Method and driver assistance system for classifying objects in a vehicle's surroundings

Abstract

The invention relates to a method for classifying objects (30) in the surroundings of a vehicle (1) using ultrasonic sensors (12, 13, 14, 15), wherein an ultrasonic signal (20) is emitted, ultrasonic echoes (22, 24) are received from objects (30) in the surroundings, and the positions of reflection points relative to the ultrasonic sensors (12, 13, 14, 15) are determined using the least squares method, and wherein the reflection points are continuously determined and assigned to objects (30) in the surroundings. Furthermore, provision is made for determining a dispersion parameter related to the position of a reflection point assigned to an object (30) and for using the dispersion parameter as a classification criterion related to the type of the object (30). Further aspects of the invention relate to a driver assistance system (100) and a vehicle (1) comprising such a driver assistance system (100).

Description

Technical Field

[0001] The present invention relates to a method for classifying objects in the surroundings of a vehicle, wherein ultrasonic signals are emitted using an ultrasonic sensor, ultrasonic echoes are received from objects in the surroundings, and the positions of reflection points relative to the ultrasonic sensor are determined using a least squares method, wherein the reflection points are continuously determined and assigned to objects in the surroundings. Further aspects of the present invention relate to a driver assistance system configured to implement the method and a vehicle comprising such a driver assistance system. Background Art

[0002] Modern vehicles are equipped with numerous driver assistance systems that assist the vehicle driver in carrying out various driving maneuvers. Furthermore, driver assistance systems are known that warn the driver of dangers in the surroundings. To perform their functions, these driver assistance systems require precise data about the vehicle's surroundings, and in particular, precise data about objects located in the vehicle's surroundings.

[0003] Ultrasonic object localization methods are often used, employing two or more ultrasonic sensors. Each ultrasonic sensor emits an ultrasonic pulse and receives an ultrasonic echo reflected by an object in the surrounding environment. The propagation time of the ultrasonic pulse until the corresponding ultrasonic echo is received, along with the known speed of sound, allows the distance between the reflecting object and the corresponding sensor to be determined. If an object is within the field of view of more than one ultrasonic sensor, i.e., if the distance from more than one ultrasonic sensor to the object can be determined, the exact position of the reflecting object relative to the sensors or the vehicle can also be determined using a least-squares algorithm. It is usually sufficient if an object is visible to both ultrasonic sensors, i.e., if both ultrasonic sensors can determine the distance to the object.

[0004] To fulfill their function, modern driver assistance systems require not only information about the position of an object but also information about its type. Objects are therefore preferably classified to distinguish between objects that are relevant for warnings or brake interventions, such as pedestrians, walls, or trees, and objects that are not relevant, such as curbs.

[0005] A method for classifying distance data from an ultrasonic distance detection system is known from DE 10 2007 061 235 A1. In this method, a measurement signal is emitted and then intercepted by a sensor, which reflects the measurement signal from a distant object. The distance is calculated based on the time elapsed between the emission and reception of the measurement signal and the known propagation velocity. Furthermore, it is provided that the statistical dispersion of the distance data is correlated with the height of the reflecting object. In the case of distance values ​​with prominent outliers, a large deviation indicates a larger object. If such an extended object has a smooth surface with a small amount of structure, a large dispersion is still observed in the distance values, but there are no prominent outliers in the measured values. If the distance values ​​lie on a line with a small dispersion, this can be considered a small, elongated object, such as a curb.

[0006] DE 10 2013 018 721 A1 discloses a method for identifying at least one parking space for a motor vehicle. Here, an occupancy grid is created as a digital model of the surroundings, with detected quantities entered into each cell. Furthermore, discrete centers are determined, representing areas where the signal emitted by the sensor is strongly reflected. Object classification is also performed, and the determined discrete centers are compared with comparison data during the object classification. This makes it possible, for example, to distinguish between a motor vehicle and other objects.

[0007] DE 10 2016 218 064 A1 discloses an operating method for an ultrasonic sensor system in which ultrasonic signals reflected by objects in the surrounding area are received and assigned to lanes. A search is performed for temporal changes in an echo image in a sequence of echo signals from ultrasonic signals emitted temporally successively. Pedestrians stand out because they generally reflect only a small amount of sound energy and are in motion.

[0008] A disadvantage of the known methods is that a high error rate occurs when classifying objects in the vehicle's surroundings. It is therefore desirable to consider the use of additional parameters for classifying objects. Summary of the Invention

[0009] The present invention proposes a method for classifying objects in the surroundings of a vehicle. In this method, an ultrasonic sensor is used to transmit ultrasonic signals, ultrasonic echoes are received from objects in the surroundings, and the positions of reflection points relative to the ultrasonic sensor are determined using the least squares method. The determination of the reflection points and their assignment to objects in the surroundings are performed continuously. Furthermore, a dispersion parameter associated with the positions of the reflection points assigned to an object is determined and used as a classification criterion associated with the type of the object.

[0010] Within the scope of the proposed method, ultrasonic signals are continuously emitted using at least two ultrasonic sensors whose fields of view at least partially overlap, and ultrasonic echoes reflected by an object are correspondingly received again. Preferably, multiple ultrasonic sensors—for example, two to six ultrasonic sensors—are arranged in a group, for example, on the bumper of a vehicle. Using the known speed of sound in air, the distances of reflecting objects in the vehicle's surroundings from each ultrasonic sensor are determined. If an ultrasonic echo is received by multiple ultrasonic sensors, it can be assumed that the object reflecting the ultrasonic signal is in the overlapping fields of view of the two ultrasonic sensors. By applying a least squares algorithm, the position of the reflecting object relative to the vehicle or to the participating ultrasonic sensors can be determined. Two ultrasonic sensors receiving echoes from the object are sufficient to determine the position in a plane.

[0011] When objects are classified, they can be sorted into different categories, such as "low, drivable objects" or "high, non-drivable objects." Furthermore, a typification can be performed, in which specific types of objects are distinguished within the scope of the classification. For example, in the case of long objects such as curb edges, point-shaped objects such as pillars or columns, and complex objects such as pedestrians, bushes, or trees.

[0012] The classification process preferably incorporates additional criteria in addition to the discreteness parameter. For example, additional criteria may include the number of echoes received for the transmitted ultrasonic signal or the behavior of the measurement data, such as the number and amplitude of echoes, as the vehicle approaches the object. The number of echoes, for example, depends on whether the object has a clearly defined reflection point. Furthermore, for tall objects, at least two ultrasonic echoes typically appear: the first echo is reflected by the object at the same height as the ultrasonic sensor, and the second echo is reflected at the transition between the object and the ground. The behavior of the received measurement data as the vehicle approaches an object can also provide conclusions about the reflecting object. For example, when approaching a wall, the amplitude of the received echo does not change or changes only slightly, whereas when approaching a curb edge—which represents a low object—the amplitude decreases as the vehicle approaches the object.

[0013] The determined dispersion parameter describes how the reflection points assigned to an object are distributed geographically. Accordingly, the mean, standard deviation, variance, or another statistical parameter known to those skilled in the art can be used as the dispersion parameter. Provision can be made to disregard outliers—that is, individual reflection points that deviate significantly from the mean—when determining the dispersion parameter. For example, a check can be performed to determine whether the reflection points are more than a multiple of the mean distance from the center point, or an outlier test known to those skilled in the art can be applied.

[0014] Preferably, the discreteness parameter represents the discreteness of the reflection point along two mutually orthogonal directions.

[0015] Alternatively, the center point of an object can be determined by averaging the reflection points assigned to the object, and the proportion of reflection points on the object that lie within or outside a predetermined radius around the object's center point can be used as a dispersion parameter. The radius of the circle can be fixed in advance, for example, from a range of 20 cm to 100 cm, preferably from 30 cm to 80 cm, and particularly preferably from 40 cm to 60 cm, and for example set at 50 cm. If the radius is set at 50 cm, for example, in the case of point-shaped objects such as pillars or columns, nearly all of the reflection points will lie within the predetermined radius. In the case of more complex objects such as pedestrians, shrubs, or trees, a larger proportion of the reflection points will still lie within the predetermined radius, but due to the increased dispersion caused by the poorly defined positions of the reflection points, a certain proportion of the reflection points will already lie outside this radius. In the case of linear objects, such as curb edges, a higher dispersion is observed, resulting in a larger proportion of the reflection points lying outside the predetermined radius.

[0016] Within the scope of the method, a bounding box is preferably determined, which describes the range in which all reflection points assigned to an object, with the exception of reflection points determined as outliers, lie, wherein the size of the bounding box is determined as the dispersion parameter.

[0017] The bounding box is preferably configured to be tolerant to statistical outliers, so that significant expansion of the bounding box due to reflection points classified as outliers is not, or at least not completely, accommodated within the bounding box. It may be provided that a certain reflection point history is stored before the initial creation of the bounding box to avoid creating a bounding box with outliers from the outset. For example, it may be provided that at least five to ten reflection points are first assigned to an object before the first bounding box is created. After the initial creation of the bounding box, the bounding box is updated as soon as additional reflection points are added to the object within the scope of the method.

[0018] The determined bounding box preferably has a longitudinal extension and a transverse extension, wherein the longitudinal extension describes the intensity of the dispersion in the longitudinal direction and the transverse extension describes the intensity of the dispersion in the transverse direction.

[0019] Alternatively or additionally, provision can be made for each object to determine the dispersion parameter by creating an occupancy grid, wherein the cells of the occupancy grid have an occupancy value indicating how many reflection points are assigned to the corresponding cell. The occupancy grid represents a grid, wherein each cell indicates how many reflection points have been detected at the position represented by the corresponding cell. Accordingly, as soon as a reflection point can be assigned to a cell of the grid or occupancy grid, the occupancy value of the corresponding cell is incremented.

[0020] Preferably, a longitudinal extent and a transverse extent are determined as dispersion parameters based on the occupancy values ​​of the cells of the occupancy grid, wherein the longitudinal extent describes the intensity of the dispersion in the longitudinal direction and the transverse extent describes the intensity of the dispersion in the transverse direction.

[0021] By independently determining the dispersion parameter for at least two orthogonal directions—for example, the longitudinal and transverse directions—it is possible to deduce whether the object is uniformly dispersed or whether the dispersion is greater in one direction than in another. For each of these directions, it is then possible to independently assess whether the dispersion occurs over a wide area or whether the reflection points, or their locations, are concentrated in a small area. To this end, limits can be predefined for each direction in which the dispersion parameter was determined, in order to distinguish between low and high dispersion.

[0022] The longitudinal extent preferably extends parallel to the direction pointing away from the vehicle, while the transverse extent preferably extends perpendicularly thereto. Alternatively, the direction of the object's maximum extent is preferably determined, the transverse extent extending parallel to this direction and the longitudinal extent extending perpendicularly thereto. Alternatively, it is preferably provided that an object model having a point geometry or a line geometry is assigned to the object by evaluating the relative positions of the reflection points assigned to the object, in the case of line geometry, the transverse extent extending parallel to the orientation of the line and the longitudinal extent extending perpendicularly thereto.

[0023] Particularly preferably, in the case where an object model is present, a bounding box is created which is oriented corresponding to the determined transverse and longitudinal directions.

[0024] The proposed method enables better classification of specific object types, in particular, better differentiation between curb edges, pedestrians, and point-like objects such as pillars or columns. During classification, it is provided that curb edges are identified by a large dispersion in the transverse direction and a small dispersion in the longitudinal direction. Point-like objects are identified by a small dispersion in the transverse direction and a small dispersion in the longitudinal direction. Objects with complex geometries, in particular pedestrians, are identified by a large dispersion in the transverse direction and a large dispersion in the longitudinal direction. For example, a distinction is made between large and small dispersions by presetting a limit value for the dispersion. This limit value can be predefined differently for the longitudinal and transverse directions, wherein a large dispersion is considered if the predefined limit value is exceeded, and a small dispersion is considered if the dispersion is equal to or below the limit value.

[0025] Furthermore, it is conceivable to perform classification using a machine learning method using the dispersion parameter determined as described. Training data is used, which includes the assignment of specific measured dispersions to specific object types. A corresponding trained model can then be used for the classification of the dispersion parameter determined within the context of this method.

[0026] Another possible method for classification is to determine a distribution of discrete values ​​for the object types to be distinguished based on training data. By continuously observing the discrete values ​​of an object, the probability of a specific object type can be derived. This probability is preferably combined with other features (such as the number and / or amplitude of echoes) to make the final classification. This approach is similar to machine learning methods, except that the decision criteria are determined by the developer and are therefore known and adaptable.

[0027] Another aspect of the present invention relates to a driver assistance system comprising at least two ultrasonic sensors with at least partially overlapping fields of view and a controller. The driver assistance system is designed and / or configured to implement any of the methods described herein.

[0028] Since the driver assistance system is constructed and / or arranged for carrying out any of the methods, the features described within the scope of any of the methods apply accordingly to the driver assistance system, and conversely, the features described within the scope of any of the driver assistance systems apply accordingly to the method.

[0029] The driver assistance system is accordingly configured to detect objects in the surroundings of the vehicle using at least two ultrasonic sensors and to classify the objects in the surroundings of the vehicle.

[0030] The field of view is the area within which the corresponding ultrasonic sensor can perceive an object. By arranging the fields of view of at least two ultrasonic sensors so that they overlap, it is possible for multiple sensors to receive corresponding ultrasonic echoes when an ultrasonic pulse is emitted. This allows more than one ultrasonic sensor to determine the distance between an object and the vehicle, or the corresponding ultrasonic sensor, and thus determine the object's position relative to the ultrasonic sensor or vehicle using a least-squares algorithm.

[0031] Furthermore, a vehicle is proposed, which comprises any of the driver assistance systems described in this document.

[0032] The method proposed according to the present invention for classifying objects in the surroundings of a vehicle makes it possible to use a new classification criterion for classifying objects, taking into account the dispersion associated with the positions of reflection points assigned to an object. Using this new classification criterion already provides information about the type of reflecting object. Furthermore, the proposed classification criterion can be combined with known classification criteria, such as the number of ultrasonic echoes received after transmitting an ultrasonic signal and the behavior of measured values ​​obtained when the vehicle approaches an object.

[0033] Particularly within the scope of this combination, in which several different criteria can be considered for classification, the newly proposed classification criterion, which is related to the discreteness of a specific location, helps to significantly reduce the error rate, in particular to reduce the false positive rate when classifying curb edges or to achieve a better true positive rate when identifying pedestrians and trees.

[0034] Furthermore, this method makes it possible to differentiate between object classes that go beyond a simple classification (drive-through / not drive-through) and thus differentiate between pedestrians, trees, pillars, and small shrubs. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Embodiments of the present invention are explained in more detail with reference to the drawings and the following description.

[0036] The accompanying drawings show:

[0037] Figure 1 : A vehicle having a driver assistance system according to the invention;

[0038] Figure 2 : Typical discrete characteristics of pedestrians;

[0039] Figure 3 : Typical discrete properties of point-like objects, and;

[0040] Figure 4 : Typical discrete characteristics of curb edges. DETAILED DESCRIPTION

[0041] In the following description of the embodiments of the present invention, identical or similar elements are denoted by the same reference numerals, wherein a repeated description of these elements in individual cases is omitted. The figures merely schematically illustrate the subject matter of the present invention.

[0042] Figure 1 A vehicle 1 is shown, which includes a driver assistance system 100 according to the present invention. In the example shown, driver assistance system 100 includes four ultrasonic sensors 12, 13, 14, 15, all of which are arranged at the front of vehicle 1 and are each connected to a controller 18. Controller 18 is accordingly configured to transmit ultrasonic signals 20 using ultrasonic sensors 12, 13, 14, 15 and to receive ultrasonic echoes 22, 24 from objects 30 in the surroundings of vehicle 1.

[0043] The ultrasonic sensors 12, 13, 14, 15 are arranged at the front of the vehicle 1 in such a way that the fields of view of at least two ultrasonic sensors 12, 13, 14, 15 at least partially overlap. Figure 1In the situation shown in FIG, an object 30 is within the field of view of both the first ultrasonic sensor 12 and the second ultrasonic sensor 13. In the illustrated example, an ultrasonic signal 20 is transmitted by the first ultrasonic sensor 12 and reflected by the object 30. The ultrasonic echo 22 received by the first ultrasonic sensor 12 is referred to as a direct echo because it is received by the same ultrasonic sensor 12, 13, 14, 15 that also transmitted the original ultrasonic signal 20. The other ultrasonic echo 24 received by the second ultrasonic sensor 13 is referred to as a cross echo because it is received by another ultrasonic sensor 12, 13, 14, 15.

[0044] For the classification of the object 30, it is provided that the position of the reflection point 44 is determined using the least squares method, with reference to Figure 2 、 Figure 3 and Figure 4 , which indicate the locations at which the emitted ultrasonic signal 20 is reflected by the object 30. This determination is carried out continuously, so that a plurality of reflection points 44 are determined. Each determined reflection point 44 is assigned to an object 30, wherein for this purpose, for example, the distance between the reflection points 44 or an object model assigned to the object 30 can be used as a criterion. This object model describes assumptions about the shape and extent of the object 30. For example, as an assumption, it can be assumed that the object 30 is a point-shaped object 34, such as a pillar. In another assumption, it can be assumed that the object 30 is a linear object, such as a curb edge or a wall. The reflection points 44 that meet these assumptions of the model are then assigned to a corresponding object 30.

[0045] The method further provides for examining the dispersion of the positions of the reflection points 44 in more detail and determining corresponding dispersion parameters, which are then used as criteria for classifying the type of object 30 .

[0046] In the following Figure 2 、 Figure 3 and Figure 4 The positions of specific reflection points 44 of different typical objects 30 are shown in FIG. Figure 2 、 Figure 3 and Figure 4 An object 30 is shown in each case, which is located in front of vehicle 1 and is therefore within the visual range of a plurality of ultrasonic sensors 12 , 13 , 14 , 15 .

[0047] Figure 2 The distribution of reflection points 44 of the object 30 is shown, and the object is a pedestrian 32. Figure 2As can be seen in the diagram, the determined reflection points 44 appear highly concentrated at the actual position of pedestrian 32. However, due to the complex shape and structure of pedestrian 32, widely scattered outliers can also be seen, which are mainly highly scattered in the transverse direction 42. In the longitudinal direction, the reflection points 44 are also scattered, but the longitudinal extent 40 of the dispersion in the longitudinal direction is much smaller.

[0048] For the reflection point 44 Figure 2 By analyzing the discreteness plotted in , for example, the direction of the longest extent of the object 30 can be determined and the lateral direction of the discreteness with the lateral extent 42 can be oriented along this direction. The direction of the longitudinal extent 40 is correspondingly perpendicular thereto. A bounding box can then be created, i.e., a box having a width corresponding to the extent of the lateral discreteness and a length corresponding to the longitudinal extent 40 of the discreteness.

[0049] Figure 3 The dispersion of the positions assigned to the reflection points 44 is shown by way of example using the point-shaped object 34 as the object 30. Figure 2 Compared to the example of a pedestrian 32, it can be seen that both the longitudinal extent 40 of the dispersion in the longitudinal direction and the transverse extent 42 of the dispersion in the transverse direction are smaller. In the case of point-shaped objects 34, the dispersion of the reflection points 44 is smaller because, in the case of such point-shaped objects 34, the positions at which the ultrasound reflections occur are well defined, so that only small deviations of the reflection points 44 occur.

[0050] exist Figure 4 In FIG, the position of the reflection point 44 is plotted using the curb edge 36 as an example of the object 30. Figure 4 As can be seen in the illustration in , the curb edge 36 typically exhibits a wide dispersion in terms of the transverse extent 42, wherein there is usually no particular concentration of reflection points 44 at a specific point. In addition, a small dispersion is observed in the longitudinal direction, so that the longitudinal extent 40 of this dispersion is correspondingly small.

[0051] The present invention is not limited to the embodiments described herein and the aspects emphasized in the embodiments, but rather various modifications are possible within the scope of the conventional technical means of a person skilled in the art within the scope of the claims.

Claims

1. A method for classifying objects (30) in the surroundings of a vehicle (1), wherein: Ultrasonic signals (20) are emitted using at least two ultrasonic sensors (12, 13, 14, 15) having at least partially overlapping fields of view, ultrasonic echoes (22, 24) are received from an object (30) in the surroundings, and the position of a reflection point (44) relative to the ultrasonic sensors (12, 13, 14, 15) is determined using a least squares method, wherein the object is located in the at least partially overlapping fields of view of the at least two ultrasonic sensors, wherein the reflection point (44) is continuously determined and assigned to the object (30) in the surroundings, It is characterized in that a dispersion parameter is determined in relation to the position of a reflection point (44) assigned to an object (30) and the dispersion parameter is used as a classification criterion in relation to the type of the object (30). The determined dispersion parameter describes how the reflection points assigned to the object (30) are distributed in place. A bounding box is determined, which describes a range in which all reflection points (44) assigned to an object (30) are located, except for reflection points (44) determined as outliers, and a size of the bounding box is determined as a dispersion parameter. The bounding box has a longitudinal extension (40) and a transverse extension (42), wherein the longitudinal extension (40) describes the intensity of the dispersion in the longitudinal direction and the transverse extension (42) describes the intensity of the dispersion in the transverse direction. and / or, An occupancy grid is created for each object (30) to determine the dispersion parameter, wherein the cells of the occupancy grid have occupancy values, which indicate how many reflection points (44) are assigned to the corresponding cell, wherein a longitudinal extension (40) and a transverse extension (42) are determined as dispersion parameters based on the occupancy values ​​of the cells of the occupancy grid, wherein the longitudinal extension (40) indicates the intensity of the dispersion in the longitudinal direction, and the transverse extension (42) indicates the intensity of the dispersion in the transverse direction.

2. The method according to claim 1, characterized in that The dispersion parameters respectively describe the dispersion of the reflection points (44) along two mutually orthogonal directions.

3. The method according to claim 1, characterized in that The center point of an object (30) is determined by averaging reflection points (44) assigned to the object (30), and the proportion of the reflection points (44) that is within or outside a predetermined radius around the center point of the object (30) is determined as a dispersion parameter.

4. The method according to claim 1, wherein The longitudinal extension (40) extends parallel to a direction pointing away from the vehicle (1), and the transverse extension (42) extends perpendicular to the direction pointing away from the vehicle, or determining a direction of the greatest extension of the object (30), the lateral extension (42) extending parallel to this direction and the longitudinal extension (40) extending perpendicular to this direction, or By evaluating the relative positions of reflection points (44) assigned to an object (30), an object model having a point geometry or a line geometry is assigned to the object (30), in which case the transverse extension (42) extends parallel to the orientation of the line and the longitudinal extension (40) extends perpendicular to the orientation of the line.

5. The method according to claim 4, characterized in that During classification, a curb edge (36) is identified by a large dispersion in the direction of the transverse extension (42) and a small dispersion in the direction of the longitudinal extension (40), identifying point-like objects (34) by a small dispersion in the direction of the transverse extension (42) and a small dispersion in the direction of the longitudinal extension (40), and Objects (30) with complex geometric shapes, in particular pedestrians (32), are recognized by a large dispersion in the direction of the transverse extension (42) and a large dispersion in the direction of the longitudinal extension (40).

6. A driver assistance system (100) comprising at least two ultrasonic sensors (12, 13, 14, 15) having at least partially overlapping fields of view, and comprising a controller (18), characterized in that The driver assistance system (100) is configured to carry out a method according to any one of claims 1 to 5.

7. A vehicle (1) comprising a driver assistance system (100) according to claim 6.

Images