Method for localizing a vehicle, sensor device, vehicle, and computer program product

Abstract

The invention relates to a method for localizing a vehicle (1) from sensor data sets which describe an environment of the vehicle (1) and which comprise in each case a plurality of Gaussian distributions (11, 12), comprising the following steps: – capturing a first sensor data set at a first vehicle position (9), the first sensor data set comprising a plurality of first Gaussian distributions (11) in a first coordinate system, – capturing a second sensor data set after a movement of the vehicle (1), the second sensor data set comprising a plurality of second Gaussian distributions (12) in a second coordinate system, – assigning the respective closest first Gaussian distribution (11) to the second Gaussian distributions (12) or the respective closest second Gaussian distribution (12) to the first Gaussian distributions (11), the assignment being effected in each case by evaluation of a distance criterion, – determining at least one transformation specification comprising a displacement and / or a rotation between the first and the second coordinate systems with minimization of the distances between the respective closest Gaussian distributions (11, 12) which are described by the distance criterion, – determining a second vehicle position (10) from the first vehicle position (9) and the transformation specification.

Description

[0001] Description

[0002] Method for locating a vehicle, sensor device, vehicle and computer program product

[0003] The invention relates to a method for localizing a vehicle from sensor data sets describing the vehicle's surroundings and recorded using at least one sensor of a sensor device of the vehicle. The sensor data sets each comprise a plurality of Gaussian distributions. The Gaussian distributions are each formed from at least one measurement point of the sensor device and, with respect to a coordinate system, each describe at least part of an object in the vehicle's surroundings. Furthermore, the invention relates to a sensor device, a vehicle, and a computer program product.

[0004] To gain knowledge of their surroundings, vehicles can be equipped with sensors that can, for example, detect objects in the vehicle's vicinity. Knowledge of the presence of these objects can facilitate the vehicle's driving maneuvers. The information obtained from the sensors can be displayed to the driver and can also enable autonomous or semi-autonomous driving.

[0005] For example, distance sensors can be used, which can determine the distance between the sensor or the vehicle and an object in the vehicle's surroundings. Distance sensors typically comprise a transmitting device that transmits signals and a receiving device that receives signals reflected from objects in the surroundings as echoes. The distance, or in other words, the distance, to the object can be determined based on the time difference between the time the signal is sent and the time the echo is received, taking into account the propagation speed of the signal.

[0006] Such distance sensors can typically be used in motor vehicles for environmental detection. Ultrasonic sensors, for example, are particularly important in this regard, as they can be used in semi-automatic or automatic driving maneuvers, especially in connection with parking applications, such as parking distance measurement, parking space searches, or parking. The motor vehicle is typically moved relative to the objects, with a measurement cycle being performed at predetermined times during the movement. During each measurement cycle, at least one signal is emitted by a distance sensor.

[0007] If a vehicle movement has occurred between two measurement cycles, the position and / or orientation of the detected objects changes relative to a vehicle-fixed coordinate system. If information about the vehicle movement is to be derived from a comparison of the relative object arrangements from different measurement cycles, it is necessary to perform an assignment, also known as registration, of the detected objects. This allows the relative position change of the objects in relation to the vehicle to be determined. The relative position change of stationary objects can then also be used to trace the movement of the vehicle.

[0008] The assignment or registration of objects presents a challenge. The applicable methods can depend, in particular, on the type of object description in the environmental data generated from the sensor measurements. The environmental data can be generated, for example, as point clouds from the measurement points or as object maps in which the detected objects are described as geometric figures, such as polygons.

[0009] One method for assigning different points in two data sets is described, for example, in the article by Besl, Paul J.; N.D. McKay "A Method for Registration of 3-D Shapes", IEEE Transactions on Pattern Analysis and Machine Intelligence, 14 (2), pp. 239-256, 1992. In this method, a transformation or mapping rule that maps a first point data set to a second point data set is determined by searching step by step for the points in the first data set to find the closest point in the second data set. Subsequently, one of the data sets is transformed in a transformation step in order to minimize the distances. These steps are carried out until a minimum of the distances is reached. The mapping rule for the data sets is then obtained from the sum of the transformation steps carried out in each case.

[0010] In addition to describing objects using point clouds of measured points or as polygons, objects can also be represented in other forms in the environment data. In this regard, descriptions that enable a compact, i.e., data-efficient, description of the objects are particularly desirable. More compact environment data reduces the resources required for its processing and / or storage, thus enabling more efficient use and / or the use of cheaper hardware.

[0011] However, even such a description of objects presents the challenge of efficiently assigning or registering objects in two different data sets.

[0012] The invention is based on the object of providing an efficient method for assigning or registering objects in sensor data sets, which in particular also enables a compact description of the objects.

[0013] To achieve this object, the invention provides for a method of the type mentioned above to comprise the following steps:

[0014] - Recording a first sensor data set at a first vehicle position, wherein the first sensor data set comprises a plurality of first Gaussian distributions in a first coordinate system,

[0015] - Recording a second sensor data set after a movement of the vehicle, wherein the second sensor data set comprises a plurality of second Gaussian distributions in a second coordinate system,

[0016] - Assigning the respective nearest first Gaussian distribution to the second Gaussian distributions or the respective nearest second Gaussian distribution to the first Gaussian distributions, wherein the assignment is carried out by evaluating at least one distance criterion describing a distance between two Gaussian distributions, - Determining at least one transformation rule comprising a shift and / or a rotation between the first coordinate system and the second coordinate system while minimizing the distances between the respective nearest Gaussian distributions described by the distance criterion,

[0017] - Determining a second vehicle position from the first vehicle position and the at least one

[0018] T ransformation rule.

[0019] A first sensor data set is therefore initially recorded at a first vehicle position. The first vehicle position can, for example, be the destination of a previously performed vehicle movement. The vehicle can have moved into the first position at least partially independently, i.e. autonomously or semi-autonomously. Alternatively, the vehicle can also have been moved into the first position by a driver. The first vehicle position can, in particular, be selected as a function of a driving maneuver to be performed with the assistance of the method according to the invention. The driving maneuver can, in particular, be performed autonomously or semi-autonomously. The driving maneuver can, for example, be a parking maneuver, i.e. parking in or out of a parking space.During a parking maneuver, the first vehicle position can, for example, be in the vicinity of a parking space; during a parking maneuver, the first vehicle position can be within the parking space.

[0020] The first sensor data set is acquired in the first vehicle position and consequently describes the objects in the surroundings of the vehicle. For this purpose, a plurality of measuring points are acquired by means of the at least one sensor, which is designed, for example, as an ultrasonic sensor. In the sensor data set, in particular one or more measuring points, which describe, for example, signals (echoes) reflected from the sensor from a similar distance and / or from a similar direction, are combined to form an ellipse described by a Gaussian distribution. The shape of the Gaussian distribution or the ellipse can in particular correspond to the spatial distribution of the reflection locations of the echoes described by the respective measuring points or can at least be approximated thereto. An ellipse or a Gaussian distribution describing the ellipse can each describe an object or a part or a section of an object in the surroundings of the vehicle.In particular, large objects and / or objects with a more complex geometry can be described by several Gaussian distributions or several ellipses.

[0021] The spatial distribution of the measurement points and / or the spatial extent of the Gaussian distributions determined from the measurement points are specified in the first sensor data set in relation to the first coordinate system. The first coordinate system can, for example, describe the arrangement of the Gaussian distributions in relation to the sensor or in relation to the vehicle. Since the vehicle is in the first vehicle position when the first sensor data set is recorded, the first coordinate system specifies the spatial distribution of the Gaussian distributions in relation to the first vehicle position.

[0022] If the vehicle now moves to a second vehicle position that differs from the first vehicle position, the reference system to which the spatial distribution of the Gaussian distributions is specified when measuring the second sensor data set by the at least one sensor changes. The Gaussian distributions described in the second sensor data set are specified in a second coordinate system, which can also refer to the vehicle. Due to the vehicle's movement to the second vehicle position, the Gaussian distributions now refer to the second vehicle position.

[0023] To enable localization of the vehicle, i.e., determination of the second vehicle position from the first vehicle position, the first sensor data set and the second sensor data set are then merged, for example by transferring the Gaussian distributions of one sensor data set into the second sensor data set without taking into account the offset between the coordinate systems. Subsequently, for each of the first Gaussian distributions in the first sensor data set, the closest Gaussian distribution in the second sensor data set can be determined. The closest distribution is determined by evaluating the distance criterion describing a distance between two Gaussian distributions. At this point, it is also possible to determine the closest first Gaussian distribution for each of the second Gaussian distributions.

[0024] After the closest Gaussian distribution of the other sensor data set has been determined, in particular for each of the Gaussian distributions of one sensor data set, a transformation rule comprising a shift and / or a rotation between the first coordinate system and the second coordinate system is determined. The transformation rule is determined by minimizing the distances between the closest Gaussian distributions described by the distance criterion. In other words, the first coordinate system and the second coordinate system are aligned relative to each other, which allows the different reference positions to be at least partially compensated.

[0025] This makes it possible to determine, for example calculate or at least approximate, the second vehicle position from the first position using the transformation rule. In particular, it is possible that after determining a first transformation rule, the Gaussian distributions are assigned again and then a further transformation rule is determined by again minimizing the distances described by the distance criterion between the now assigned pairs of Gaussian distributions. The steps of assigning the Gaussian distributions and determining a transformation rule can be repeated until a minimum for the distances is reached. The assignment and the subsequent determination of a transformation rule can therefore be repeated iteratively.

[0026] The second vehicle position can then be determined, in particular, depending on all determined transformation rules. For example, an overall transformation rule can be determined that maps all displacements and / or all rotations, which, when applied to the first vehicle position, can be used to determine the second vehicle position.

[0027] Regarding the assignment of the Gaussian distributions of one sensor data set to the Gaussian distributions of the other sensor data set, it should be noted that the closest Gaussian distribution, i.e., the one located at the shortest distance described by the distance criterion, is considered in each case. This does not necessarily have to correspond to a Gaussian distribution assigned to the same object. In particular, the closest Gaussian distributions can also change for different assignment steps. Iteratively, a shift and / or rotation between the first coordinate system and the second coordinate system can advantageously be determined, thus enabling the vehicle to be localized based on the first sensor data set and the second sensor data set.

[0028] The vehicle's localization can occur, in particular, as part of a (partially) automated driving process. The first sensor data set and the second sensor data set can each be referred to as maps or map information. The Gaussian distributions can each be a multidimensional probability density distribution. In the sensor data sets, the Gaussian distributions can be described, for example, as covariance matrices.

[0029] The method according to the invention advantageously enables a simple mapping of environmental information. The determined sensor data sets are memory-efficient and require little computing power for their creation. The localization of the vehicle can also be advantageously determined within a predetermined time with comparatively low computing power. This advantageously results in the method being suitable for sensor devices that, in addition to at least one sensor, comprise a computing device with comparatively low computing capacity. The method according to the invention makes it possible to perform a localization of the vehicle based on a first and a second sensor data set within an acceptable time, even with such a sensor device.Furthermore, localizing the vehicle from the sensor data sets has the advantage of enabling precise positioning, particularly during a driving maneuver involving multiple vehicle movements. Compared with position determination using odometry, the method according to the invention does not exhibit any accumulating errors. Furthermore, the method can be used to precisely determine the position even during movements in a confined space and / or during a driving maneuver involving a significant rotational and / or lateral movement of the vehicle, such as a parking maneuver. The method can also be used, in particular, to determine a vehicle position multiple times during a driving maneuver.

[0030] In a preferred embodiment of the invention, it can be provided that the Gaussian distributions in the first sensor data set and the second sensor data set each describe an ellipse summarizing one or more measurement points. Summarizing, in particular, several measurement points as an ellipse represents an efficient way of describing the objects in the sensor data sets. In this context, an ellipse has the advantage that its mathematical description is comparatively simple compared to a polygon or a point cloud and thus requires only a small data volume. Furthermore, an ellipse can assume different shapes, for example, depending on the ratio of its semiaxes, and thus describe the geometry of different objects.

[0031] Such an ellipse can advantageously be described by a Gaussian distribution. The ellipse can, for example, be assigned to a probability, and a particularly two-dimensional Gaussian distribution can be determined in such a way that it maps the corresponding ellipse based on its covariance for this probability. Compared to describing ellipses using their semi-axes and / or foci, describing ellipses using Gaussian distributions represents a more efficient type of mathematical description.

[0032] According to the invention, it can be provided that the Gaussian distributions are each at least two-dimensional, wherein a distance criterion is used which specifies at least one distance in relation to the at least two dimensions. The two dimensions can correspond to a plane of a road surface on which the vehicle is standing. In this way, the Gaussian distributions can each describe an ellipse lying in this plane. It is also possible for three-dimensional Gaussian distributions to be used, in particular if the at least one sensor and / or a sensor device comprising the sensor is also designed to determine an object height assigned to the respective measuring points.

[0033] According to the invention, a Mahalanobis distance, a Bhattacharyya distance, a Kullback-Leibler distance and / or a probability-based distance between two Gaussian distributions can be used as a distance criterion, wherein the probability-based distance indicates the probability associated with the mean of one of the Gaussian distributions in the other Gaussian distribution.

[0034] For example, the expected value of a Gaussian distribution can be used as the mean of a Gaussian distribution. The probability-based distance can indicate the probability that the mean or expected value of a first Gaussian distribution lies within a second Gaussian distribution, with the second Gaussian distribution being selected as the closest Gaussian distribution with the highest probability. This also applies conversely to the assignment of a first Gaussian distribution as the closest distribution to a second Gaussian distribution.

[0035] Alternatively or additionally, mathematical distances that specify a distance between two distributions, such as the Bhattacharyya distance or the Kullback-Leibler distance, can also be used as a distance criterion. A mathematical distance such as the Mahalanobis distance, which specifies the distance of a point, for example, the center of one of the distributions, from a distribution, can also be used additionally or alternatively as a distance criterion.

[0036] According to the invention, it can be provided that the assignment of the nearest Gaussian distributions takes place depending on at least one assignment criterion, wherein the assignment criterion used is a ratio between the amplitudes of the measured values ​​described by the Gaussian distributions, a ratio between directions assigned to the measured values ​​described by the Gaussian distributions, and / or a ratio between the curvatures of the objects described by the Gaussian distribution. The assignment criteria can each be used in addition to the distance criterion, or one or more higher-dimensional distance criteria can be used which, in addition to a spatial distance, for example in the x-direction and y-direction, also comprise at least one further dimension corresponding to an assignment criterion. The use of the assignment criteria can improve the quality of the assignments.Advantageously, the number of required iteration steps can be reduced by using one or more assignment criteria, thus enabling a rapid determination of the assignment rule.

[0037] By using one or more matching criteria, it becomes more likely that a first Gaussian distribution will be matched to a second Gaussian distribution (or vice versa) that includes measurements reflected from the same object. Since the properties of the object influence the echoes reflected from the object, properties of the echoes can be considered to determine suitable Gaussian distributions or to avoid obvious or at least highly probable mismatches.

[0038] Depending on the assignment criteria, it can therefore be decided whether, instead of a Gaussian distribution having the smallest distance described by the distance criterion, another Gaussian distribution, for example the one having the second smallest distance, is chosen as the closest Gaussian distribution if the Gaussian distribution having the smallest distance only has a poor degree of agreement with one or more of the assignment criteria considered.For example, it is possible for several assignable second Gaussian distributions to lie within a distance range around a first Gaussian distribution, for example at distances that are smaller than a predetermined distance limit, wherein the closest Gaussian distribution is selected from this group of second Gaussian distributions depending on the at least one assignment criterion, in particular depending on the degree of agreement of the properties of the Gaussian distributions and / or their measured values ​​considered as assignment criteria. This example also applies accordingly when assigning one of several first Gaussian distributions to a second Gaussian distribution.

[0039] The ratio between the amplitudes of the measured values ​​described by the Gaussian distributions can be used as the assignment criterion. A Gaussian distribution can be assigned as the nearest Gaussian distribution if the amplitudes are as similar as possible or if the amplitude ratio lies within a predetermined interval. Additionally or alternatively, the ratio between directions assigned to the measured values ​​described by the Gaussian distributions can also be considered as an assignment criterion, with an assignment occurring in particular if the measured values ​​are assigned the same or at least a similar direction. The direction can, if necessary, be corrected using movement information describing the movement of the vehicle in order to take into account a relative change in direction due to the vehicle movement.In particular, the azimuth angle at which the echoes forming the measured values ​​are received by the sensor can be used as the direction, whereby a similar direction can exist if the ratio of the directions or the respective azimuth angles lie within a predetermined interval.

[0040] Additionally or alternatively, it is also possible to use the ratio between the curvatures of the objects described by the Gaussian distribution as the assignment criterion. The curvature refers to the shape of the object or to the shape of a side surface of the object facing the sensor. In particular, Gaussian distributions can be considered as possible closest Gaussian distributions, which describe objects with the same or a similar curvature, i.e., those lying within a predetermined curvature interval. In a preferred embodiment, the invention can provide for the use of several assignment criteria, wherein the assignment criteria are weighted against one another. The weighting factors can be determined, for example, using machine learning.

[0041] According to the invention, it can be provided that at least some of the first Gaussian distributions and at least some of the second Gaussian distributions are each assigned a label, wherein only those Gaussian distributions having a matching label are used to determine the nearest Gaussian distribution. The label can, for example, indicate an object type and / or an object class of an object described by the Gaussian distribution. It is possible for a physical object to be described by several Gaussian distributions, wherein the Gaussian distributions each describe a part or section of the object. Each of the Gaussian distributions can be assigned a label describing the object type and / or the object class of the entire object.

[0042] To determine the nearest Gaussian distribution, the invention can provide that only Gaussian distributions are considered whose mean lies within a multiple of the variance of the Gaussian distribution for which the nearest Gaussian distribution is being determined. In this way, the computational effort required to determine the respective distances resulting from the one or more distance criteria can be reduced. Of those Gaussian distributions that lie within the multiple of the variance, the Gaussian distribution with the shortest distance can then be selected as the nearest Gaussian distribution. When using one or more assignment criteria, the Gaussian distribution can also be selected for which at least one assignment criterion matches and which, of several Gaussian distributions with matching assignment criteria, has the shortest distance.

[0043] In a preferred embodiment, the invention can provide for the transformation rule to be determined in relation to mean values ​​of the respective Gaussian distributions and / or in relation to covariance matrices of the respective Gaussian distributions. Using the mean values ​​exclusively can advantageously reduce the computational effort required to determine the transformation rule. With the additional or alternative consideration of the covariance matrices, all of the information derivable from the respective Gaussian distributions can advantageously be taken into account, thus enabling a more precise and / or faster determination of the transformation rule. Taking the covariance matrices into account therefore also allows the particularly elliptical shape of the Gaussian distributions to be taken into account when determining the transformation rule describing a shift and / or a rotation.Considering the covariance matrices can be implemented as a weighted determination of the transformation rules. For example, the inverses of the covariance matrices can be used as weighting factors.

[0044] According to the invention, it can be provided that a transformation rule comprising a scaling is determined, in particular by means of singular value decomposition. The transformation rule can comprise a scaling to compensate for a different scale between the first coordinate system and the second coordinate system. In addition to a shift and / or a rotation, the transformation rule can comprise, for example, a magnification or reduction factor with respect to one of the coordinate systems as scaling. The use of singular value decomposition (SVD) has the advantage that the computational effort for determining the transformation rule is comparatively low.

[0045] In a preferred embodiment, it can be provided that a transformation rule comprising an error covariance matrix is ​​determined, in particular using a Gauss-Helmert model. The error covariance matrix can provide an indication of the accuracy of the transformation rule. By using a Gauss-Helmert model to determine the transformation rule, not only a transformation rule describing, for example, a shift, a rotation, and a scaling, but also an associated error covariance matrix can be efficiently determined. Furthermore, efficient transferability and efficient application of the method can be enabled, even with sensor data sets comprising more than two dimensions.

[0046] Before determining the nearest Gaussian distributions, an initial transformation can be applied to at least one of the coordinate systems according to the invention, wherein the initial transformation is determined as a function of movement information describing the movement. The initial transformation can be carried out in particular before a first assignment of the Gaussian distributions in order to at least partially compensate for an offset between the first coordinate system and the second coordinate system right from the start. The initial transformation can be carried out in particular if the offset between the first coordinate system and the second coordinate system can be estimated at least approximately on the basis of movement information describing the movement of the vehicle from the first vehicle position to the second vehicle position.

[0047] According to the invention, it can be provided that the movement information is determined by means of odometry and / or by means of a navigation method, in particular by means of satellite navigation. The movement information can be determined, for example, by a control device designed to carry out the method. The control device can be part of the vehicle and / or a sensor device that also includes the at least one sensor. For determining the movement information, the control device can be connected, for example, by means of an odometry device providing odometry data and / or a navigation device providing navigation data, for example a satellite navigation device, or can comprise an odometry device and / or a navigation device.

[0048] A sensor device according to the invention is provided for comprising at least one sensor and at least one control device, wherein the control device is designed to carry out a method according to the invention. The control device can form a structural unit with the at least one sensor or be designed separately from the sensor and communicatively connected thereto. The control device can be part of the vehicle or a vehicle-external control device communicatively connected to the at least one sensor.

[0049] A vehicle according to the invention is provided with a sensor device according to the invention. The control device of the sensor device can be arranged within the vehicle. Alternatively, the control device can also be arranged externally of the vehicle and communicate with the at least one sensor via a communication connection, in particular a wireless one.

[0050] A computer program product according to the invention is provided to include instructions that cause a control device to execute a method according to the invention. The computer program product according to the invention can be stored on a non-transient data carrier, for example, a CD, a floppy disk, or a USB stick. It is also possible for the computer program product to be stored on a computing device from which it can be retrieved via a communications connection, for example, the Internet, and / or stored on a control device.

[0051] All advantages and configurations described above in relation to the method according to the invention apply accordingly to the sensor device according to the invention, the vehicle according to the invention and the computer program product according to the invention and vice versa.

[0052] Further advantages and embodiments of the invention will become apparent from the exemplary embodiments described below and from the drawings. These are schematic representations and show:

[0053] Fig. 1 shows an embodiment of a vehicle according to the invention with an embodiment of a sensor device according to the invention,

[0054] Fig. 2 is a flowchart of an embodiment of a method according to the invention, Fig. 3 is a diagram comprising first Gaussian distributions and second Gaussian distributions to explain the embodiment of a method according to the invention and

[0055] Fig. 4 shows a further diagram comprising the measuring points of the sensor device used to form the Gaussian distributions.

[0056] Figure 1 shows an exemplary embodiment of a vehicle 1. Two other vehicles 2, 3 are located in the vicinity of the vehicle 1, which is designed as a passenger car, with a parking space 4 between them. The vehicle 1 comprises an exemplary embodiment of a sensor device 5, which includes a control device 6 and one or more sensors 7. For the sake of clarity, only a single sensor 7 is shown. The sensor 7 is designed as an ultrasonic sensor and serves to detect part of the surroundings of the vehicle 1.

[0057] The control device 6 is configured to perform a method for locating the vehicle 1 in a sensor data set describing the surroundings of the vehicle 1. The sensor data set can be recorded using the sensor 7 or the sensor device 5. The control device 6 can form a structural unit with the at least one sensor 7 or be implemented separately from the sensor and communicatively connected thereto. The control device can be part of the vehicle 1 or a vehicle-external control device that is connected, in particular wirelessly, to the at least one sensor 7.

[0058] The sensor 7 has a detection area 8 in which signals emitted by the sensor 7 are reflected by objects in the vicinity of the vehicle 1, for example, the vehicles 2, 3, and received again as echoes. From these echoes, the sensor device 5 can determine the distance to the objects or vehicles 2, 3, for example, from a time-of-flight measurement.

[0059] When the vehicle 1 moves from a first vehicle position 9 to a second vehicle position 10 (shown in dashed lines), the relative arrangement between the vehicle 1 and the other vehicle 2, 3 changes. For example, when performing this movement during an assisted or at least partially automated driving maneuver, it may be necessary to determine the second vehicle position 10 with high accuracy. This requires localization of the vehicle 1, which can be performed by the control device 6 based on sensor data sets determined by the sensor 7.

[0060] Fig. 2 shows a flowchart of an exemplary embodiment of a method for locating a vehicle in a sensor data set describing the vehicle's surroundings and recorded with the aid of at least one sensor of the vehicle. The method can be carried out by the control device 5.

[0061] At the first vehicle position 9, a first sensor data set is recorded in a step S1 via the sensor device 5 or the sensor 7. The first sensor data set comprises a plurality of measurement points, each of which corresponds to one of the signals emitted by the sensor 7 and received as echoes. The first sensor data set comprises a plurality of first Gaussian distributions, each of which is formed from one or more measurement points of the sensor device 5. The measurement points or the Gaussian distributions are specified in relation to a first coordinate system.

[0062] After recording the first sensor data set, the vehicle moves to a second vehicle position in a step S2. Subsequently, in a step S3, a second sensor data set is recorded via the sensor device 5 or the sensor 7. Analogous to the first sensor data set, the sensor data set recorded in the second vehicle position 10 also comprises a plurality of second Gaussian distributions, which are specified in relation to a second coordinate system. Both the first coordinate system and the second coordinate system can be specified in relation to the vehicle 1, but the coordinate systems have an offset from one another, which corresponds to the movement of the vehicle 1, i.e. the movement between the first vehicle position 9 and the second vehicle position 10.

[0063] For localization, the control device 6 determines a transformation rule that specifies the offset between the first coordinate system of the first sensor data set and the second coordinate system of the second sensor data set. From the first vehicle position 9 and the transformation rule, the second vehicle position 10 and thus the current vehicle orientation relative to the surroundings or the other vehicles 2, 3 at that time can then be determined.

[0064] In step S4, the first sensor data set and the second sensor data set are combined. This can be done, for example, by transferring the measurement points from the second sensor data set to the first sensor data set. The positions assigned to the respective measurement points differ due to the offset between the first and second coordinate systems.

[0065] Fig. 3 shows a diagram comprising first Gaussian distributions 11 and second Gaussian distributions 12. The diagram represents, for example, the first coordinate system, with an x-coordinate plotted on the abscissa and a y-coordinate plotted on the ordinate. The first Gaussian distributions 11 are each formed from the measurement points of the first sensor data set and are represented by closed points. Similarly, the second Gaussian distributions 12 are each formed from the measurement points of the second sensor data set. The second Gaussian distributions 12 are represented by X-shaped points.

[0066] The first Gaussian distributions 11 in the first sensor data set and the second Gaussian distributions 12 in the second sensor data set each describe an ellipse summarizing one or more measurement points. The first Gaussian distributions 11 and the second Gaussian distributions 12 are each represented as ellipses, in this case, for example, as equi-probability ellipses with a probability of p = 0.7.

[0067] The measuring points used to form the Gaussian distributions 11, 12 are shown in Fig. 4, with symbols corresponding to the representation of the Gaussian distributions being used for the measuring points. The measuring points describe the surroundings of the vehicle 1. For example, the measuring points located in an area 14 can describe the parking space 4 or a structure that borders the parking space in the longitudinal direction, such as a curb or the like, whereas the measuring points in area 15 describe, for example, the other vehicle 2. The measuring points shown in Fig. 4 are for illustrative purposes only. Preferably, the measuring points are not part of the sensor data sets, which in particular only contain the Gaussian distributions instead of the measuring points, in order to enable simpler data processing and data storage.

[0068] The Gaussian distributions 11, 12 extend in the x-direction and the y-direction, respectively, and are thus two-dimensional. It is evident that there is a spatial offset between the second Gaussian distributions 12 and the first Gaussian distributions 11.

[0069] Based on the combined sensor data sets, in a step S5, each of the first Gaussian distributions 11 from the first sensor data set recorded at the first vehicle position 9 is assigned the closest second Gaussian distribution 12 from the second sensor data set. The assignment is performed by evaluating at least one distance criterion describing a distance between two Gaussian distributions. For example, the second Gaussian distribution 12 that has the shortest distance or separation from the first Gaussian distribution 11 is selected as the closest Gaussian distribution.

[0070] For example, in step S5, the second Gaussian distributions 12_1, ..., 12_K are assigned to the first Gaussian distributions 11_1, ..., 11_N. In particular, each first Gaussian distribution 11_1, ..., 11_N is assigned a second Gaussian distribution 12. The number K of second Gaussian distributions 12 may differ from the number N of first Gaussian distributions. To determine the nearest second Gaussian distribution 12_1 , ... 12_K to each of the Gaussian distributions 11_1 , ... , 11_N, the distances d1 , ... , dN are calculated, where dj = distance(12_i, 11_j) with i=1 , ... , K and j=1 , ... , N, and distance(a, b) describes the distance criterion between the Gaussian distributions a and b. For example, the second Gaussian distribution 12 with the smallest distance is assigned to each of the first Gaussian distributions 11.For example, a Mahalanobis distance, a Bhattacharyya distance, a Kullback-Leibler distance, and / or a probability-based distance between two Gaussian distributions can be used as a distance criterion. The probability-based distance indicates the probability associated with the mean of one of the Gaussian distributions in the other Gaussian distribution. The aforementioned distances each enable the calculation of a spatial distance between a first Gaussian distribution 11 and a second Gaussian distribution 12 with respect to the x-coordinate and the y-coordinate.

[0071] To reduce the computational effort required to determine the nearest Gaussian distribution, for example, only those second Gaussian distributions 12 whose mean lies within a multiple of the variance of the first Gaussian distribution 11 for which the nearest Gaussian distribution is to be determined can be considered. Of the second Gaussian distributions 12 that lie within the multiple of the variance, the second Gaussian distribution 12 that is closest can then be selected as the nearest Gaussian distribution.

[0072] In step S6, a transformation rule is subsequently determined, which describes a shift and / or a rotation of the second Gaussian distributions 12 with respect to the first Gaussian distributions 11. The second Gaussian distributions 13 transformed according to the transformation rule are represented in Fig. 3 by O-shaped points. Points corresponding to the transformed Gaussian distributions 13 are also shown in Fig. 4. These measurement points also serve only for illustration purposes. Preferably, the measurement points corresponding to the transformed measurement points are also not part of the sensor data sets, which in particular contain the Gaussian distributions instead of the measurement points.

[0073] Clearly, the transformed second Gaussian distributions 13 exhibit a smaller spatial offset from the first Gaussian distributions 11. The transformation rule is determined, in particular, by minimizing the distances described by the distance criterion between the first Gaussian distributions 11 and their respective, nearest second Gaussian distributions 12. In particular, a sum of the distances can be minimized.

[0074] In addition to a translation and / or rotation, the transformation rule can also describe a scaling to compensate for a different scale between the first coordinate system and the second coordinate system. The transformation rule can be determined, for example, using singular value decomposition. It is possible to determine a transformation rule comprising an error covariance matrix, whereby the transformation rule can be determined, for example, using a Gauss-Helmert model.

[0075] The transformation rule can be determined, for example, with respect to the mean values ​​of the respective first and second Gaussian distributions 11, 12 and / or with respect to covariance matrices of the respective first and second Gaussian distributions 11, 12. Taking the covariance matrices into account can be implemented as a weighted determination of the transformation rules, using, for example, the inverses of the covariance matrices as weighting factors.

[0076] It is possible for steps S5 and S6 to subsequently be performed again at least once to further reduce the spatial offset between the first Gaussian distributions 11 and the transformed second Gaussian distributions 13. In particular, it is possible for the first Gaussian distributions 11 to be assigned, at least in part, different second Gaussian distributions than in the first iteration of this step. In particular, each time this step is performed, it is also possible for a second Gaussian distribution 12, 13 to be assigned to more than one first Gaussian distribution 11 as the closest Gaussian distribution.

[0077] If, for example, a minimum of a distance described by the distance criterion between all or at least some of the Gaussian distributions has been reached, the second vehicle position 10 is determined in a step S7 based on the first vehicle position 9 and the one or more transformation rules determined in step S6. Before determining the nearest Gaussian distributions in step S5, an initial transformation can be applied to at least one of the coordinate systems in an optional step S1 to improve the convergence of the method. The initial transformation can be determined as a function of movement information describing the movement of the vehicle 1 from the first vehicle position 9 to the second vehicle position 10.The initial transformation can be applied in particular before the first execution of step S5 and allows at least partial compensation of the offset between the first coordinate system and the second coordinate system.

[0078] The movement information can be determined, for example, by means of odometry and / or by means of a navigation method, in particular by means of satellite navigation. The movement information can be determined, for example, by the control device 6. To determine the movement information, the control device 6 can be connected, for example, to an odometry device (not shown) of the vehicle 1 that provides odometry data and / or to a navigation device (not shown) of the vehicle 1 that provides navigation data. Alternatively, it is possible for the control device 6 to comprise an odometry device and / or a navigation device.

[0079] To increase the speed of the process or reduce the number of required iteration steps, the assignment of the nearest Gaussian distributions can additionally or alternatively be performed based on at least one assignment criterion. The assignment criterion can be, for example, a ratio between the amplitudes of the measured values ​​described by the Gaussian distributions, a ratio between directions assigned to the measured values ​​described by the Gaussian distributions, and / or a ratio between the curvatures of the objects described by the Gaussian distribution.

[0080] The assignment criteria can be used in addition to the distance criterion, or one or more higher-dimensional distance criteria can be used, which, in addition to a spatial distance relative to the x-coordinate and the y-coordinate, also include at least one additional dimension corresponding to an assignment criterion. It is possible to use multiple assignment criteria, which are weighted against each other using weighting factors.

[0081] By using one or more assignment criteria, it becomes more likely that a first Gaussian distribution 11 will be assigned a second Gaussian distribution 12, which includes measured values ​​reflected from the same object, for example, one of the other vehicles 2, 3, or from the same part or section of an object. Since the properties of the object influence the echoes reflected from the object, properties of the echoes can be considered to determine suitable Gaussian distributions or to avoid obvious or at least very likely misassignments.

[0082] Depending on the assignment criteria, it can therefore be decided whether, for a first Gaussian distribution 11, instead of a second Gaussian distribution 12 having the smallest distance described by the distance criterion, another second Gaussian distribution 12, for example the second smallest distance, is selected as the closest Gaussian distribution if the second Gaussian distribution 12 having the smallest distance has a degree of agreement that is assessed as insufficient for one or more of the assignment criteria.For example, it is possible for several assignable second Gaussian distributions 12 to lie within a predetermined distance range around a first Gaussian distribution 11, wherein the closest Gaussian distribution is selected from this group of second Gaussian distributions 12 depending on the at least one assignment criterion, in particular depending on the degree of agreement between the properties of the Gaussian distributions and / or their measured values ​​considered as assignment criteria. Alternatively, other assignment specifications can also be used in relation to the assignment criteria and the distances between the Gaussian distributions described by the respective distance criterion.

[0083] The ratio between the amplitudes of the measured values ​​described by the Gaussian distributions can be used as the assignment criterion. A second Gaussian distribution 12 can be assigned as the closest Gaussian distribution to a first Gaussian distribution 11 if the amplitudes are as similar as possible or if the amplitude ratio of the measured values ​​lies within a specified interval. The amplitudes of individual measured values ​​or the averaged amplitudes of all measured values ​​forming the respective Gaussian distribution 11, 12 can be considered.

[0084] Additionally or alternatively, the relationship between directions assigned to the measured values ​​described by the Gaussian distributions 11, 12 can also be considered as an assignment criterion, with an assignment occurring in particular when the measured values ​​are assigned the same or at least a similar direction. The direction can, if necessary, be corrected based on the or further movement information describing the movement of the vehicle 1 in order to take into account a relative change in direction due to the vehicle movement.

[0085] In particular, the azimuth angle of the sensor detection area 8 in the xy plane, at which the echoes forming the measured values ​​are received by the sensor 7, can be used as the direction. The directions of individual measured values ​​or the averaged directions of all measured values ​​forming the respective Gaussian distribution can be considered. A similar direction can be assumed, for example, if the ratio of the directions or the respective azimuth angles lies within a specified interval.

[0086] Additionally or alternatively, it is also possible to use the ratio between the curvatures of the objects described by the Gaussian distribution as the assignment criterion. The curvature refers to the shape of the object or to the shape of a side surface of the object facing the sensor 7. In particular, second Gaussian distributions 12 can be considered as possible closest Gaussian distributions, which describe objects or object sections with the same or a similar curvature, i.e., curvature lying within a predetermined curvature interval. It is possible for at least some of the first Gaussian distributions 11 and at least some of the second Gaussian distributions 12 to be assigned a label, wherein only those second Gaussian distributions 12 which have a matching label are used to determine the closest second Gaussian distribution 12 to a first Gaussian distribution 11.For example, the label can describe an object type and / or an object class of an object described by a Gaussian distribution. For example, the label can indicate that the object is a third-party vehicle 2, 3 or the boundary of a parking space 4.

[0087] Alternatively, in the preceding exemplary embodiment, the respectively closest first Gaussian distribution 11 can also be assigned to the second Gaussian distributions 12, in particular to each of the second Gaussian distributions 12. The above statements apply analogously to such an assignment. In addition to the two-dimensional Gaussian distributions in relation to the x-coordinate and the y-coordinate, sensor data sets with more than two dimensions, for example three-dimensional or more than three-dimensional, as well as corresponding coordinate systems can be used. In this case, the Gaussian distributions 11, 12 can also be three-dimensional or correspondingly multi-dimensional. In such cases, the transformation rule can advantageously be determined as a covariance matrix, or comprising a covariance matrix.

[0088] The method for locating the vehicle 1 can be carried out, in particular, within the framework of an assisted, semi-autonomous, or autonomous driving maneuver. The control device 6 or another control device of the vehicle or an external computing device communicating with the control device 6 can be used to determine a travel trajectory for a semi-autonomous or autonomous vehicle movement based on the determined second position information. Depending on the determined second vehicle position 10 and / or a trajectory derived therefrom, at least one actuator of the vehicle 1 can be controlled to perform a longitudinal and / or transverse movement of the vehicle 1.Additionally or alternatively, a display device designed to display information to a driver of the vehicle 1 can also be controlled as a function of the second vehicle position 10, in particular to display information determined as a function of the second vehicle position 10.

Claims

Patent claims 1. Method for localizing a vehicle (1) from sensor data sets describing an environment of the vehicle (1) and recorded using at least one sensor (7) of a sensor device (5) of the vehicle (1), wherein the sensor data sets each comprise several Gaussian distributions (11, 12), wherein the Gaussian distributions (11, 12) are each formed from at least one measurement point of the sensor device (5) and each describe at least a part of an object in the environment of the vehicle (1) with respect to a coordinate system, comprising the steps: - Recording an initial sensor data set at a first vehicle position (9), wherein the first sensor data set comprises several first Gaussian distributions (11) in a first coordinate system, - Acquiring a second sensor data set after a movement of the vehicle (1) , wherein the second sensor data set comprises several second Gaussian distributions (12) in a second coordinate system, - Assigning the nearest first Gaussian distribution (11) to the second Gaussian distributions (12) or the nearest second Gaussian distribution (12) to the first Gaussian distributions (11), wherein the assignment is carried out by evaluating at least one distance criterion describing a distance between two Gaussian distributions, - Determine at least one transformation rule comprising a translation and / or a rotation between the first coordinate system and the second coordinate system, minimizing the distances between the nearest Gaussian distributions described by the distance criterion (11 , 12), - Determining a second vehicle position (10) from the first vehicle position (9) and the at least one Transformation regulation. A method according to claim 1, characterized in that the Gaussian distributions in the first sensor data set and the second sensor data set each describe an ellipse encompassing one or more measurement points. A method according to claim 1 or 2, characterized in that the Gaussian distributions (11, 12) are each at least two-dimensional, wherein a distance criterion is used which specifies at least one distance with respect to the at least two dimensions. A method according to any one of the preceding claims, characterized in that a Mahalanobis distance, a Bhattacharyya distance, a Kullback-Leibler distance, and / or a probability-based distance between two Gaussian distributions (11, 12) is used as the distance criterion, wherein the probability-based distance specifies the probability that is assigned to the mean of one of the Gaussian distributions (11, 12) in the other Gaussian distribution (11, 12).Method according to one of the preceding claims, characterized in that the assignment of the nearest Gaussian distributions (11 , 12) is carried out depending on at least one assignment criterion, wherein the assignment criterion is a ratio between the amplitudes of the measured values ​​described by the Gaussian distributions (11 , 12), a ratio between directions which are assigned to the measured values ​​described by the Gaussian distributions (11 , 12), and / or a ratio between the curvatures of the objects described by the Gaussian distribution (11 , 12).

6. Method according to one of the preceding claims, characterized in that several allocation criteria are used, wherein the allocation criteria are weighted among themselves.

7. Method according to one of the preceding claims, characterized in that at least a part of the first Gaussian distributions (11) and at least a part of the second Gaussian distributions (12) are each assigned a label, wherein for the determination of the nearest Gaussian distribution (11, 12) only those Gaussian distributions (11, 12) which have a matching label are used.

8. Method according to one of the preceding claims, characterized in that for the determination of the nearest Gaussian distribution (11 , 12) only Gaussian distributions (11 , 12) are taken into consideration whose mean lies within a multiple of the variance of the Gaussian distribution (11 , 12) to which the nearest Gaussian distribution (11 , 12) is determined.

9. Method according to one of the preceding claims, characterized in that the transformation rule is determined in relation to means of the respective Gaussian distributions (11 , 12) and / or in relation to covariance matrices of the respective Gaussian distributions (11 , 12).

10. Method according to one of the preceding claims, characterized in that, in particular by means of singular value decomposition, a transformation rule comprising a scaling is determined.

11. Method according to one of the preceding claims, characterized in that, in particular by means of a Gauss-Helmert model, a transformation rule comprising an error covariance matrix is ​​determined.

12. Method according to one of the preceding claims, characterized in that, before determining the nearest Gaussian distributions (11 , 12) an initial transformation is applied to at least one of the coordinate systems, wherein the initial transformation is determined as a function of motion information describing the motion, wherein in particular the motion information is determined by means of odometry and / or by means of a navigation procedure, in particular by means of satellite navigation.

13. Sensor device comprising at least one sensor (7) and at least one control device (6), wherein the control device (6) is configured to carry out a method according to one of the preceding claims.

14. Vehicle comprising a sensor device (5) according to claim 13.

15. Computer program product comprising instructions that a cause the control device (6) to execute a method according to any one of claims 1 to 12.

Images