Method for classifying objects in the environment of a vehicle and driver assistance system

Abstract

The invention relates to a method for classifying objects in the surroundings of a vehicle (1) using ultrasonic sensors (10), which transmit ultrasonic pulses and receive ultrasonic echoes reflected by the objects, wherein the distance between the respective ultrasonic sensor (10) and the object reflecting the ultrasonic pulse in the surroundings is determined by at least two ultrasonic sensors (10) having at least partially overlapping fields of view (30), and, in order to distinguish between extended objects and point-like objects, the reflecting object is positionally determined by means of edge detection and the received ultrasonic echoes are attributed to object hypotheses. Furthermore, the point-like objects are classified in terms of height on the basis of classification parameters. The invention also relates to a driver assistance system (100) which is designed to implement the method.

Description

TECHNICAL FIELD

[0001] The invention relates to a method for classifying objects in the surroundings of a vehicle using ultrasonic sensors, which emit ultrasonic pulses and receive ultrasonic echoes reflected by the objects, wherein the distance between the respective ultrasonic sensor and the object in the surroundings that reflects the ultrasonic pulse is determined by at least two ultrasonic sensors with at least partially overlapping fields of view, and in order to distinguish between extended objects and point-like objects, the reflecting object is positionally determined by means of Lateration and the received ultrasonic echoes are attributed to object hypotheses. A further aspect of the invention relates to a driver assistance system configured to implement the method. BACKGROUND

[0002] Modern vehicles are equipped with a large number of driver assistance systems that assist the driver of the vehicle in implementing various driving maneuvers. Furthermore, driver assistance systems are known that warn the driver of hazards in the surroundings. Driver assistance systems require precise data about the surroundings of the vehicle, in particular about objects in the surroundings of the vehicle, in order to function.

[0003] Ultrasonic-based object localization methods are often used in which two or more ultrasonic sensors are used. The ultrasonic sensors here each emit an ultrasonic pulse and receive ultrasonic echoes reflected by objects in the surroundings. From the propagation time of the ultrasonic pulse until the respective ultrasonic echo is received and from the known speed of sound, the distance between the reflecting object and the respective sensor can each be determined. If an object is in the field of view of more than one ultrasonic sensor, i.e. distances to the object can be determined by multiple ultrasonic sensors, the exact position of the reflecting object relative to the sensors or relative to the vehicle can also be determined by a Lateration algorithm.

[0004] Due to the increasing field of view and sensitivity of the sensors, objects on the ground, such as kerbs, bumps or manhole covers, are also increasingly recognized. Here, it is important for the proper functioning of the driver assistance system to be able to distinguish between objects that are relevant for a collision, such as a column, a wall or a traffic sign, and objects that can be driven over and are not relevant for a collision, such as a kerb, a bump or a manhole cover.

[0005] A method for identifying objects having a low height is known from DE 10 2009 046 158 Al. In this arrangement, the distance to an object is continuously detected by means of a distance sensor, and it is checked whether, in the approach of the vehicle, the object is still detected by the distance sensor below a predefined distance or whether the object disappears from the detection region of the distance sensor. If it is identified that the object disappears from the detection region of the distance sensor in the approach, the object is classified as an object having a low height.

[0006] Furthermore, it is known in the art of technology to make use of the fact that high and extended objects generally do not have a unique, well-defined reflection point, but can thereby cause multiple reflections on a single ultrasound pulse and thus multiple ultrasound echoes temporally following one another. In the case of a high object, a reflection extends, for example, simply horizontally, i.e. parallel to the ground, from the sensor to the object and back to the high object. A further reflection is reflected back from the internal angle between the ground and the high object. This second ultrasound echo arrives temporally after the first ultrasound echo, since the path from the installation position of the sensor to the point of contact between the object and the ground must be covered by a longer path compared to the path extending simply parallel to the ground. Furthermore, it is known that specific objects, such as bushes or pedestrians, and planar objects, such as drainage grates or manhole covers, cause a large number of reflections, which are regarded as echoes as noise-like signals.

[0007] DE 10 2007 061 235 Al describes a method for classifying the height of an object by making use of the statistical dispersion, which is caused, inter alia, by multiple reflections of the measurement signal.

[0008] A problem of the known methods for height classification is that small objects and objects which can be regarded as point-like on a plane, such as posts or traffic signs, cause almost no multiple reflections due to their low reflectivity, and the ultrasound echoes reflected by these objects also only have a low amplitude, so that the use of the amplitude as the sole decision criterion for classifying between low objects and high objects cannot be considered. There is therefore a need, inter alia, for a robust method for height classification of objects, with regard to such point-like objects. SUMMARY

[0009] A method for classifying objects in the surroundings of a vehicle using ultrasonic sensors which emit ultrasonic pulses and receive ultrasonic echoes reflected by the objects is proposed. In this arrangement, the distance between the respective ultrasonic sensor and the object in the surroundings which reflects the ultrasonic pulse is determined by at least two ultrasonic sensors having at least partially overlapping fields of view, and in order to distinguish between extended objects and point-like objects, the position of the reflecting object is determined by means of trilateration and the received ultrasonic echoes are attributed to the object hypotheses. Furthermore, point-like objects represented by the object hypotheses are classified in terms of height on the basis of the following classification parameters: the update rate of the object hypotheses, the stability of the position of the object represented by the object hypotheses, the amplitude of the ultrasonic echoes attributed to the object hypotheses and the probability of the ultrasonic sensor obtaining ultrasonic echoes from the object represented by the object hypotheses. Point-like objects are understood here to mean objects which are essentially point-like in a plane parallel to the ground, i.e. have only a small extension, such as a post or a traffic sign. Furthermore, protruding parts of larger extended objects are also considered to be point-like objects, such as the edges of a house, the corners of a vehicle, the corners of a curbstone, the corners of a road hump or the like. Thus, objects whose extension visible to the sensor is less than 10 cm are in particular considered to be point-like objects. Conversely, objects having long edges which are considered to be extended in a plane parallel to the ground, such as a wall, a fence or another vehicle, are considered to be extended objects. Thus, objects having visible edges which are 10 cm or longer in a plane parallel to the ground are in particular considered to be extended objects.

[0010] Within the scope of the proposed method, ultrasonic pulses are continuously emitted and ultrasonic echoes reflected by the objects are continuously received in the case of the use of at least two ultrasonic sensors having at least partially overlapping fields of view. For this purpose, a plurality of ultrasonic sensors, for example 2 to 5 ultrasonic sensors, are arranged as a group, for example on the bumper of a vehicle. Using the known speed of sound in the atmosphere, the distance of the reflecting object in the surroundings of the vehicle to the respective ultrasonic sensor is determined. If an ultrasonic echo is received by a plurality of ultrasonic sensors, it can be assumed that the object which reflected the ultrasonic pulse is in the overlapping field of view of the two ultrasonic sensors. By applying a trilateration algorithm, the relative position of the reflecting object with respect to the vehicle or to the ultrasonic sensors can be determined. Here, two ultrasonic sensors which receive echoes from the object are sufficient for determining the position in a plane.

[0011] In this method, object hypotheses are created. The object hypotheses here make a synthesis of all distances determined by means of the ultrasonic sensor and other measured values, such as recorded ultrasonic echo amplitudes, which can be attributed to an object in the vehicle's surroundings. Each object hypothesis thus represents an object in the vehicle's surroundings. Here, if a measurement yields a position of an object that reflects the ultrasound that coincides with or lies in the vicinity of the position attributed to an object hypothesis, then in particular the measured values obtained one after the other in time, i.e. the distance values determined one after the other in time, can be attributed to the corresponding same object hypothesis. By analyzing the totality of the measurements attributed to an object hypothesis or by analyzing the distances and positions determined by means of the ultrasonic sensor, it is possible to infer the profile of the object. If, for example, the vehicle moves uniformly in one direction forwards and all the positions attributed to an object hypothesis lie on a line, or if all the positions attributed to an object hypothesis by all the ultrasonic sensors of a bumper lie on a line, then it is possible to infer that the object attributed to the object hypothesis is an extended object, such as a wall or another vehicle. Conversely, if the positions do not change approximately, then there can be a point-like object, which has only a small geometric extension in a plane parallel to the ground. This is, for example, a post, a traffic sign or a characteristic corner of another object, such as a vehicle corner or a house corner, or even a kerb edge corner. The piecing together of the individually measured distances to an extended object is described, for example, in DE 10 2007 051 234 A1.

[0012] In the case of an object hypothesis that is to be regarded as a point-like object, a height classification is then carried out in accordance with the proposed method. It is preferably provided here to distinguish between a passable object and a non-passable object. This distinction is meaningful, since, for example, a passable object can be driven over when carrying out a parking maneuver, whereas in the case of a non-passable object a driving maneuver must be aborted or a warning must be issued.

[0013] It is provided here in accordance with the application that a combination of different classification parameters can be taken into account for classifying a point-like object in terms of its height. Here, in accordance with the application the update rate of an object hypothesis, the stability of the position of the object represented by the object hypothesis, the amplitude of the ultrasonic echoes attributed to the object hypothesis and the probability of an ultrasonic sensor obtaining an echo from the object represented by the object hypothesis are used as classification parameters.

[0014] The probability of an ultrasonic sensor obtaining an ultrasonic echo from the object represented by the object hypothesis is preferably determined on the basis of the position of the object relative to the field of view of the respective ultrasonic sensor, the extent of the object ascertained and / or the detection threshold of the ultrasonic sensor.

[0015] In determining the probability, the position of the object relative to the field of view of the ultrasonic sensor has a large influence on the detection probability, since the amplitude of the emitted ultrasonic signal decreases with distance on the one hand and continuously falls off towards the edge of the field of view or rather towards the edge of the sound beam emitted by the ultrasonic sensor on the other hand. If the object is, for example, exactly in the center of the field of view, the amplitude of the ultrasonic waves impinging on the object is usually greatest, whereas the further the object is from the center of the field of view, the further the amplitude falls off. Furthermore, the extension of the object has a large influence on the size of the amplitude of the reflected ultrasonic echo. Large extended objects reflect more acoustic energy than small objects. Furthermore, a detection threshold is usually provided at the ultrasonic sensor in order to not classify ultrasonic echoes caused by the ground or the ground surface as ultrasonic echoes of objects. The ultrasonic echo is only classified as an ultrasonic echo reflected by an object when its amplitude is greater than a predefined threshold.

[0016] In this preferred provision, the detection threshold is matched to the currently prevailing ambient conditions, so that the detection threshold is lowered in the case of low ambient noise or low number of ground echoes, whereas the detection threshold is increased in a noisy ambient environment with many disturbing signals and greater noise and / or a large number of ground echoes, for example due to a rough ground surface such as gravel. In order to adapt the detection threshold, for example an algorithm can be used which adjusts the detection threshold such that a constant false alarm rate (CAFR) is achieved.

[0017] As a further decision criterion, the amplitude of the ultrasonic echo assigned to the object hypothesis is preferably provided for height classification. Here, on the one hand it can be taken advantage of that large extended objects usually have a higher amplitude than small objects. On the other hand, the change in amplitude can be monitored when the object approaches the vehicle or rather when the object approaches the ultrasonic sensor, as is known, for example, from DE 10 2009 046 158 Al, to determine whether the object continues to be detected or disappears from the field of view of the ultrasonic sensor. This "dive under" of the object below the range of the ultrasonic sensor is an indicator of a low object. Here, the analysis of the amplitude in the course of the change in amplitude as the object approaches the ultrasonic sensor in particular also involves normalizing the amplitude taking into account the object extension represented by the object hypothesis and / or the detection probability.

[0018] The stability of the position of the object represented by the object hypothesis is preferably taken into account as a decision criterion for the height classification of point-like objects. It is fully exploited here that high point-like objects such as posts and traffic signs have well-defined reflection points which are always reliably detected independently of the relative position of the object to the vehicle. In the case of low objects, for example kerb corners which behave as point-like objects, there is no clearly defined reflection point for the ultrasonic waves which are emitted, so that the specific position of the point-like object appears to be inconstant when the object approaches the vehicle or rather the respective ultrasonic sensor. Furthermore, this apparent inconstancy can lead to the fact that it becomes difficult to distinguish between an extended object and a point-like object as a result of this apparent inconstancy. This can be taken into account by attributing a confidence value to the classification as point-like object or rather as extended object, wherein the confidence value is preferably taken into account as a classification parameter for the height classification. Here, a greater unreliability in the classification indicates a low object, while a low unreliability or rather a high confidence value indicates a high point-like object.

[0019] The update rate of the object hypotheses is preferably used as a classification parameter for the height classification. It is fully exploited here that the probability of an object being detected simultaneously by more than one ultrasonic sensor is higher or lower depending on the nature of the object. In the case of extended objects it is usually ensured that the object is simultaneously in the field of view of more than one ultrasonic sensor, so that edge detection can be carried out frequently. This makes it possible to determine the position of the object which reflects the ultrasonic waves frequently, and thus to attribute the measured distance values to the object hypothesis and thus to update the object hypothesis. In the case of small point-like objects, the probability of the object being recognised simultaneously by more than one ultrasonic sensor, i.e. the probability of the ultrasonic echo reflected by the point-like object being intercepted by at least two ultrasonic sensors, is correspondingly lower. The respective object hypothesis for a point-like object can thus be updated less frequently. If the point-like object is a high object, direct acoustic reflection is usually possible, so that the probability of the echo of the high point-like object being intercepted by at least two ultrasonic sensors simultaneously is higher than in the case of a low point-like object. A low update rate of the object hypothesis thus indicates a low point-like object.

[0020] The object hypothesis is preferably always updated when a further ultrasonic echo is added to the object hypothesis. This usually always occurs when successful edge detection is possible, i.e. an ultrasonic echo of the object represented by the object hypothesis is received by at least two ultrasonic sensors, from which the position of the object can be determined by means of edge detection and attributed to the object hypothesis.

[0021] High classification of point-like objects can be performed using the mentioned classification parameters, in particular using statistical analysis processing methods or machine learning methods. Here, weighting factors are created and a correlation between the classification parameters is created, in particular on the basis of a training data set. For the case in which known objects are present, this training data set contains, in addition to the classification as point-like high object or point-like low object, also the corresponding classification parameter measurement. A suitable machine learning method here is the so-called random forest method, in which a large number of decision trees is created using the training data set. In the subsequent application with unknown data, the results of all decision trees are taken into account and the then most probable result is selected.

[0022] Another aspect of the application relates to a driver assistance system comprising at least two ultrasonic sensors having at least partially overlapping fields of view and comprising a controller. The driver assistance system is configured and / or arranged for carrying out any of the methods described herein.

[0023] Since the driver assistance system is configured and / or arranged for carrying out any of the methods, the features described in the context of any of the methods apply accordingly to the driver assistance system and vice versa, the features described in the context of any of the driver assistance systems apply to the methods.

[0024] The driver assistance system is accordingly arranged for identifying objects in the surroundings of the vehicle using the at least two ultrasonic sensors and classifying them as extended objects and point-like objects and, in the presence of a point-like object, high-classifying the point-like object.

[0025] The driver assistance system is preferably arranged for providing various different assistance functions using the data obtained on the objects in the surroundings of the vehicle. The driver assistance system preferably comprises a display function and a safety function. In the case of the display function, the distance to objects in the surroundings of the vehicle that are relevant for a collision is displayed, for example on a display screen, acoustically or by means of a light display. In the case of the safety function, intervention into the driving function is preferably provided when a dangerous situation is present. Such intervention into the driving function can be, for example, a braking intervention or a steering intervention. A dangerous situation is present, in particular, when a collision with an impassable object is imminent.

[0026] In the proposed driver assistance system, in a preferred embodiment, different weights of the classification parameters are used for the display function and the safety function, respectively, when high-classifying point-like objects. The weights of the classification parameters are here preferably predefined such that the probability of classification as an impassable object is higher for the display function than for the safety function.

[0027] Further, a vehicle is proposed, which comprises any one of the driver assistance systems described herein.

[0028] By means of the method according to the application, a highly reliable classification of point-like objects for a distance sensor is achieved. A reliable height classification, in particular a reliable classification into passable objects and non-passable objects, is crucial for the reliable functioning of numerous driver assistance systems. Driver assistance systems should not trigger a warning or even a braking intervention in the case of flat passable objects, such as kerbs, bumps or manhole covers, whereas objects which are relevant for a collision, such as pillars, walls, traffic signs or edges of other objects, such as house corners or vehicle corners, must be reliably identified.

[0029] The proposed method can advantageously be applied to all systems in which the ultrasonic sensors have at least partially overlapping fields of view and are capable of edge detection. Additional sensors are not necessary.

[0030] By classifying point-like objects as high objects which are relevant for a collision and low objects which are passable and do not require a reaction by the driver assistance system, the number of false warnings or even false system reactions in the absence of objects which are relevant for a collision is reduced, in particular, and the acceptance of the driver assistance system by the driver is improved.

[0031] It is further possible to select the weight of the individual classification parameters for the height classification differently depending on the application. For example, in the case of a driver assistance system which only has a display function, a higher rate of false classification of passable low objects as high objects, i.e. as non-passable objects, is accepted than in the case of a driver assistance system which has a safety function and is capable of braking interventions, for example. BRIEF DESCRIPTION OF DRAWINGS

[0032] Embodiments of the application are further explained according to the drawings and the following description.

[0033] The drawings show:

[0034] Figure 1 a side view of a vehicle with a driver assistance system according to the application;

[0035] Figure 2 a top view of the fields of view of a plurality of ultrasonic sensors at a sensor installation height; and

[0036] Figure 3 a top view of the fields of view of ultrasonic sensors at ground level. DETAILED DESCRIPTION

[0037] In the following description of embodiments of the invention, the same or similar elements are designated by the same reference numerals, and in some cases, repeated descriptions of these elements are omitted. The drawings are for illustrative purposes only, showing the subject matter of the invention.

[0038] Figure 1 A vehicle 1 is shown in a side view on road 22. Vehicle 1 includes a driver assistance system 100, which has ultrasonic sensors 10 and a controller 20. Figure 1 In the side view, only one ultrasonic sensor 10 is visible; however, vehicle 1 includes multiple ultrasonic sensors 10, compared to... Figure 2 and Figure 3 .exist Figure 1 In the illustrated embodiment, the driver assistance system 100 also includes a display device 28 connected to the controller 20. The controller 20 is also configured to implement braking intervention. This is in Figure 1 The connection between the controller 20 and the pedal 29 is shown in the diagram.

[0039] exist Figure 1 The ultrasonic sensor 10, visible in the image, is mounted at a height h at the rear of the vehicle 1. The ultrasonic sensor 10 has a field of view 30 within which it can identify objects such as traffic signs 26 or ridges 24. Figure 1 Another bulge 24', also shown in the image—this bulge is closer to vehicle 1 than bulge 24—is in Figure 1 In the situation shown, the bump 24' can no longer be detected by the ultrasonic sensor 10 because it is outside the field of view 30 of the ultrasonic sensor 10. When the vehicle 1 approaches the bump 24, the height classification of the bump 24 can be identified by changes in amplitude or detection behavior. If the vehicle 1 slowly reverses towards the bump 24, the bump leaves the field of view 30 of the ultrasonic sensor 10 at a specific point, which can be identified by a strong drop in the amplitude of the corresponding ultrasonic echo. The moment when the ultrasonic sensor 10 can no longer detect the bump 24, or the distance between the bump 24 and the vehicle 1 at that moment, can be used to infer the height of the bump 24. If the bump 24 is a tall object, similar to a traffic sign 26, it will not leave the field of view 30 of the ultrasonic sensor 10 upon approach. This situation of leaving the field of view 30 upon approach only occurs with low, normally drivable objects.

[0040] However, because the area capable of reflecting the ultrasonic waves from the ultrasonic sensor 10 is relatively small, and therefore the amplitude of the received ultrasonic echo is relatively small, the traffic sign 26 cannot be reliably classified as a high-value object solely based on amplitude. Therefore, other criteria must be considered. According to the present invention, the update rate of the object hypothesis representing the object, the amplitude of the ultrasonic echo, the stability of the object location determination, and the probability that the ultrasonic sensor 10 obtains an ultrasonic echo from the object are used as classification parameters.

[0041] If an object of significant importance to the collision is identified, i.e. a tall, impassable object, a warning can be issued and / or braking intervention can be performed via display device 28.

[0042] Figure 2 The rear of vehicle 1 is shown schematically. Figure 2 In the example shown, four ultrasonic sensors 10 are mounted at the tail. Figure 2 This schematically illustrates the field of view assigned to ultrasonic sensors 11 to 14 at mounting heights 31 to 34 of ultrasonic sensor 10, in contrast to... Figure 1 .

[0043] Figure 3 The same arrangement as the ultrasonic sensor 10 of vehicle 1 is shown. (And...) Figure 2 The difference indicates the field of view at ground heights of 41 to 44.

[0044] exist Figure 2 and Figure 3 The comparison shows that the field of view at installation heights 31 to 34 is larger than the corresponding field of view at ground heights 41 to 44. In particular, the area where the fields of view 31 to 34 and 41 to 44 of at least two ultrasonic sensors 10 overlap is significantly larger when considered along the installation height h than when considered along the ground height.

[0045] from Figure 2 A comparison of the field of view at installation heights 31 to 34 and at ground heights 41 to 44 shows that, when the object has a small height above the ground, there is a lower probability of it being "simultaneously within the field of view 30 of at least two ultrasonic sensors 10" compared to an object at the same location with a height at least equivalent to the installation height h of the ultrasonic sensor 10. Figure 1 .

[0046] It is only possible to perform an edge measurement and thus to determine the position of the object reflecting the ultrasound waves if at least two ultrasound sensors 10 receive ultrasound echoes reflected by one object. It is only possible to create and / or update an object hypothesis actually representing an object in the environment of the vehicle 1 if the position of the object reflecting the ultrasound waves is known. Thus, the probability of recognizing a high object is higher when continuously performing measurements using ultrasound sensors 10 than in the case of a low object. Thus, if an object is recognized once and an object hypothesis is created accordingly, the object hypothesis is updated accordingly with a higher probability if the object relates to a high object than if the object relates to a low object. Thus, it can be considered to use the update rate of an object hypothesis as a decision criterion for performing a height classification.

[0047] Furthermore, according to Figure 3 It can be derived from the depicted field of view diagrams at ground level 41 to 44 and the field of view diagrams at mounting height 31 to 34 that the relative position of an object with respect to the fields of view 31 to 34 and 41 to 44 also has an influence on the detection probability. Since the sound amplitude continuously decreases from the center of the fields of view 31 to 34 and 41 to 44 towards the edges, the probability of recognizing an object is higher if the object is in the center of one or more fields of view 31 to 34 and 41 to 44 than in the case of the same object being at the edge of the fields of view 31 to 34 and 41 to 44. Thus, it is preferred to consider the detection probability given by the relative position of an object at the fields of view 31 to 34 and 41 to 44 when classifying.

[0048] The application is not limited to the embodiments described herein and aspects emphasized therein. Rather, a large number of modifications within the scope of the skilled person are possible, which are described by the claims.

Claims

1. A method for classifying an object in the surroundings of a vehicle (1) using an ultrasonic sensor (10) that emits ultrasonic pulses and receives ultrasonic echoes reflected by the object, wherein, The distance between the respective ultrasonic sensor (10) and an object in the surroundings which reflects ultrasonic pulses is determined by at least two ultrasonic sensors (10) having at least partially overlapping fields of view (30), and for distinguishing between extended objects and point-like objects, the reflecting object is positionally determined by means of edge detection and the received ultrasonic echoes are attributed to an object hypothesis, wherein objects having visible edges with a length of 10 cm or more, viewed in a plane parallel to the ground, are regarded as the extended objects, wherein objects having an extent of less than 10 cm, viewed in a plane parallel to the ground, are regarded as the point-like objects, wherein the point-like objects represented by the object hypotheses are classified in terms of height on the basis of the update rate of the object hypotheses, the stability of the position of the object represented by the object hypotheses, the amplitude of the ultrasonic echoes attributed to the object hypotheses and the probability of the ultrasonic sensor (10) obtaining ultrasonic echoes from the object represented by the object hypotheses as classification parameters, wherein the update rate of the object hypotheses is higher if the object is involved in a high point-like object than if the object is involved in a low point-like object, wherein the probability of the ultrasonic sensor (10) obtaining ultrasonic echoes from the object represented by the object hypotheses is determined on the basis of the position of the object relative to the field of view (30) of the ultrasonic sensor (10), the determined extent of the object and / or a detection threshold of the ultrasonic sensor (10), wherein a confidence value for the classification as a point-like object is taken into account as a further classification parameter for the classification in terms of height, wherein a high confidence value indicates a high point-like object, wherein the classification in terms of height is carried out using a machine learning method, wherein weighting factors are created and a correlation between the classification parameters is created on the basis of a training data set.

2. The method of claim 1, wherein, The respective detection threshold of the ultrasonic sensor (10) is thus matched to the current noise level such that the rate of false classification of ultrasonic echoes as object echoes is constant.

3. The method of any one of claims 1-2, wherein, The amplitude of the ultrasonic echoes is corrected on the basis of the determined extent of the object represented by the object hypotheses.

4. The method according to any one of claims 1 to 3, characterized in that, The object hypotheses are updated when a further ultrasonic echo is added to the object hypotheses.

5. The method according to any one of claims 1 to 4, characterized in that, A random forest method is used as the machine learning method.

6. A driver assistance system (100) comprising at least two ultrasonic sensors (10) with overlapping fields of view (30) and having a controller (20), characterized in that, The driver assistance system (100) is designed to implement the method according to any one of claims 1 to 5.

7. The driver assistance system (100) according to claim 6, wherein The driver assistance system (100) comprises a display function and a safety function, wherein the display function shows a report on objects in the surroundings of the vehicle (1) on a display device (28), and the safety function is designed to intervene in a driving function in the event of a dangerous situation, characterized in that different weights of the classification parameters are assigned to the display function and the safety function, respectively.

Images