DE102018219255A1 - Training system, data set, training method, evaluation device and deployment system for a road vehicle for recording and classifying traffic noise - Google Patents

Abstract

Training system (10) for a road vehicle (1) for detecting and classifying traffic noises (17), comprising an input interface (11) in order to receive actual training data (12), the actual training data (12) recording the traffic noises (17 ) that can detect oncoming traffic (2) while driving the road vehicle (1) on a mountain road with at least one microphone (3a, 3b, 3c, 3d) that can be arranged on the road vehicle (1), and each associated target identifier (18a) this oncoming traffic (2), and wherein the training system (10) is designed to feed an artificial neural network (13) with the actual training data (12) forward and in the forward feed with the artificial neural network (13 ) ascertained actual identifier (18b) of the respective oncoming traffic (2), and weighting factors (14) for connections (15) of neurons (16) of the artificial neural network factory (13) by feeding back a deviation (19) between the actual indicator (18b) and the target indicator (18a) to minimize the deviation (19) in order to learn the target indicator (18a) of oncoming traffic. The invention also relates to a data set, a training method, an evaluation device (20) and an application system (30).

Description

The invention relates to a training system according to claim 1 for a road vehicle for detecting and classifying traffic noise. Furthermore, the invention relates to a data record according to claim 2. Furthermore, the invention relates to a training method according to claim 3 for an artificial neural network for detecting and classifying traffic noise. Furthermore, the invention relates to an evaluation device according to claim 4 for a road vehicle for detecting and classifying traffic noise. The invention also relates to an application system for a road vehicle for detecting and classifying traffic noises.

Known radar, lidar and image sensors from driver assistance systems detect objects in a visible environment of a vehicle by means of corresponding electromagnetic waves. However, if an obstacle to these waves is arranged between such a sensor and the object, the object cannot be detected with these sensors.

On narrow, winding mountain roads in particular, oncoming vehicles are hidden until shortly before their appearance due to the tight bends and / or rocks. In such situations, human drivers often use horns to make themselves known. However, there is a risk that the horn will be activated too late. In general, operating the horn in critical driving maneuvers can be challenging for the driver and accidents cannot be avoided.

The DE 10 2012 218 482 A1 discloses a driver assistance system that detects ambient noise that hits the vehicle from the outside.

This is where the invention comes in. The invention is based on the object of providing detection and classification of oncoming traffic on mountain roads by means of traffic noise and of giving an indication of oncoming traffic.

• • ein Trainingssystem für ein Straßenfahrzeug zum Erfassen und Klassifizieren von Verkehrsgeräuschen mit den Merkmalen des Anspruchs 1,
• • einen Datensatz mit den Merkmalen des Anspruchs 2,
• • ein Trainingsverfahren für ein künstliches neuronales Netzwerk zum Erfassen und Klassifizieren von Verkehrsgeräuschen mit den Merkmalen des Anspruchs 3,
• • eine Auswerteeinrichtung für ein Straßenfahrzeug zum Erfassen und Klassifizieren von Verkehrsgeräuschen einer Verwendung eines erfindungsgemäßen Einsatzsystems als ein Fahrerassistenzsystem mit den Merkmalen des Anspruchs 4 und
• • ein Einsatzsystem für ein Straßenfahrzeug zum Erfassen und Klassifizieren von Verkehrsgeräuschen mit den Merkmalen des Anspruchs 7.

This task is solved by

• A training system for a road vehicle for detecting and classifying traffic noise with the features of claim 1,
• A data record with the features of claim 2,
• A training method for an artificial neural network for detecting and classifying traffic noise with the features of claim 3,
• • An evaluation device for a road vehicle for detecting and classifying traffic noises using an application system according to the invention as a driver assistance system with the features of claims 4 and
• An application system for a road vehicle for recording and classifying traffic noises with the features of claim 7.

Further developments and advantageous refinements are specified in the subclaims.

The training system according to the invention for a road vehicle is used to record and classify traffic noise. The training system includes an input interface to receive actual training data. The actual training data include a recording of the traffic noises that can be detected by oncoming traffic with at least one microphone that can be arranged on the road vehicle while the road vehicle is traveling on a mountain road. Furthermore, the actual training data each include target identifiers of this oncoming traffic belonging to the recording. The training system is designed to feed an artificial neural network forward with the actual training data and to receive in the forward feed an actual indicator of the respective oncoming traffic determined with the artificial neural network. In addition, the training system is designed to adapt weighting factors for connections of neurons of the artificial neural network by feeding back a deviation between the actual indicator and the target indicator to minimize the deviation in order to learn the target indicator of the oncoming traffic. Furthermore, the training system includes an output interface in order to provide the actual indicator.

The following definitions apply to the entire subject matter of the invention.

Road vehicles are land vehicles that can maintain or change their direction of travel by means of static friction on an accessible surface. Motor vehicles, in particular motor vehicles, such as, for example, passenger cars, commercial vehicles or motorcycles, are road vehicles.

Traffic noise from these road vehicles is a sound event that arises, among other things, from the static friction with the surface being driven on. Depending on the flooring, for example Gravel, asphalt or sand, of the type of contact of the road vehicle with the surface, e.g. number of wheels, and of the road vehicle itself, e.g. size of the engine, weight of the road vehicle, size of the road vehicle and / or driving condition of the road vehicle, in particular speed, Gear selection, braking or accelerating create typical traffic noises. The traffic noise depends on the weather. For example, the driving noise of a road vehicle on a wet surface is different from a dry surface or a surface covered with snow or ice. The object of the invention is designed to classify these traffic noises depending on the respective weather.

A mountain road is a road that leads to a mountain and / or runs in a mountain range and in particular has a pronounced height profile. Other features of a mountain road are, for example, many curves, U-turns, serpentines and / or at least some narrow lane widths.

An interface is a component between at least two functional units on which an exchange of logical quantities, e.g. Data or physical quantities, e.g. electrical signals, either unidirectional or bidirectional. The exchange can be analog or digital. The exchange can be wireless or wired. An interface can exist between software and software, hardware and hardware as well as software and hardware and hardware and software.

The microphones that can be arranged on the road vehicle are microphones that are suitable for automotive applications, that is to say in particular are correspondingly weatherproof, functionally reliable, and are wind and splash water resistant. These microphones preferably have filters and / or amplifiers in order to be particularly sensitive to the noise of oncoming traffic on mountain roads. For example, engine noises and / or braking noises from oncoming vehicles are amplified. At least one microphone is preferably arranged on each side of the road vehicle, that is to say at the front, rear, left and right, that is to say a specific arrangement of microphones. Alternatively, two microphones are arranged in the front area and in the rear area. The microphones are preferably connected to one another, that is to say the microphones are arranged in an array. The microphones are arranged in such a way that the arrangement optimizes the signal acquisition and / or signal processing of the microphones. For example, the microphones are arranged protected from the wind. The respective microphones are preferably directional microphones.

Actual training data are positive training data with which a learning mechanism, for example an artificial neural network, learns information, comparable to the learning process of the brain. Actual training data are labeled with the meaning of the information, that is, labeled, so that the learning mechanism can semantically record information. For example, an oncoming vehicle, that is, a downhill vehicle, usually drives in a low gear with a high engine speed in order to achieve an additional braking effect. The high engine speed is a target indicator for oncoming traffic. The target indicator is the real, given information. This target number labels the recording of the traffic noise of the high engine speed with the meaning that a downhill braking vehicle is approaching. The amplitude, frequency and other variables of the traffic noise are recorded with at least one microphone. The actual training data also include traffic noises from the same oncoming traffic under different conditions, for example when the road is covered with snow or when the road is wet and at different altitudes. For example, pass roads, mountain roads or other mountain roads have multiple turns and / or switchbacks. The straight sections of the mountain road between turns and / or serpentines are then on different floors in relation to the mountain. The training system is also used to teach in which floor, i.e. at what altitude, the oncoming traffic is located.

It is also within the scope of the invention that the training system has a second input interface in order to receive error training data. Error training data is negative training data through which the learning mechanism learns to react to an error.

An actual indicator is the indicator that the learning mechanism receives after processing the target information, that is, outputs it as an output. At the beginning of the training phase, the actual indicator received is usually different from the target indicator entered. In the training phase, the error between the actual and target indicators is minimized, for example using the least squares method. After the training phase, the learning mechanism, for example the artificial neural network, has completed the learning process and is trained. In the case of a trained artificial neural network, the actual indicators obtained are approximately the same as the target indicators.

Artificial intelligence is a generic term for the automation of intelligent behavior. For example, an intelligent algorithm learns to react appropriately to new information. An intelligent algorithm is, for example, an artificial neural network, referred to as an artificial neural network.

An artificial neural network is an algorithm that is executed on an electronic circuit and is modeled on the neural network of the human brain. As the developer also makes decisions based on the tests that influence the algorithm, at the end of the training phase another data set, the test data set, is used for a final evaluation.

Functional units of an artificial neural network are artificial neurons. An output of a neuron generally results as the value of an activation function evaluated via a sum of the inputs weighted with weighting factors plus a systematic error, the so-called bias. An artificial neural network with several hidden layers is a deep neural network.

A trained artificial neural network is characterized by responding appropriately to new information. Examples of artificial neural networks are perceptrons, convolutional or recurrent neural networks.

Weighting factors are used to evaluate connections between neurons. Forward feeding means that information is fed into the input layer of the artificial neural network, goes through the following layers and is output in the output layer. Backward feeding, referred to as backward propagation, means that information is input into the output layer and output in the input layer. By successively feeding the error back from the output layer into the previous layer up to the input layer, the errors of the respective layers are obtained. The errors are a function of the weighting factors. By minimizing the error in the training phase, the weighting factors are changed. This means that when the input is fed in again, the desired output is approximated. Reverse feeding is described in detail in Michael A. Nielsen, Neural Networks and Deep Learning, Determination Press, 2015.

The subject of the invention classifies traffic noises from oncoming traffic on mountain roads even before oncoming traffic is visible to the driver of the road vehicle or to camera, lidar and / or radar sensors of driver assistance systems. This is particularly advantageous if oncoming traffic is only visible shortly before the appearance due to a route rich in curves, in particular a mountain road between rocks. The traffic noises can be detected, however, if the oncoming traffic is still far away in front of your own road vehicle, for example hidden behind a curve.

With the object of the invention, the traffic noises of oncoming traffic can be detected and classified even under the most varied weather conditions.

The object of the invention further increases driving safety and driving comfort in that the driver of the road vehicle is warned of oncoming traffic before the oncoming traffic is visible. Since the driver is warned in good time by the subject matter of the invention, for example, severe braking and / or abrupt steering interventions, each of which increases the risk of an accident, are avoided. The object of the invention ensures that the driver does not have to perform critical driving maneuvers if he has seen the oncoming traffic too late.

The data record according to the invention consists of recordings of traffic noises that can be detected by oncoming traffic with at least one microphone that can be arranged on the road vehicle while a road vehicle is traveling on mountain roads, and the associated target number of the oncoming traffic. The data record is designed in such a way that an evaluation device can process a respective recording together with the associated target identifier in order to train an artificial neural network to classify the traffic noise as a function of the target identifier. The data record is characterized by the respective traffic noise under different conditions, for example under different weather conditions, in combination with the respective target number plate. This data record is advantageously designed to be made available to different architectures of artificial neural networks for the purpose of training.

• • Bereitstellen von während einer Fahrt mit einem Straßenfahrzeug aufgenommen Verkehrsgeräuschen, die während einer Fahrt des Straßenfahrzeuges auf einer Bergstraße von entgegenkommenden Verkehr mit wenigstens einem an dem Straßenfahrzeug anordenbaren Mikrofon erfassbar sind, als Ist-Trainingsdaten und jeweils zugehörige Soll-Kennzeichen dieses entgegenkommenden Verkehrs,
• • Vorwärtsspeisen des künstlichen neuronalen Netzwerks mit den Ist-Trainingsdaten,
• • Erhalten eines mit dem künstlichen neuronalen Netzwerk ermittelten Ist-Kennzeichens des jeweiligen entgegenkommenden Verkehrs,
• • Rückwärtsspeisen des künstlichen neuronalen Netzwerks mit der Abweichung zwischen dem ermitteltem Ist-Kennzeichen und zugehörigem Soll-Kennzeichen, und
• • Erhalten von Gewichtungsfaktoren für Verbindungen von Neuronen des künstlichen neuronalen Netzwerks in einer Rückwärtsspeisung zur Minimierung der Abweichung, um das Soll-Kennzeichen des entgegenkommenden Verkehrs zu lernen.

The training method according to the invention for an artificial neural network for recording and classifying traffic noises comprises the following method steps:

• Providing traffic noises recorded during a journey with a road vehicle, which can be detected by oncoming traffic with at least one microphone which can be arranged on the road vehicle while the road vehicle is traveling on a mountain road, as actual training data and respectively associated target identifiers of this oncoming traffic,
• Feeding the artificial neural network forward with the actual training data,
• Obtaining an actual identifier of the respective oncoming traffic determined with the artificial neural network,
• • Reverse feeding of the artificial neural network with the deviation between the determined actual indicator and the associated target indicator, and
• Obtaining weighting factors for connections of neurons of the artificial neural network in a reverse feed to minimize the deviation in order to learn the target indicator of the oncoming traffic.

The training system and the training method thus provide training for an artificial neural network in order to learn to record and classify traffic noises from oncoming traffic on mountain roads before the oncoming traffic starts from the road vehicle using the human eye, camera, radar, lidar or another imaging sensor is visible.

A training system according to the invention is advantageously used to carry out the training method.

It is also within the scope of the invention that the training system and / or the training method are used or carried out at different altitudes and / or countries and in different weather conditions. The artificial neural network thus learns traffic noise under different conditions and can thus record and classify more specifically oncoming traffic on mountain roads.

The evaluation device according to the invention for a road vehicle is used to record and classify traffic noises. The evaluation device comprises an input interface in order to receive the traffic noise of oncoming traffic on a mountain road. The input interface is in particular an interface to a microphone suitable for automotive use. The evaluation device is designed to feed a trained artificial neural network with these traffic noises forward. The artificial neural network is trained to record and classify oncoming traffic based on traffic noise. Furthermore, the evaluation device comprises an output interface in order to provide the detection and classification of the oncoming traffic.

An evaluation device is a device that processes incoming information and outputs a result resulting from this processing. In particular, an electronic circuit, such as a central processor unit or a graphics processor, is an evaluation device.

The output interface is advantageously a human-machine user interface, in English referred to as a human machine interface, abbreviated HMI, which can be integrated, for example, into an infotainment system of the road vehicle. A human-machine user interface is the place or action with which a human comes into contact with a machine, for example a touchscreen.

The recording and classification or an indication of oncoming traffic is given optically, e.g. by symbolically displaying oncoming traffic in a display, haptically, e.g. by vibration of a steering wheel, acoustically, for example by a warning tone which warns of a vehicle appearing behind a curve, or generally electronically. This notifies a driver of oncoming traffic before the oncoming traffic is visible, based on the recording and classification of traffic noise.

The evaluation device is intended for non-automated and automatically operated road vehicles. In the case of road vehicles which cannot be operated automatically, the evaluation device supports the human driver to perceive oncoming traffic on mountain roads, in particular perception over longer distances. If oncoming traffic also has the object according to the invention, honking is no longer absolutely necessary for the purpose of attracting attention when cornering on mountain roads. The horn continues to serve as an additional warning signal, which is recognized by the evaluation device. Drivers receive information about oncoming traffic via the output interface of the evaluation device. In the case of road vehicles which can be operated automatically, the evaluation device supports and improves the overall system of road vehicles in the detection of oncoming traffic. With limited detection of the environment by radiation sensors such as Lidar, radar and camera, the oncoming traffic is detected before its visibility with the object of the invention.

The artificial neural network is preferably trained in accordance with the training method according to the invention.

According to a development of the invention, the artificial neural network is trained to use the classified oncoming traffic to determine a vehicle control command in order to avoid an impending collision of the road vehicle with the oncoming traffic. The output interface is designed to issue this vehicle control command to a vehicle control device.

For example, if an oncoming vehicle approaches its own road vehicle in a narrow area of a mountain road that is not wide enough that both vehicles can pass each other, the artificial neural network detects the traffic noise of this vehicle and classifies it. The evaluation device is preferably designed to merge the traffic noises with data from environment detection sensors that detect the width of the road. The artificial neural network determines a braking operation as a vehicle control command so that the own road vehicle comes to a standstill and the oncoming vehicle can pass the narrow area. The output interface issues this vehicle control command to brake actuators of the road vehicle. As a result, the road vehicle brakes. There is no collision with the oncoming vehicle. This is particularly advantageous when the road vehicle is moving towards a U-turn and the oncoming vehicle is moving towards the road vehicle from an area of the U-turn that is not visible to the road vehicle. This oncoming vehicle cannot be detected for known environment detection sensors of driver assistance systems, for example radar, lidar or camera. However, if the oncoming vehicle shifts down or brakes a gear shortly before the U-turn, the evaluation device can process the noise of braking. The artificial neural network then determines an appropriate vehicle control command to avoid a collision. The evaluation device thus improves traffic safety on mountain roads. If the oncoming vehicle and your own road vehicle are equipped with the subject matter of the invention, the vehicle that drives uphill has priority. Alternatively, it can be provided that the road vehicle that drives downhill has priority.

The application system according to the invention is designed for a road vehicle for detecting and classifying traffic noises. The deployment system comprises at least one microphone that can be arranged on the road vehicle in order to receive traffic noise from oncoming traffic while driving the road vehicle on a mountain road. The application system also includes an evaluation device that can be integrated into the road vehicle. The evaluation device is designed to receive the traffic noise received with the microphone as an input. In addition, the evaluation device is designed to feed a trained artificial neural network forward with these traffic noises. The artificial neural network is trained to classify oncoming traffic based on the traffic noise. In addition, the evaluation device is designed to provide, as an output, an indication of the detected and classified oncoming traffic of a human-machine user interface, in particular to inform a human driver of the oncoming traffic.

In contrast to the training system according to the invention, the deployment system according to the invention has an already trained artificial neural network and provides the at least one identifier as an output. The application system is a system with the individual components of the microphone and evaluation device, which are in a functional unit due to the targeted use.

The evaluation device is preferably an evaluation device according to the invention.

The invention also relates to a use of the application system according to the invention as a driver assistance system. The driver assistance system cannot be a purely assistive or a fully automated driver assistance system. The fully automated driver assistance system merges data from other environment detection sensors and is designed to control the road vehicle autonomously, i.e. without human intervention.

The invention also relates to the use of the training system, training method, evaluation device and application system when driving on mountain roads for detecting and classifying traffic noises that come from oncoming traffic while driving on the mountain road.

The use of the object according to the invention is also within the scope of the invention in a vehicle which is traveling down a mountain road.

• 1 ein Ausführungsbeispiel einer erfindungsgemäßen Auswerteeinrichtung, integriert in ein Ausführungsbeispiel Straßenfahrzeug,
• 2 ein Ausführungsbeispiel eines erfindungsgemäßen Trainingssystems,
• 3 eine schematische beispielhafte Darstellung eines erfindungsgemäßen Trainingsverfahrens und
• 4 ein Ausführungsbeispiel eines erfindungsgemäßen Einsatzsystems.

The invention is described in detail in the following figures using exemplary embodiments. Show it:

• 1 an embodiment of an evaluation device according to the invention, integrated in an embodiment of a road vehicle,
• 2nd an embodiment of a training system according to the invention,
• 3rd a schematic exemplary representation of a training method according to the invention and
• 4th an embodiment of an application system according to the invention.

Unless otherwise stated, the same reference symbols in the figures denote the same or functionally similar reference parts. For the sake of clarity, the relevant reference parts are numbered in the respective figures.

1 shows as a road vehicle 1 a passenger car. The road vehicle 1 drives up a mountain road. In a front area of the road vehicle 1 is like a microphone 3a for example, a first directional microphone, aligned in the direction of travel. In a first side area of the road vehicle 1 is like a microphone 3b a second directional microphone, aligned towards the first side, arranged. In a second side area of the road vehicle 1 is like a microphone 3d a third directional microphone, aligned towards the second side, arranged. In a rear area of the road vehicle 1 is like a microphone 3c a fourth directional microphone, aligned against the first directional microphone, is arranged. The microphones record traffic noise 17th of in relation to the road vehicle 1 oncoming traffic 2nd on a mountain road.

Regarding the road vehicle 1 is oncoming traffic 2nd a second passenger car driving down the mountain road. The second passenger car is located relative to the road vehicle 1 behind a U-turn 7 the mountain road and is due to rocks in the U-turn 7 neither by a driver of the road vehicle 1 visual environment detection sensors such as radar, lidar and / or image sensor of the road vehicle 1 detectable. The lane width in the U-turn 7 is not wide enough to let two road vehicles pass each other. While maintaining the movements of the road vehicle 1 and the second passenger car it turns around 7 to a collision between the road vehicle 1 and the second passenger car.

Such a collision is with an evaluation device according to the invention 20th avoided as follows. The movement of oncoming traffic 2nd in the U-turn 7 the mountain road causes specific traffic noises 17th , for example braking noises, engine noises due to high engine speed and / or tire noises when entering and exiting the U-turn 7 . This traffic noise 17th are from the microphones 3a , 3b , 3c and 3d recorded, that is recorded. In an analysis of the recording of the traffic noise, in particular by means of the two spatially separated microphones 3a and 3d , the direction from which the traffic noise is coming can also be determined 17th arrive. This means that if there is more than one microphone, the microphones do not necessarily have to be directional microphones.

Via an input interface 21st , at the input of which the outputs of the microphones 3a , 3b , 3c and 3d concern, the inclusion of traffic noise in an artificial neural network 13 fed forward. The artificial neural network 13 is trained, traffic noise 17th to classify. For example, the artificial neural network recognizes 13 depending on the amplitude, frequency and duration of the traffic noise 17th whether it is a braking sound or an engine sound. The artificial neural network 13 has a fully connected architecture, i.e. every neuron 16 one layer is about connections 15 with every neuron 16 connected to the next layer. The artificial neural network 13 has four layers, namely an input layer with four neurons 16 , a first intermediate layer with three neurons 16 , a second intermediate layer with two neurons 16 and an output layer with a neuron 16 . The neurons 16 the input layer are only active, for example, when a noise exceeds a certain volume level. The neurons 16 the first intermediate layer are only active, for example, when the noises are traffic noises 17th are. The neurons 16 the second intermediate layer are only active, for example, when the traffic noise 17th specific traffic noise 17th of road vehicles, for example braking noises. The output layer outputs, for example, that the traffic noise is 95% probability a braking noise, 3% probability an engine noise and 2% probability no traffic noise 17th is. This architecture corresponds to a filter for classifying oncoming traffic 2nd based on traffic noise 17th . The filter can be used depending on the layer structure of the artificial neural network 13 , for example by several intermediate layers, can be refined as desired.

The activity of the neurons is determined by weighting factors 14 of the artificial neural network 13 learned. The weighting factors 14 are set via error minimization, i.e. the artificial neural network 13 is trained. The weighting factor w ³ ₁₁ evaluates the connection 15 between the first neuron 16 the third layer and the first layer in front of it, i.e. the second, i.e. three minus 1, layer.

During a training phase, the artificial neural network 13 learned in the traffic noise 17th oncoming traffic 2nd classify. Depending on the classified traffic noise 17th , and preferably depending on a detected lane width, a relative position and direction of movement of the second passenger car to the road vehicle 1 , determines the artificial neural network 13 a vehicle control command 4th . The vehicle control command 4th is via an output interface 22 to a vehicle control device 5 spent. The vehicle control device 5 regulates depending on the vehicle control command 4th the longitudinal and lateral control of the road vehicle 1 to avoid collision with the second passenger car. For example, the vehicle control device receives 5 the vehicle control command 4th Decelerate to a standstill. The road vehicle 1 so come before entering the U-turn 7 to a standstill. That is, the U-turn 7 is passed by only the second passenger car at a given time and a collision in the U-turn 7 with the road vehicle 1 is avoided.

The artificial neural network 13 is in the training phase with an in 2nd shown training system 10th trained. Via an input interface 11a maintains the artificial neural network 13 Actual training data 12th . The actual training data 12th are traffic noises 17th , namely braking noise from a vehicle tire 8th of the second passenger car when braking in the U-turn 7 going out. The braking sound attribute of a vehicle tire 8th of a passenger car braking in a U-turn 7 is a target indicator 18a . The target indicator 18a is shown as a symbol, but can ultimately be a simple number label with which the artificial neural network 13 know the traffic noise 17th a braking noise of a vehicle tire 8th of a passenger car braking in a U-turn 7 is. The artificial neural network 13 with the actual training data 12th fed forward. The artificial neural network receives in the forward feed 13 as a determined actual indicator 18b a horn sound that is different from the brake sound. The actual indicator 18b is shown simply as a symbol. Using subtraction becomes an error 19th between the actual indicator 18b and the target indicator 18a educated. The mistake 19th is fed backwards into the artificial neural network to determine the weighting factors via backward propagation 14 set so that the artificial neural network as an actual indicator when a brake sound is entered again 18b also receives the braking noise. This is called learning or training. The artificial neural network is designed to learn other traffic noises as well.

The training process is in 3rd shown. In the procedural step T1 be while driving the road vehicle 1 Traffic noise recorded on a mountain road 17th of the road vehicle 1 oncoming traffic 2nd at least one microphone 3a , 3b , 3c and or 3d and the associated target indicator 18a oncoming traffic 2nd as actual training data 12th provided. In the procedural step T2 becomes the artificial neural network 13 with the actual training data 12th fed forward. The one with the artificial neural network 13 determined actual indicator 18b oncoming traffic 2nd is in the process step T3 receive. In the procedural step T4 becomes the artificial neural network 13 with the mistake 19th between the determined actual indicator 18b and associated target indicator 18a fed backwards to the weighting factors 14 in the procedural step T5 adjust.

4th shows an operational system 30th that can be used as a driver assistance system. The operational system 30th has at least one microphone as components 3a , 3b , 3c and or 3d and an evaluation device 20th with a correspondingly based on detection and classification of traffic noise 17th of oncoming traffic 2nd artificial neural network trained on mountain roads 13 on. The traffic noise 17th a braking vehicle tire 8th is via the at least one microphone 3a , 3b , 3c and or 3d into the trained artificial neural network 13 fed. The evaluation device 20th leads the trained artificial neural network 13 out. The trained artificial neural network 13 with the traffic noise 17th fed forward. The trained artificial neural network is the output 13 at least one mark 18c oncoming traffic 2nd . The license plate 18c that with the trained artificial neural network 13 the corresponding target indicator 18a is then output, for example via a human machine interface. Alternatively or in addition to the issue of the license plate 18c is based on the indicator 18c , a position and / or a direction of movement of the oncoming traffic 2nd a vehicle control command 4th certainly, about an impending collision of the road vehicle 1 with oncoming traffic 2nd in the U-turn 7 to avoid.

Reference symbol list

1

Road vehicle

2nd

oncoming traffic

3a

microphone

3b

microphone

3c

microphone

3d

microphone

4th

Vehicle control command

5

Vehicle control device

7

U-turn

8th

Vehicle tires

10th

Training system

11

Input interface

12

Actual training data

13

artificial neural network

14

Weighting factor

15

connection

16

Neuron

17th

Traffic noise

18a

Debit indicator

18b

Actual indicator

18c

Mark

19th

deviation

20th

Evaluation device

21

Input interface

22

Output interface

30th

Operational system

T1-T5

Procedural steps

QUOTES INCLUDE IN THE DESCRIPTION

This list of documents listed by the applicant has been generated automatically and is only included for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• DE 102012218482 A1 [0004]

Claims (7)

Training system (10) for a road vehicle (1) for detecting and classifying traffic noise (17) comprising An input interface (11) in order to obtain actual training data (12), • the actual training data (12) • Include a recording of the traffic noise (17), which can be detected while the road vehicle (1) is traveling on a mountain road from oncoming traffic (2) with at least one microphone (3a, 3b, 3c, 3d) that can be arranged on the road vehicle (1) , and • each associated target identifier (18a) of this oncoming traffic (2), and wherein • the training system (10) is implemented, • to feed an artificial neural network (13) forward with the actual training data (12) and to receive in the forward feed an actual identifier (18b) of the respective oncoming traffic (2) determined with the artificial neural network (13), and • Adjust weighting factors (14) for connections (15) of neurons (16) of the artificial neural network (13) by feeding back a deviation (19) between the actual identifier (18b) and the target identifier (18a) in order to minimize the deviation (19) to learn the target number (18a) of the oncoming traffic. Data set consisting of recordings of traffic noises (17), which can be detected while driving a road vehicle (1) on mountain roads of oncoming traffic (2) with at least one microphone (3a, 3b, 3c, 3d) that can be arranged on the road vehicle (1) and the associated target identifier (18a) for the oncoming traffic (2), the data record being designed such that an evaluation device can process a respective recording together with the associated target identifier (18a) in order to train an artificial neural network to classify the traffic noise as a function of the target number plate (18a). Training method for an artificial neural network (13) for recording and classifying traffic noise (17) with the following method steps: • Providing traffic noises (17) recorded during a journey with a road vehicle (1), which, during a journey of the road vehicle (1) on a mountain road, from oncoming traffic (2) with at least one microphone (3a, 3a) that can be arranged on the road vehicle (1) 3b, 3c, 3d) can be recorded as actual training data (12) and respectively associated target identifier (18a) of this oncoming traffic (2) (T1), Feeding the artificial neural network (13) forward with the actual training data (12) (T2), Obtaining an actual identifier (18b) of the respective oncoming traffic (2) (T3) determined with the artificial neural network (13), • Backfeeding of the artificial neural network (13) with the deviation (19) between the determined actual identifier (18b) and the associated target identifier (18a) (T4), and • Obtaining weighting factors (14) for connections (15) of neurons (16) of the artificial neural network (13) in a reverse feed to minimize the deviation (19) in order to learn the target identifier (18a) of the oncoming traffic (T5 ). Comprising evaluation device (20) for a road vehicle (1) for detecting and classifying traffic noises (17) An input interface (21) in order to receive the traffic noise (17) oncoming traffic (2) on a mountain road, Wherein the evaluation device (20) is designed to feed a trained artificial neural network (13) forward with these traffic noises, • wherein the artificial neural network (13) is trained to use the traffic noise (17) to detect and classify the oncoming traffic (2), and • an output interface (22) to provide the detection and classification of the oncoming traffic (2). Evaluation device (20) according to Claim 4 , characterized in that the artificial neural network (13) according to the training method Claim 3 trained artificial neural network (13). Evaluation device (20) according to Claim 4 or 5 , characterized in that • the artificial neural network (13) is trained to use the classified oncoming traffic (2) to determine a vehicle control command (4) in order to avoid an impending collision of the road vehicle (1) with the oncoming traffic (2), and • the output interface (22) is designed to issue this vehicle control command (4) to a vehicle control device (5). Deployment system (30) for a road vehicle (1) for detecting and classifying traffic noises (17), comprising • at least one microphone (3a, 3b, 3c, 3d) that can be arranged on the road vehicle (1), in order to be able to drive the road vehicle (1 ) to receive traffic noise (17) from oncoming traffic (2) on a mountain road, and • an evaluation device (20) that can be integrated into the road vehicle (1), the evaluation device (20) being designed, • receiving the traffic noise (17) obtained with the microphone (3a, 3b, 3c, 3d), • a trained one to feed the artificial neural network (13) forward with these traffic noises (17), the artificial neural network (13) being trained to classify the oncoming traffic (2) on the basis of the traffic noises (17), and • as an exit a reference to the provide recorded and classified oncoming traffic (2) to a human-machine user interface.

Images