Method for checking artificial neural network for analyzing echo of motor vehicle detected by means of sensor

Abstract

A method for checking an artificial neural network (9) for analyzing echoes (8) of a motor vehicle (1) detected by a sensor (2). The method comprises: emitting (S1), by means of the sensor (2), at least two pulses (7) which differ from each other in at least one characteristic; detecting (S2) an echo (8) for each emitted pulse (7) by means of the sensor (2); and, for each detected echo (8), determining (S3) a term of result information (11) by applying an artificial neural network (9) to the detected echo (8), where the term of result information (11) describes a result of a particular problem (10) that the artificial neural network (9) has been trained to solve. It is then checked (S4) whether the determined result information (11) describes the same result, and the analysis carried out by the artificial neural network (9) is regarded as reliable (S5) only if so.

Description

Technical Field

[0001] The invention relates to a method for testing an artificial neural network for analyzing echoes of a motor vehicle detected by means of a sensor. The invention also relates to a motor vehicle having the sensor, a control device for the motor vehicle, and a computer program product for carrying out the method. Background Art

[0002] A motor vehicle may include at least one sensor whose measuring principle is based on the fact that the sensor transmits at least one pulse or pulses into the sensor's environment and then detects an echo, which is generated, for example, by reflection of the pulse from an object in the environment. Examples of sensors based on this pulse emission measuring principle are ultrasonic sensors, radar devices, and / or lidar devices. The echoes detected and provided by the sensor can be analyzed, for example, to classify objects in the environment. This analysis can be based on machine learning methods, in particular artificial neural networks. The results of the analysis can then be provided, for example, to a function of the motor vehicle, such as a driver assistance system for at least partially automated, in particular fully automated, control of the motor vehicle.

[0003] Document DE 10 201 9 218 349 A1 discloses a method for classifying at least one ultrasound echo from an echo signal. In this method, a classified echo image based on the received echo signals is generated with the aid of a trained neural encoder-decoder network and an input matrix formed from a plurality of detected echo signals.

[0004] When analyzing at least one detected echo using an artificial neural network, it has not been possible to reliably determine whether the result of the classification performed by the artificial neural network is actually correct or whether it returns an incorrect result, for example due to noise. Therefore, the artificial neural network should be checked in order to reliably exclude negative influences on the results, for example due to noise. Summary of the Invention

[0005] The object of the present invention is to provide a solution by means of which an echo analysis performed by an artificial neural network can be reliably verified.

[0006] This object is achieved by the subject-matter of the independent claims.

[0007] A first aspect of the invention relates to a method for testing an artificial neural network for analyzing echoes detected by a motor vehicle using a sensor. For example, the sensor is an ultrasonic sensor. Alternatively or additionally, the sensor may be a radar device and / or a lidar device. The artificial neural network may be a neural network of any design that has at least been trained to perform classification of data provided to it that is relevant for a motor vehicle-related task. The data provided to it describe at least one echo detected by the sensor, the at least one echo being received from the environment of the sensor and thus detected. The environment of the sensor preferably corresponds to at least a part of the environment of the motor vehicle in which the sensor is arranged. Known methods for training and for designing the structure of the artificial neural network can be used.

[0008] The method includes emitting at least two pulses by means of a sensor, the at least two pulses differing from one another in at least one characteristic. If the sensor is, for example, an ultrasonic sensor, the at least two ultrasonic pulses are emitted by the ultrasonic sensor. Preferably, multiple pulses, i.e., more than two pulses, are emitted consecutively. Alternatively or additionally, the sensor may emit pulses. For the purposes of the present invention, a pulse may alternatively or additionally be understood as an impulse. The pulses are generated by the sensor. For example, a characteristic that may differ between the emitted pulses is frequency. Thus, for example, it is possible to emit a first pulse at 50 kHz and then a second pulse at, for example, 52 kHz. These frequencies are to be understood purely as examples. Different variations in the characteristics of the at least two pulses, or even multiple characteristics, are possible.

[0009] In a further method step, the sensor detects a corresponding echo for each transmitted pulse. For example, if a pulse reflects off an object in the sensor's surroundings, a corresponding echo signal returns, referred to herein as a detected echo. For example, if exactly two pulses are transmitted, two echoes are detected: one for the first pulse and one for the second pulse. The more pulses transmitted, the more echoes are detected in each case.

[0010] An item of result information is obtained for each detected echo. This is performed by applying an artificial neural network to the detected echoes. Thus, if exactly two echoes are detected, both echoes are analyzed by the artificial neural network. The result information obtained describes the results of the specified problem that the artificial neural network has been trained to solve. For example, the problem might be to classify objects. In this example, the problem to be solved by the artificial neural network is to which category an object in an environment should be assigned. The result determined by the network is then one of the known categories, i.e., the category to which the object is assigned. It is assumed that even for multiple pulses that differ from one another in at least one characteristic, the same result, and therefore consistent result information, will still be obtained from the analysis of their echoes by means of the artificial neural network. Items of result information are considered consistent if they describe the same result, at least in terms of content. Therefore, it is expected that if the artificial neural network is able to reliably analyze the echoes, all obtained results will be consistent.

[0011] The method also includes checking whether the items of result information obtained describe the same result. Only if this is the case, that is, only if the items of result information obtained describe the same result, is the analysis of the echoes detected by the sensor using the artificial neural network considered reliable. Therefore, to verify the reliability of the analysis using the artificial neural network, it is ultimately proposed to vary the artificially transmitted pulse so that the results of the corresponding echoes can be compared. Since noise occurs statistically, while the detected echo is always a function of the transmitted pulse, noise can be easily identified. If the items of result information obtained are consistent despite variations in the transmitted pulse, this can be considered evidence that the results of the artificial neural network are reliable, regardless of, for example, ambient noise and / or other noise sources. This assumes that ambient noise or other noise sources, for example, will be located within a specific frequency range. By varying the characteristics of the pulses during the echo analysis, it is possible to determine whether the received echo is primarily or exclusively due to noise or whether it contains information about the sensor's environment. Thus, the analysis of the corresponding echoes performed using the artificial neural network is reliably verified.

[0012] An advantageous exemplary embodiment provides that the resulting information items obtained are provided to at least one function of the motor vehicle only if the analysis performed by the artificial neural network is assessed as reliable. Preferably, the function to which the information is provided is a driver assistance system. For example, a driver assistance system of the motor vehicle designed for at least partially automated, in particular fully automated, driving, which may control at least the longitudinal and / or lateral guidance of the motor vehicle, can then be operated taking into account the resulting information items. For example, the function can rely on the classification provided by the artificial neural network and consider it reliable and accurate information, for example, for its own application. This ensures that only truly reliable analysis results influence other functions of the vehicle.

[0013] Another exemplary embodiment provides that the at least one characteristic that distinguishes the emitted pulses from one another is one of the following: pulse duration, frequency, shape, and / or a code imprinted on the pulses. For example, the pulse length can be varied so that the pulse duration of a first pulse, for example, 50 microseconds, is doubled, resulting in a second pulse having a pulse duration of, for example, 100 microseconds. For the additional second pulse, a pulse duration twice as long or equal to the pulse duration distance between the first two pulses can also be selected, meaning that in this example, the third pulse can have a pulse duration of 200 microseconds or 150 microseconds. If the frequency varies, the frequency variation can be specified, for example, in steps of 1 kHz to 2 kHz between each pulse. Regarding the shape, for example, the pulse width can be varied, in particular deformed. Furthermore, a predetermined information item in the form of a code can be imprinted on the pulses. Alternatively or additionally, the sweep, or sampling rate, can be varied to specify, for example, the frequency and / or pulse duration differences of the emitted pulses within a frequency range or pulse duration range. For example, the sweep can specify the emission of a certain number of additional pulses, each with a specific deviation from the output pulse. The deviation between any two of the additional pulses is preferably always of the same magnitude. In addition to the aforementioned characteristics, variations in these characteristics are possible. Known methods for varying pulse characteristics can be used. Ultimately, various characteristics that characterize the pulses emitted by the sensor can be used. It is crucial that the at least two emitted pulses differ from one another in at least some manner. This allows the characteristics to be influenced in a particularly versatile manner.

[0014] A preferred exemplary embodiment provides that an F-measure value is determined for each item of result information, and then the F-measure value is checked to determine whether it is less than a minimum value. If so, at least one additional pulse is transmitted that differs from at least two previously transmitted pulses in at least one characteristic. Result information is also obtained and checked for the echo detected for the additional pulse. This checking process verifies whether the result information obtained for the further detected echo describes the same result as the item of result information obtained for the echo previously detected after the transmission of the at least two pulses. The F-measure value is, in particular, the F1-measure value. The F-measure value provides a method for evaluating measurement values. For example, the F-measure value is the harmonic mean of accuracy and hit rate, while in the F1-measure value, accuracy and hit rate are given equal weight. The F-measure value is generally dependent on the signal-to-noise ratio of the echo. Therefore, the F-measure value is a measure of the reliability of the artificial neural network. If the classification (i.e., the result information obtained using the artificial neural network) is correct, the F-measure value, and in particular the F1-measure value, is exactly equal to 1. However, if the F-measure value, in particular the F1-measure value, is not equal to 1, for example, is less than 1, this indicates that the result information is not completely reliable. If it is now determined that the F-measure value (particularly the F1-measure value) is not high enough, a new pulse is emitted, thereby increasing the number of echoes analyzed. For example, a minimum value specified for this purpose could be 0.9. Larger or smaller minimum values ​​are possible. The fact that at least one additional pulse is emitted is based on the observation that the measurement results, and in particular the signal-to-noise ratio of the measurement, can be improved by a factor equal to the square root of the number of measurements performed. As the number of measurements increases, the error and noise level decrease. For this reason, based on the F-measure value, it is possible to determine whether additional echoes should be analyzed or whether enough echoes have already been analyzed. The F-measure value is typically determined by an artificial neural network, making it very easy to determine and analyze.

[0015] It can be provided that if the F measured value is less than a minimum value, the entire measurement is always repeated, i.e., the previously acquired echo is discarded and pulses are retransmitted, with an increased number of pulses compared to the previous measurement. The described method for determining the reliability of the analysis is then performed using the newly detected echoes of the newly transmitted pulses. In other words, in repeated measurements, the artificial neural network can be checked repeatedly with increasing numbers of transmitted pulses.

[0016] Furthermore, exemplary embodiments may provide that the artificial neural network has been trained to solve multiple problems. For example, it may have been trained for two, three, or even more different problems. Items of result information are obtained and checked for each of the multiple problems. This means that for each problem, the individual items of result information from the multiple echoes are checked to see whether they all describe the same result. If, during the checking of at least one of the multiple problems, it is determined that the items of result information obtained do not describe the same result, the check is repeated, with a greater number of pulses being emitted for the repeated check. If the result for at least one problem indicates that the corresponding analysis performed by the artificial neural network is not reliable enough, the entire method is preferably repeated, i.e., the artificial neural network is checked for all problems, even those that already provided consistent results. In this new and therefore repeated check, the number of pulses emitted is also increased to obtain even more accurate results. For example, the analysis performed by the artificial neural network is then checked again to see whether it is actually not reliable enough, for example, because only noise sources were previously present. Prior to the recheck, the artificial neural network can be retrained, in particular for artificial neural networks that are not reliably solving the problem. This provides a readily implementable means for rechecking artificial neural networks that have previously been assessed as unreliable.

[0017] Another exemplary embodiment provides for the artificial neural network to be checked repeatedly at predetermined time intervals during the active state of the sensor. For example, the predetermined time intervals are 1 minute, 2 minutes, 3 minutes, 5 minutes, or in particular 10 minutes. Larger or smaller time intervals, and in particular time intervals between the aforementioned time intervals, are possible. Thus, it can be provided that during operation of the motor vehicle, during which echoes are received and analyzed by the sensor, and during which, for example, the motor vehicle's functionality is effective based on the provided echoes, a change in at least two pulses occurs repeatedly, for example once per minute, and is analyzed according to the above-described method to check whether the artificial neural network continues to deliver reliable results. This means that an additional safety check can be provided, in particular for the functionality of a vehicle designed for at least partially autonomous operation.

[0018] Furthermore, according to one exemplary embodiment, the predetermined time interval between two checks is determined by the current speed of the motor vehicle on which the sensor is mounted. For example, it can be provided that during highway travel, such as when traveling in an urban area or on rural roads, the artificial neural network is checked at shorter intervals than during lower highway travel. This assumes that at high vehicle speeds, even higher reliability of the artificial neural network is required compared to lower speeds, particularly if the motor vehicle is controlled at least in part based on the echoes detected by the sensor. For this reason, the frequency of checks depends on the current journey and, therefore, on the speed. This also saves resources, as checks are only performed at short intervals when particularly relevant, and at longer intervals otherwise.

[0019] The speed of the motor vehicle can be detected and provided, for example, by means of a speed sensor of the motor vehicle.

[0020] Furthermore, one exemplary embodiment provides that each item of result information at least describes whether the detected echo is real or noise. The first task of the artificial neural network can therefore be to classify whether an echo or noise is present. In other words, the artificial neural network may have been trained to distinguish between data obtained that is typical of noise and data obtained that describes the echo of the transmitted pulse. This assumes that noise is received by the sensor in the same manner as an echo and is therefore detected. The artificial neural network can thus analyze the data provided to it by the sensor to determine whether it is typical of a noise spectrum or typical of an echo. This is a particularly useful problem for the artificial neural network, for example, in order to decide whether further analysis is useful if an echo is actually detected, or not if only noise is detected.

[0021] Furthermore, one exemplary embodiment provides that each item of result information at least describes whether the detected echo describes an object relevant to the intended target application. Thus, a second question for the artificial neural network can be to classify whether the echo (particularly an echo that has been classified in this manner) describes an object relevant to the intended target application or whether the object described by the echo is irrelevant to the intended target application. This allows for determining whether the object is relevant for further analysis in the motor vehicle. It can be assumed that the echo is not noise, but rather a real echo describing an object actually located in the environment. The target application is, for example, a function of the motor vehicle and / or the object class of the object. This allows, for example, using a further neural network, to directly determine whether the object is another road user. This is a useful question for the artificial neural network, for example, in order to decide whether further analysis of the echo should be performed.

[0022] In an exemplary embodiment, it can also be provided that the corresponding result information at least describes whether the detected echo describes an object with a height and / or width that is less than a predetermined minimum value. Therefore, the third problem of the artificial neural network can be to classify whether the height and / or width of the object described by the echo is less than the minimum value on the one hand and greater than or equal to the minimum value on the other hand. Different minimum values ​​can be specified for height and width. For example, the minimum height and / or width can be 20 centimeters. It is possible that it is greater or less than the minimum value of this example. This problem is particularly suitable for determining, for example, whether an object in the environment of the sensor is relevant to the function of a motor vehicle, in particular to the driving function. For example, it can be assumed that the object less than the minimum value is a speed bump, a curb and / or some other object that can be driven over. Therefore, classifying objects according to their height and / or width may be particularly useful.

[0023] Furthermore, one exemplary embodiment provides that the corresponding result information item also includes whether the detected echo describes an object having a height and / or width less than a predetermined maximum value. The maximum value is greater than the minimum value. The maximum value can be, for example, 60 centimeters. Alternatively, the maximum value can be any value greater than or less than 60 centimeters. Different minimum values ​​can be specified for height and width. The evaluation of the object relative to the maximum value preferably results in a third question being provided in another category. Thus, it can now be determined whether the object is less than the minimum value, greater than or equal to the minimum value but less than or equal to the maximum value, or greater than the maximum value. This is used to accurately assess whether the object could constitute an obstacle for the motor vehicle and should therefore, for example, be considered by the function.

[0024] Furthermore, exemplary embodiments can be provided that include the fact that the corresponding result information item also includes whether the detected echo describes an object with a height and / or width between a predetermined minimum value and a predetermined maximum value. In the above example, it is thus possible to check whether the object's height and / or width is between 20 centimeters and 60 centimeters, or whether the height and / or width is outside this value range. In particular, an intermediate value is specified between the minimum and maximum values. For example, the intermediate value could be 40 centimeters. The result information can then indicate whether the received echo describes an object with a height and / or width between the minimum and intermediate values, or whether the height and / or width is between the intermediate value and the maximum value. Ultimately, this adds a fourth category to the third question, allowing determination of whether the height and / or width is less than the minimum value, greater than or equal to the minimum value but less than or equal to the intermediate value, greater than the intermediate value but less than or equal to the maximum value, or greater than the maximum value. A different intermediate value can be assigned to the width than to the height. Ultimately, this allows for a particularly clear and specific classification of the object's height into different categories, which can, for example, be associated with different object classes or types in subsequent analysis steps. This means that, for example, for an ultrasonic sensor as a sensor, based on its measurement data, a description of the sensor environment in which the detected objects are evaluated can be generated at least with respect to its spatial range.

[0025] Another aspect of the present invention relates to a motor vehicle having a sensor. The motor vehicle is designed to transmit at least two pulses differing from one another in at least one characteristic using the sensor, and to detect echoes of the corresponding transmitted pulses using the sensor. Furthermore, the motor vehicle is designed to obtain an item of result information for each detected echo by applying an artificial neural network to the detected echoes, wherein the result information describes the result of a predetermined problem that the artificial neural network has been trained to solve; and to check whether the obtained items of result information describe the same result. If so, the motor vehicle is designed to assess the analysis performed by the artificial neural network as reliable.

[0026] The motor vehicle is, for example, a passenger car, truck, bus, motorcycle, and / or moped. The sensor is preferably an ultrasonic sensor. Alternatively or additionally, the sensor is a radar device and / or a lidar device. The sensor is preferably arranged in the front and / or rear area of ​​the motor vehicle, for example, in the bumper. Alternatively or additionally, the sensor may be arranged, for example, on the windshield, rear window, in a door, and / or in a rearview mirror of the motor vehicle. Alternatively or additionally, the sensor is located in the interior of the motor vehicle.

[0027] The present invention also relates to a control device for a motor vehicle. This device is designed to perform the above-described method. In this context, it is assumed that the control device can provide a command to a sensor, which, for example, causes the sensor to emit the at least two pulses and provide the echoes detected by the sensor to the control device. The artificial neural network is preferably stored in a memory unit of the control device and can be applied by the device to the echoes provided by the sensor. The control device performs the described method, in particular, an exemplary embodiment or a combination of exemplary embodiments of the method. The control device includes a processor device. The processor device may include at least one microprocessor and / or at least one microcontroller and / or at least one FPGA (field programmable gate array) and / or at least one DSP (digital signal processor). Furthermore, the processor device may contain program code, which may alternatively be referred to as a computer program product. The program code may be stored in a data memory of the processor device.

[0028] The present invention further relates to a computer program product. The computer program product is designed to perform an inspection at least with the aid of an artificial neural network. To this end, the computer program product preferably receives detected echoes from a sensor. Thus, the computer program product includes at least the steps according to the present invention, which are not exclusively performed by the sensor.

[0029] The exemplary embodiments described in conjunction with the method according to the invention, whether individually or in combination with one another, apply accordingly, to the extent applicable, to the motor vehicle according to the invention, the control device according to the invention, and the computer program product according to the invention. The present invention includes combinations of the exemplary embodiments described. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic diagram showing a motor vehicle having a plurality of sensors; and

[0031] Figure 2 A schematic diagram of a method for testing an artificial neural network is shown for the purpose of analyzing echoes of a motor vehicle detected by means of sensors. DETAILED DESCRIPTION

[0032] Figure 1A motor vehicle 1 is shown which has a plurality of sensors 2 in the front and rear areas. The sensors 2 are arranged in the bumper of the motor vehicle 1. Six sensors 2 are shown in the front area and six sensors 2 in the rear area, for example. The motor vehicle 1 can have more or fewer sensors 2 than those shown here. The respective sensors 2 are preferably ultrasonic sensors. Alternatively or in addition, the sensors 2 are radar devices, in other words radar sensors, and / or LIDAR devices. Finally, the sensors 2 are pulse-transmitting sensors 2 whose measuring principle is based on receiving echoes 8 (see Figure 2 8 in FIG) as a pulse 7 emitted, for example (see Figure 2 7 ) of the reflections and / or pulses.

[0033] The motor vehicle 1 has a control device 3, which is, for example, a central processing unit of the motor vehicle 1. The control device 3 is preferably designed to control the corresponding sensors 2 of the motor vehicle 1, i.e., for example, to provide them with control commands and then receive from the sensors 2, for example, echoes 8 detected by the sensors 2. The control device 3 can also be designed to provide functions 4 for the motor vehicle 1, such as a driver assistance system for at least partially automatic, in particular fully automatic, driving of the motor vehicle 1.

[0034] The motor vehicle 1 may comprise an output device 5 which is used, for example, to output a test artificial neural network 9 (see Figure 2 The output device 5 can be, for example, a screen, in particular a touch-sensitive screen, and / or a loudspeaker of the motor vehicle 1. The output device 5 can, for example, indicate whether the function 4 can currently be activated.

[0035] Figure 2 The steps of a method for testing an artificial neural network 9 for the purpose of analyzing an echo 8 of a motor vehicle 1 detected by means of a sensor 2 are shown. In method step S1, at least two pulses 7 are emitted by means of the sensor 2, said pulses differing from one another in at least one characteristic. The characteristics by which the at least two pulses 7 differ from one another are, for example, pulse duration, frequency, shape, and / or a code printed on the pulses 7.

[0036] It is also outlined here that two pulses 7 are emitted in the direction of an object 6, which is located in the environment of the sensor 2. The environment of the sensor 2 corresponds here to at least a part of the environment of the motor vehicle 1 in which the sensor 2 is arranged.

[0037] In method step S2 , a corresponding echo 8 is detected for each emitted pulse 7 . Here, for example, exactly two echoes 8 are shown, which are caused by two emitted pulses 7 and which move from the object 6 in the direction of the sensor 2 and are detected by the sensor 2 .

[0038] In method step S3, artificial neural network 9 is applied to each of the detected echoes 8. In this process, an item of result information 11 is obtained. Result information 11 describes the results of a predetermined question 10 that artificial neural network 9 has been trained to solve. The artificial neural network 9 specified here may, for example, include three questions 10, i.e., may have been trained to solve three questions 10. A first question 10 specifies whether the obtained result information 11 describes at least whether the detected echo 8 is real or noise. A second question 10 may, for example, specify whether the corresponding item of result information 11 describes whether the detected echo 8 describes an object 6 relevant to a predetermined target application of motor vehicle 1. Object 6 may, for example, be relevant to the target application if it is another motor vehicle 1, a pedestrian, a cyclist, and / or an obstacle to motor vehicle 1. Another possible question 10 may specify whether the corresponding item of result information 11 describes at least whether the received echo 8 describes an object 6 having a height and / or width that is less than a corresponding predetermined minimum value, or greater than or equal to the minimum value but less than or equal to a median value, or greater than the median value but less than or equal to a maximum value, or greater than the maximum value. For the width, different minimum values, maximum values, and / or median values ​​can be specified than for the height. The minimum value is smaller than the median value, which in turn is smaller than the maximum value. Thus, the third question 10 can distinguish, for example, four different categories, so that by applying the artificial neural network 9, it can be checked to which of these four height-related and / or width-related categories the object 6 should be assigned. In the first two questions 10 mentioned, only two possible categories were distinguished, namely, for example, the category "real echo" and the category "noise," or the category "object relevant to the specified target application" and the category "object not relevant to the specified target application." Each question 10 can be solved by a common artificial neural network 9, or a separate artificial neural network 9 can exist for each question 10. In this case, items of result information 11 can be obtained for multiple artificial neural networks 9 and then, for example, compared with each other. The artificial neural network 9 can include even more questions 10 than the examples mentioned.

[0039] In method step S4, a check is performed to determine whether the items of result information 11 obtained describe the same result. This means, for example, whether two echoes 8 each produce the "echo true" answer to first question 10. If this is the case, i.e., if all items of result information 11 describe the same result, then in method step S5, the analysis performed by artificial neural network 9 is assessed as reliable. It is therefore determined that artificial neural network 9 is reliably suitable for analyzing and classifying echoes 8.

[0040] If this has been evaluated in this way in method step S5, the items of result information 11 obtained can be provided in method step S6 to a function 4 of the motor vehicle 1. Thereafter, for example, the items of result information 11 and / or echoes 8 obtained by the sensor 2 and optionally analyzed by the artificial neural network 9 can be used to operate a driver assistance system as function 4.

[0041] Provision can be made for determining an F-measured value, in particular an F1-measured value, for the result information 11. A check is then performed to determine whether the determined F-measured value is less than a minimum value. If this is the case, at least one further pulse 7 is emitted that differs from at least two previously transmitted pulses 7 in at least one characteristic. Result information 11 is also obtained and checked for the echoes 8 detected for the further pulses 7. If, for example, a higher F-measured value is achieved for the further pulses 7 by analyzing the corresponding echoes 8, this can result in the evaluation of the artificial neural network 9 being assessed as reliable.

[0042] Method step S7 can also be provided, assuming that the artificial neural network has already been trained to solve a plurality of problems 10. Items of achievement information 11 are then obtained for the plurality of problems 10. A check is then performed to see whether the items of result information 11 obtained for at least one of the plurality of problems 10 do not describe the same result. In this case, the check is performed again, wherein a greater number of pulses 7 is emitted than before. Thus, in method step S8, five pulses 7 are emitted, purely as an example, and five echoes 8 are thus received, which are then provided to the artificial neural network 9 for solving the respective problems 10. This results in, for example, five items of result information 11 being obtained for each problem 10, which can then be compared with one another regarding their results. If this greater number of pulses 7 ultimately leads to the situation where all problems 10 can be reliably solved and the artificial neural network 9 is therefore assessed as reliable for each problem 10, the method can be terminated and the assessment in method step 5 updated. Otherwise, it can be provided that, for example, the entire neural network 9 is assessed as unreliable because it was determined that at least one of the problems 10 of the artificial neural network 9 could not be reliably solved.

[0043] In general, it is possible to repeat the described checks, i.e., at least method steps S1 to S9, continuously while, for example, sensor 2 is active. Preferably, the checks are repeated at predetermined time intervals, such as 1 minute, 2 minutes, 3 minutes, 5 minutes, or in particular 10 minutes. The time interval between two checks can depend on the current speed of motor vehicle 1. For example, the higher the speed, the more frequent the checks can be performed.

[0044] In summary, the example shows the quality inspection of a pulse distance measuring sensor.

[0045] An additional and novel solution involves working with different echo forms of echo 8. The principle proposed here involves performing measurements with different echo forms. For example, varying the frequency, duration, shape, sweep, etc. While the noise is statistical, the received echo 8 is a function of the transmitted pulse (pulse 7). This allows for unique classification after multiple measurements (F1-measurement = 1). The number of measurements and the variation in the transmitted pulse are performed by concurrently measuring the signal-to-noise ratio. This proposed solution can be applied to any type of pulse-echo detection sensor.

Claims

1. A method for checking an artificial neural network (9) for analyzing an echo (8) of a motor vehicle (1) detected by means of a sensor (2), the method comprising: - emitting (S1) at least two pulses (7) by means of the sensor (2), the at least two pulses (7) differing from one another in at least one characteristic; - detecting ( S2 ) the echo ( 8 ) of each emitted pulse ( 7 ) by means of the sensor ( 2 ); and - for each detected echo (8), obtaining (S3) an item of result information (11) by applying said artificial neural network (9) to the detected echo (8), wherein said result information (11) describes the result of a specified problem (10) that said artificial neural network (9) has been trained to solve; It is characterized in that It is checked (S4) whether the items of result information (11) obtained describe the same result, and only if this is the case it is assessed (S5) that the analysis performed by the artificial neural network (9) is reliable.

2. The method according to claim 1, characterized in that Only if the analysis performed by the artificial neural network (9) is assessed as reliable, the item of result information (11) obtained is provided (S6) to at least one function (4) of the motor vehicle (1), in particular a driver assistance system of the motor vehicle (1).

3. The method according to any one of the preceding claims, characterized in that The at least one characteristic by which the emitted pulses (7) differ from one another is one of the following: - Pulse duration; - frequency; - shape; and / or - A code printed on said pulse (7).

4. A method according to any one of the preceding claims, wherein An F measurement value, in particular an F1 measurement value, is determined for each item of result information (11) and a check is performed to determine whether the determined F measurement value is less than a minimum value, wherein, if this is the case, at least one further pulse (7) is emitted, which differs from the at least two pulses (7) emitted so far in at least one characteristic, and the result information (11) of the echo (8) detected from the further pulse (7) is obtained and checked.

5. A method according to any one of the preceding claims, wherein The artificial neural network (9) has been trained to solve a plurality of problems (10), and items of result information (11) are obtained and checked for the plurality of problems (10), wherein if, for at least one of the plurality of problems (10), the obtained items of result information (11) do not describe the same result after the check, the check is performed again, wherein for the repeated check a greater number of pulses (7) is emitted than before.

6. The method according to any one of the preceding claims, characterized in that During the active state of the sensor (2), the checking of the artificial neural network (9) is repeated at predetermined time intervals.

7. The method according to claim 6, characterized in that The predetermined time interval between two checks depends on the current speed of the motor vehicle (1) on which the sensor (2) is arranged.

8. The method according to any one of the preceding claims, characterized in that Each item of the result information (11) at least describes whether the detected echo (8) is real or noise.

9. The method according to any one of the preceding claims, characterized in that Each item of result information (11) describes at least whether the detected echo (8) describes an object (6) that is relevant to the predetermined target application.

10. The method according to any one of the preceding claims, characterized in that The corresponding result information (11) describes at least whether the detected echo (8) describes an object (6) having a height and / or width smaller than a predetermined minimum value.

11. The method according to claim 10, wherein: Each item of result information (11) further includes whether the detected echo (8) describes an object (6) having a height and / or width greater than a predetermined maximum value, wherein the maximum value is greater than the minimum value.

12. The method according to claims 10 and 11, characterized in that The corresponding result information (11) also includes whether the detected echo (8) describes an object (6) having a height and / or width between a minimum value and a maximum value, in particular between a minimum value and an intermediate value, or between an intermediate value and a maximum value, wherein the intermediate value is between the minimum value and the maximum value.

13. A motor vehicle (1) having a sensor (2), characterized in that The motor vehicle (1) is designed to: - emitting at least two pulses (7) by means of the sensor (2), the at least two pulses (7) differing from one another in at least one characteristic; - detecting the echo (8) of each emitted pulse (7) by means of the sensor (2), - for each detected echo (8), an item of result information (11) obtained by applying an artificial neural network (9) to the detected echo (8), wherein the result information (11) describes the result of a specified problem (10) that the artificial neural network (9) has been trained to solve, - check that the items of the obtained result information (11) describe the same result, and - If this is the case, the analysis performed by the artificial neural network (9) is assessed as reliable.

14. A control device (3) for a motor vehicle (1), wherein: The control device (3) is designed to execute the steps provided for the control device (3) of the method according to any one of claims 1 to 12.

15. A computer program product comprising commands which, when executed by a control device (3) of a motor vehicle (1), cause the control device to carry out the steps of the method according to any one of claims 1 to 12 provided for the control device (3).

Images