a processor or other computing system to execute an aspect of the present invention. Computer-executable code for performing operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object-oriented programming language such as Java, Smalltalk, C++, or similar languages, and conventional procedural programming languages ​​such as the programming language "C" or similar languages, and compiled into machine-executable instructions. In some cases, the computer-executable code may be in the form of a high-level language or in pre-compiled form and used in conjunction with an interpreter that generates the machine-executable instructions on the fly.In other cases, the machine-executable instructions or computer-executable code may be in the form of programming for programmable logic gate arrays.

[0092] The executable computer code can be executed entirely on the user's computer unit, partially on the user's computer unit, as a standalone software package, partially on the user's computer unit and partially on a remote computer unit, or entirely on the remote computer unit or server. In the latter case, the remote computer unit can be connected to the user's computer unit via any network, including a local area network (LAN) or a wide area network (WAN), or the connection can be established with an external computer unit (for example, via the internet using an internet service provider).

[0093] Aspects of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the invention. It is understood that each block or part of the blocks of the flowchart, illustrations, and / or block diagrams can be implemented by computer program instructions in the form of computer-readable or computer-executable code, where applicable. It is further understood that combinations of blocks in different flowcharts, illustrations, and / or block diagrams can be combined, provided they are not mutually exclusive.These computer program instructions can be provided to a computing system of a general-purpose computer, a special-purpose computer, or any other programmable data processing device to create a machine such that the instructions executed through the computing system of the computer or other programmable data processing device provide means for implementing the functions / actions specified in the flowchart and / or block diagram block or blocks.

[0094] These machine-executable instructions or computer program instructions may also be stored in a computer-readable medium capable of instructing a computer, other programmable data processing device, or other apparatus to operate in a particular manner, such that the instructions stored in the computer-readable medium produce a manufactured item containing instructions to perform the function / action specified in the flowchart and / or block diagram block or blocks.

[0095] The machine-readable or machine-executable instructions or computer program instructions can also be loaded onto a computer, other programmable data processing device, or other devices to initiate a series of procedural steps that are executed on the computer, other programmable device, or other devices to create a computer-implemented process, so that the instructions executed on the computer or other programmable device provide processes for implementing the functions / actions specified in the flowchart and / or block diagram block or blocks.

[0096] A "user interface," as used here, is an interface that allows a user or operator to interact with a computer or computer system. A "user interface" can also be called a "human interface device." A user interface can provide information or data to the user and / or receive information or data from the user. A user interface can allow the computer to receive input from the user and provide output from the computer to the user. In other words, the user interface can allow a user to control or manipulate a computer, and the interface can allow the computer to display the effects of the user's control or manipulation.Displaying data or information on a screen or graphical user interface is an example of providing information to a user. Receiving data via a keyboard, mouse, trackball, touchpad, pointer, graphics tablet, joystick, gamepad, webcam, headset, pedals, wired glove, remote control, and accelerometer are all examples of user interface components that enable a user to receive information or data.

[0097] A "hardware interface," as used here, comprises an interface that enables the processor or computing system of a computer unit or computer system to interact with and / or control an external computer device and / or external apparatus. A hardware interface can allow a computer unit to send control signals or commands to an external computer device and / or external apparatus. A hardware interface can also allow a computer unit to exchange data with an external data processing system and / or external device.Examples of hardware interfaces include: a universal serial bus, an IEEE-1394 port, a parallel port, an IEEE-1284 port, a serial port, an RS-232 port, an IEEE-488 port, a Bluetooth connection, a wireless local area network connection, a TCP / IP connection, an Ethernet connection, a control voltage interface, a MIDI interface, an analog input interface, and a digital input interface.

[0098] A "display," "indicator," or "display device," as used here, comprises an output device or user interface capable of displaying images or data. A display can output visual, auditory, and / or tactile data. Examples of a display include, but are not limited to: a computer monitor, a television screen, a touchscreen, a tactile electronic display, and a Braille display.

[0099] Cathode ray tube (CRT), storage tube, bistable display, electronic paper, vector display, flat panel display, vacuum fluorescent display (VF display), light-emitting diode display (LED), electroluminescent display (ELD), plasma display panel (PDP), liquid crystal display (LCD), organic light-emitting diode display (OLED), projector and head-mounted display.

[0100] The term "machine learning" (ML) refers to a computer algorithm used to extract useful information from training datasets by automatically creating probabilistic models called machine learning modules or models. A machine learning module can also be called a predictive model. Machine learning algorithms create a mathematical model based on example data, called "training data," to make predictions or decisions without being explicitly programmed to perform the task. The machine learning module can be implemented using a learning algorithm, such as supervised or unsupervised learning. The machine learning module can use various techniques, such as clustering, classification, linear regression, and reinforcement learning.A machine learning module can be, for example, a data structure or a program such as a neural network, in particular a convolutional neural network (CNN), a support vector machine, a decision tree, a Bayesian network, etc. The machine learning module can be adapted, i.e., trained, to predict an unmeasured value. The trained machine learning module can thus be enabled to predict the unmeasured value as output from other known values ​​as input.

[0101] A machine learning module to be trained can be, for example, an untrained machine learning module, a pretrained machine learning module, or a partially trained machine learning module. The machine learning module to be trained can be an untrained module being trained from scratch. Alternatively, the machine learning module to be trained can be a pretrained or partially trained machine learning module. Generally, it is not necessary to start with an untrained machine learning module, for example, in deep learning. One can, for example, start with a pretrained or partially trained machine learning module. The pretrained or partially trained machine learning module may have been pretrained or partially trained for the same or a similar task.Using a pre-trained or partially trained machine learning module can, for example, enable faster training of the machine learning module being trained; that is, the training can converge more quickly. Transfer learning, for instance, can be used to train a pre-trained or partially trained machine learning module. Transfer learning refers to a machine learning module where the learning process doesn't start from scratch but builds upon patterns already learned from solving another problem. This allows, for example, the use of previous learning experiences without having to start from zero. A pre-trained machine learning module is a machine learning module that has been previously trained, for example, on a large benchmark dataset, to solve a problem similar to the one to be solved through the additional learning.In the case of a pre-trained machine learning module, a previous learning process has been successfully completed. A partially trained machine learning module is a machine learning module that has been partially trained, meaning the learning process may not yet be complete. A pre-trained or partially trained machine learning module can, for example, be imported and trained for use for the purposes disclosed here. LIST OF REFERENCE MARKS 102 Computer unit 104 processor units 106 Hardware interface 108 User interface 110 storage units 120 machine-readable program instructions 122 Module for Machine Learning 124 recurrent neural network 126 Pre-processing module 128 Post-processing module 130 first reflection data 132 second reflection data 134 Crosstalk data 136 first noise data 138 second noise data 140 deviation 142 parameter values 160 Arrangement of radar sensors 162 Radar sensor 164 Radar sensor 166 distance 168 inclination QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited non-patent literature

[0000] Xiaoru Xie et al., “Fast-ABC: A Fast Architecture for Bottleneck-Like Based Convolutional Neural Networks,” 2019 IEEE Computer Society Annual Symposium on VLSI (ISVLSI), pages 1 to 6

[0044] Olaf Ronneberger, „U-Net: Convolutional Networks for Biomedical Image Segmentation“ (https: / / arxiv.org / abs / 1505.04597

[0045] Kaiming He et al., „Deep Residual Learning for Image Recognition“ (https: / / arxiv.org / abs / 1512.03385

[0046]

Claims

[1] Computer-implemented method for dynamically determining deviations (140) of a relative arrangement (160) of at least a first radar sensor (162) and a second radar sensor (164) to each other, the method comprising: • Receiving first reflection data (130) of a first reflection signal determined using the first radar sensor (162), which results from reflections of a first radar signal emitted by the first radar sensor (162) at one or more surrounding objects and comprises distance information about one or more distances of the first radar sensor (162) to the one or more surrounding objects, • Receiving second reflection data (132) of a second reflection signal determined using the second radar sensor (164), which results from reflections of a second radar signal emitted by the second radar sensor (164) at one or more surrounding objects and includes distance information about one or more distances of the second radar sensor (164) to the one or more surrounding objects, • Receiving at least crosstalk data (134) of a crosstalk signal determined using the first radar sensor (162), which results from a crosstalk of the second radar signal of the second radar sensor (164) to the first radar sensor (162), • receiving first noise data (136) of a first noise signal of an ambient noise of the first sensor determined using the first radar sensor (162), • receiving second noise data (138) of a second noise signal of an ambient noise of the second sensor (164) determined using the second radar sensor (164), • Determining a deviation (140) of the relative arrangement (160) of the first and second radar sensors (162, 164) using the first reflection data (130), the second reflection data (132), the crosstalk data (134), the first noise data (136) and the second noise data (138). [2] The method according to claim 1, wherein the deviation (140) of the relative arrangement (160) comprises a deviation (140) of a relative positioning and / or a relative orientation of the at least two radar sensors (162, 164) to one another. [3] Method according to one of the preceding claims, wherein the method further comprises dynamically adapting a combined evaluation of third radar signals received using the at least two radar sensors (162, 164), which comprises dynamically adapting one or more parameter values ​​(142) used for the combined evaluation, which describe the relative arrangement (160) of the at least two radar sensors (162, 164). [4] The method of claim 3, wherein the one or more parameter values ​​(142) of the relative arrangement of the at least two radar sensors (162, 164) comprise a distance (166) between the at least two radar sensors (162, 164) and / or a relative inclination (168) of the at least two radar sensors (162, 164) to each other. [5] The method of any preceding claim, wherein determining the deviation (140) of the relative arrangement (160) is performed using a machine learning module trained to receive the first reflection data (130), the second reflection data (132), the crosstalk data (134), the first noise data (136) and the second noise data (138) as inputs and to output the deviation (140) of the relative arrangement (160) of the first and second radar sensors (162, 164) as output. [6] Method according to claim 5, wherein the machine learning module (122) comprises at least one neural network, in particular at least one recurrent neural network (124). [7] Method according to one of the preceding claims, wherein a time course of the deviation (140) of the relative arrangement (160) of the first and second radar sensors (162, 164) is determined. [8] Method according to claim 7, wherein a periodically repeating pattern of the temporal course of the deviation (140) is determined, wherein the determined deviation (140) is used as part of the periodically repeating pattern for a periodically repeating dynamic adaptation of the combined evaluation of the third radar signals received using the at least two radar sensors (162, 164). [9] Computer program for dynamically determining deviations (140) of a relative arrangement (160) of at least a first radar sensor (162) and a second radar sensor (164) to each other, which comprises machine-readable program instructions (122), wherein execution of the machine-readable program instructions (122) by a processor unit (104) of a computer unit (102) causes the processor unit (104) to control the computer unit (102) to: • Receiving first reflection data (130) of a first reflection signal determined using the first radar sensor (162), which results from reflections of a first radar signal emitted by the first radar sensor (162) at one or more surrounding objects and includes distance information about the one or more distances of the first radar sensor (162) to one or more surrounding objects, • Receiving second reflection data (132) of a second reflection signal determined using the second radar sensor (164), which results from reflections of a second radar signal emitted by the second radar sensor (164) at one or more surrounding objects and includes distance information about one or more distances of the second radar sensor (164) to the one or more surrounding objects, • Receiving at least crosstalk data (134) of a crosstalk signal determined using the first radar sensor (162), which results from a crosstalk of the second radar signal of the second radar sensor (164) to the first radar sensor (162), • receiving first noise data (136) of a first noise signal of an ambient noise of the first sensor determined using the first radar sensor (162), • receiving second noise data (138) of a second noise signal of an ambient noise of the second sensor determined using the second radar sensor (164), • Determining a deviation (140) of the relative arrangement (160) of the first and second radar sensors (162, 164) using the first reflection signal, the second reflection signal, the crosstalk signal, the first noise signal and the second noise signal. [10] Computer unit (102) for dynamically determining deviations (140) of a relative arrangement (160) of at least a first radar sensor (162) and a second radar sensor (164) to one another, wherein the computer unit (102) comprises a processor unit (104) and a memory unit (110) with machine-readable program instructions (122), wherein execution of the machine-readable program instructions (122) by the processor unit (104) causes the processor unit (104) to control the computer unit (102) to: • Receiving first reflection data (130) of a first reflection signal determined using the first radar sensor (162), which results from reflections of a first radar signal emitted by the first radar sensor (162) at one or more surrounding objects and includes distance information about the one or more distances of the first radar sensor (162) to one or more surrounding objects, • Receiving second reflection data (132) of a second reflection signal determined using the second radar sensor (164), which results from reflections of a second radar signal emitted by the second radar sensor (164) at one or more surrounding objects and includes distance information about one or more distances of the second radar sensor (164) to the one or more surrounding objects, • Receiving at least crosstalk data (134) of a crosstalk signal determined using the first radar sensor (162), which results from a crosstalk of the second radar signal of the second radar sensor (164) to the first radar sensor (162), • receiving first noise data (136) of a first noise signal of an ambient noise of the first sensor determined using the first radar sensor (162), • receiving second noise data (138) of a second noise signal of an ambient noise of the second sensor determined using the second radar sensor (164), • Determining a deviation (140) of the relative arrangement (160) of the first and second radar sensors (162, 164) using the first reflection data (130), the second reflection data (132), the crosstalk data (134), the first noise data (136) and the second noise data (138).

Images