Method for detecting objects for a motor vehicle

Abstract

The invention relates to a method for detecting objects for a motor vehicle (2), wherein at least one environmental sensor (7, 9, 18) of the motor vehicle (2) detects at least a partial area of ​​the surroundings of the motor vehicle (2), and in the process, first sensor data (19) are generated, wherein on the basis of the first sensor data (19) a plurality of primary data points (dk, dm) are generated in a digital map (30), each of which corresponds to a detection of at least one partial object at a specific location in said surroundings, wherein a grid (32) is generated in the digital map (30), which comprises a plurality of grid cells (ak), and the primary data points (dk, dm) are each assigned to a grid cell (ak), wherein at least for a subset of the plurality of grid cells (ak) a grid data point (pd, pj, pk) is created, for which purpose on the basis of the primary data points (dk, dm) contained in the respective grid cell (ak)dm) at least one piece of information (I(k)) relating to a position in the digital map (30) and to a density of the primary data points (dk, dm) in the respective grid cell (ak) is determined, wherein a density-based clustering method is applied to the grid data points (pd, pj, pk), which uses the respective said information (I(k)) relating to the density of the primary data points (dk, dm), and from this a cluster of grid data points (pd, pj, pk) is identified which corresponds to an object in the said environment.

Description

[0001] The invention relates to a method for detecting objects for a motor vehicle, wherein at least a partial area of ​​an environment of the motor vehicle is detected by at least one environment sensor of the motor vehicle, and first sensor data are generated in the process, wherein a plurality of primary data points are generated in a digital map on the basis of the first sensor data, each of which corresponds to a detection of at least one partial object at a specific location in said environment.

[0002] For the increasing automation of motor vehicle driving functions, it is desirable to have the most detailed digital maps possible of a vehicle's surroundings. To improve automation, precise detection of structural boundaries in the surroundings, such as walls or pillars, is particularly necessary. Precise detection of vehicles, especially stationary vehicles, in the surrounding area can, in turn, be used to improve the mapping of individual parking spaces in parking areas in the vicinity of the vehicle.

[0003] Individual objects (e.g. walls, pillars, and vehicles) can be recognized using data generated by environmental sensors on motor vehicles. Each environmental sensor scans the area surrounding the vehicle and uses this data to generate data points in the space surrounding the vehicle, which can then be recorded in a digital two- or three-dimensional map of the environment. For actual object recognition, the data points are usually checked to determine the extent to which spatially adjacent data points can represent a common object. The data points that belong together and represent a specific object are examined for the specific spatial dimensions of the respective detected object, and the associated information about the boundaries of the object is then used to create the digital map.

[0004] Radar sensors are typically used as environmental sensors. These emit a measurement signal in a specific direction and determine the reflections from that direction with distance resolution, for example, by frequency modulating the measurement signal, so that the distance can be determined based on changes in the frequency modulation in the received signal. However, such methods for measuring the distance of objects can be subject to comparatively strong sensor noise, which means that the determination of the distance of a data point belonging to an object is also subject to a non-negligible amount of noise. To improve data quality, the measured data points are accumulated over time using odometry.In order to prevent the amount of data to be processed from growing too quickly, the data points are recorded in a grid that is placed over the digital map, and data points from individual grid cells are viewed as common object points and processed accordingly.

[0005] This allows the overall number of data points to be processed to be kept to a minimum. However, this also results in the loss of potentially important information regarding the individual original data points of a given grid cell.

[0006] CN 108 256 577 A describes cluster processing of lidar data, which involves grid filtering the data points, filtering out individual raw points, and correcting any erroneous raw points. Density-based DBSCAN clustering is then applied to the grid-filtered point clouds for object detection.

[0007] CN 109 101 892 A describes a laser radar target acquisition method based on a grid and density clustering algorithm. The data points are projected onto a grid superimposed on the digital map. It is then assessed whether and at which cells the grid is "dense," and the sparsely populated grid cells are deleted. The grid cells deemed dense are replaced by representative points, to which a density-based clustering method is applied.

[0008] CN 115 861 966 A discloses a method for obstacle detection using lidar data from a motor vehicle. The lidar data is associated with the position data of an integrated navigation system. Individual point clouds are determined based on the real-time positioning of the vehicle, and overlapping point clouds are downscaled to a local feature point cloud, to which a density-based clustering method is applied for obstacle detection.

[0009] CN 114 266 801 A discloses a method for ground segmentation for a mobile robot in a terrain environment based on three-dimensional laser radar. The method comprises the following steps: meshing the obtained point cloud data, finding a maximum density area, determining a maximum density point as a starting point for clustering, finding a core point in the point cloud according to a density clustering algorithm, and clustering the core point; and searching for reachable density points in the surrounding area according to the clustering radius of each core point, using the starting point as the first point of density clustering; finally, terminating clustering and judging points that cannot be clustered as outliers for filtering. Downsampling processing is performed on point clouds using voxel filtering based on a neighboring centroid.Unordered point clouds are sorted by a radius and a vertical angle. The expected distance under actual road conditions is determined by the relationship between the slope angle and the vertical angle. The local threshold height and the global threshold height are determined by the relationship between the expected distance, the radar installation height, and the slope angle, and the global threshold height is calculated. By assessing the radius of each point and the actual expected distance, as well as the height of each point and the global threshold and local threshold heights, it is determined whether the point belongs to the ground.

[0010] The object of the invention is to improve the aforementioned methods with regard to their performance and, in particular, to enable object recognition that is both precise and resource-efficient.

[0011] The stated object is achieved according to the invention by a method for detecting objects for a motor vehicle (motor vehicle), wherein at least a partial area of ​​an environment of the motor vehicle is detected by at least one environment sensor of the motor vehicle, and first sensor data is generated in the process, wherein on the basis of the first sensor data a plurality of primary data points are generated in a digital map, each of which corresponds to a detection of at least one partial object at a specific location in said environment, and wherein a grid is generated in the digital map, which comprises a plurality of grid cells, and the primary data points are each assigned to a grid cell.

[0012] According to the method, a grid data point is created for at least a subset of the plurality of grid cells, for which purpose at least one piece of information on a position in the digital map and on a density of the primary data points in the relevant grid cell is determined on the basis of the primary data points contained in the relevant grid cell, and that a density-based clustering method is applied to the grid data points, which uses the respective said information on the density of the primary data points, and from this a cluster of grid data points is recognized which corresponds to an object in the said environment, wherein at least for some grid data points the respective information on the position in the digital map is determined on the basis of a geometric center of gravity of the primary data points in the respective grid cell.Advantageous and partly inventive embodiments are the subject of the dependent claims and the following description.

[0013] A motor vehicle in this case particularly includes any vehicle with an internal combustion engine in the drive train and / or an electric drive (i.e. also a vehicle with a hybrid drive), whereby a priori any of the usual vehicle sizes can be present, i.e. in particular a car or a truck. The surroundings of the motor vehicle in this case particularly include the immediate surroundings which at a given point in time are formed by the field of vision of the vehicle (i.e. the totality of objects which are visible from a point on the vehicle). According to this definition, a detected object can preferably still be in the surroundings of the motor vehicle if it temporarily disappears from the field of vision of the vehicle (e.g. because it is temporarily obscured by another object such as a pillar or another stationary vehicle), or if it was still in the field of vision up to a point in time which is only insignificantly in the past, i.e. preferably e.g.60 seconds, preferably 30 seconds.

[0014] An environmental sensor here includes any device that is configured to detect the surroundings of the motor vehicle using physical means, i.e. in particular with the aid of electromagnetic waves, and from this to determine at least a distance from the sensor for individual objects in the surroundings. The environmental sensor is preferably configured to generate a two- or three-dimensional point cloud from the determined distances of objects in the surroundings of the vehicle. In particular, an environmental sensor here is an active sensor in that the electromagnetic radiation used to detect objects and / or distances to objects is emitted by the sensor itself or by an auxiliary device associated with the sensor, for example a radar or lidar sensor.However, a sensor that determines distances to objects based on ultrasound, structure-from-motion, VisualSLAM, and / or direct sparse odometry can also be considered as an environmental sensor.

[0015] The at least one environment sensor detects at least a partial area which covers a polar angle range of preferably at least one quadrant, particularly preferably at least two quadrants (i.e. a half-space), and in particular the entire azimuthal space around the motor vehicle.

[0016] A primary data point in a digital map, which corresponds to a detection of at least one partial object at a specific location in the surroundings of the motor vehicle, is to be understood in particular as meaning that the environment sensor initially detects at least part of an object as being “present” at a definable point in the surroundings, without the object being completely detected in terms of its dimensions or type, and that the detected part of the object is assigned the location or point in the surroundings and is represented by the data point in the digital representation of the surroundings, which the digital map represents.

[0017] In other words, when any part of any object is detected by the environment sensor, the point in the environment detected by the environment sensor is assigned, and this point is recorded as a data point in the digital map. To generate a data point, the initial sensor data can be transformed from polar coordinates in the vehicle's reference system into stationary Cartesian coordinates (of the environment). Driving data (current instantaneous driving speed and direction of travel) are also used for this purpose. This allows for the location and / or movement information of the vehicle to be taken into account for a distance from the said partial object detected by the environment sensor in a specific angular direction.

[0018] Creating a grid in the digital map particularly includes dividing the preferably Cartesian digital map, in which the primary data points are recorded, into individual grid cells of preferably equal dimensions, thus superimposing a preferably uniform (or uniformly meshed) "grid" over the digital map. The individual grid cells (which in particular do not overlap) thus preferably cover at least the aforementioned partial area of ​​the vehicle's surroundings.

[0019] The primary data points, each of which is recorded at a specific location in the digital map, can thus be uniquely assigned to a specific grid cell (with the possible exception of primary data points that lie exactly on a grid line separating the grid cells; however, this special case can be neglected due to its rarity).

[0020] A grid data point is understood, in particular, to be a data point in the digital map that lies within the grid cell associated with the grid data point and that represents all primary data points located in the grid cell. In other words, all primary data points of a grid cell are preferably mapped to the grid data point whose position lies further within the grid cell, and which represents the said primary data points for the subsequent density-based clustering process and, if applicable, for any upstream or downstream processing steps. The grid data points can thus be understood as a type of secondary data point, each derived from the primary data points.

[0021] For this representation of the primary data points of a grid cell by the associated grid data point, the latter is linked to (or contains) information which at least indicates its concrete position in the digital map (i.e. in particular its position within its grid cell) and also makes a statement about the density of the primary data points of the grid cell, which is preferably monotonic in the number of primary data points of the grid cell.

[0022] This means, in particular, that a grid data point contains information about its position in the grid cell (and thus in the digital map), which can be determined based on the positions of the primary data points in the grid cell as the center of gravity, and about the density of the primary data points within the grid cell. The latter information can, in particular, be formed monotonically based on the mere number of said primary data points. However, this density information can also take into account the distribution of the primary data points within the grid cell and assign a higher value to a more uniform distribution (with respect to a suitable dispersion or distribution measure) than to a more concentrated distribution of the primary data points in the grid cell.

[0023] The aforementioned generation of grid data points based on the primary data points contained in the associated grid cell is performed at least for a subset of all grid cells. In particular, this subset can include those grid cells that each contain at least two primary data points. In particular, grid cells with only one primary data point can be assigned as a grid data point. Preferably, no grid data point is generated for grid cells that do not have a primary data point, so that grid cells without primary data points are not considered for density-based clustering.

[0024] A density-based clustering method such as DBSCAN or OPTICS is applied to the grid data points obtained in this way in order to identify related objects in the surroundings of the vehicle, whereby the information on the density of the primary data points assigned to each grid data point can be taken into account.

[0025] The density of the grid data points is determined based on neighborhoods (so-called “ε-neighborhoods”) around the individual grid data points, and on the number of other grid data points within such a neighborhood, i.e. within a distance of ε.

[0026] Within the framework of the method, a core distance can be determined for each of the grid data points, which is given by the smallest neighborhood of the respective grid data point, which contains a predetermined minimum number of grid data points that were weighted according to the aforementioned density information. Thus, a minimum number minPts is specified for the evaluation of the grid data points (e.g., minPts := 5 or minPts := 6), and then, for each individual grid data point pj, the density information is first processed, for example, in the form of a function fk(pk) of the corresponding grid data point. Subsequently, the distance or the ε-neighborhood is determined within which the predetermined number minPts of grid data points fk(pk) weighted as described lies (including the grid data point pj to be considered), i.e. min ε:∑k fk(pk)≥minPts with|pk−pj|≤ε

[0027] Based on these ε-neighborhoods of the grid data points, the density-based clustering procedure (hereinafter also referred to as “clustering”) can then be carried out.

[0028] The advantage of the approach described here is that by combining various primary data points into grid data points, a significant reduction in the number of data points to be processed by clustering can initially be achieved. Even if, for example, each of the grid data points is only weighted by the number of its underlying primary data points as a multiplicity in the clustering, a comparable clustering result can be expected as with clustering of the primary data points, although a significant reduction in numerical complexity can already be achieved due to the smaller number of points. In addition, additional information about the grid data points, such as the signal-to-noise ratio (SNR), can be taken into account when determining and weighting the primary data points.

[0029] Furthermore, the method becomes even more precise by taking into account the density of primary data points within the individual grid cells that underlie the grid data points. This reduces the ε-neighborhoods, in particular, because a given number of minPts can be achieved with just a few grid data points due to their weighting. Using smaller ε-neighborhoods, in turn, reduces the probability of erroneously grouping unrelated data points into one object, as could potentially be the case with larger ε-neighborhoods (e.g., with unweighted grid data points).

[0030] In order to obtain individual object clusters, the grid data points can be clustered according to their minimum core distance, starting from a selected starting point, as in the DBSCAN method, or ordered according to their so-called “reachability distances” (which are calculated using the core distances, among other things), as in the OPTICS method.

[0031] On the one hand, the process can be carried out entirely in a single vehicle, so that the digital map only contains the detections of the respective vehicle. The vehicle can also transmit a fully identified object (including all calculated information on its position in the digital map, its dimensions, and, if applicable, the object type) to a server, e.g., a backend of the vehicle manufacturer or similar, in which all objects fully identified by all participating vehicles are entered into a digital master map. This digital master map, after post-processing of the identified objects using the proposed process (and, if necessary, further post-processing steps), is then made available to all participating vehicles.

[0032] Preferably, at least for some grid data points, the number of primary data points is used for the respective information on the density of the primary data points in the respective grid cell. While said information on the density of the primary data points can also be determined based on their spatial distribution or spatial concentration within their grid cell, a function of the number is easier to calculate and sufficiently precise for the purposes of the method, especially for small grid cells (with an edge length between 5 cm and 15 cm, for example).

[0033] The number Nk of primary data points dk in the grid cell k can be used directly as said information, or a function fk(Nk) can be used, which is preferably monotonic in Nk, e.g. a product of Nk with a weighting factor, which can take into account further information about the primary data points in the respective grid cell.

[0034] According to the invention, the respective position information in the digital map is determined for at least some grid data points based on a geometric center of gravity of the primary data points in the respective grid cell. In particular, said center of gravity can be used directly as position information. While this information could also be provided by the center of the grid cell, for example, the center of gravity accounts for the spatial distribution in a mathematically simple manner. Furthermore, updates, i.e., new primary data points, can be integrated into the grid cell and thus into the grid data point with minimal complexity.

[0035] Advantageously, a DBSACN algorithm (“Density-Based Spatial Clustering of Applications with Noise”) and / or an OPTICS algorithm (“Ordering Points To Identify the Clustering Structure”) is applied to the grid data points as the density-based clustering method. With a DBSCAN algorithm, for each data point to be processed (i.e., in this case, for each grid data point), it is determined whether a specified minimum number of data points (minPts) is contained within the ε-neighborhood (with a specified distance parameter ε) (whereby, in this case, the grid data points are weighted according to their density information, as described). The (grid) data points for which this is the case are “dense” in the sense of the algorithm. Clusters thus occur within the coverage of dense data points.In the OPTICS algorithm, a core distance is determined for each data point to be processed (i.e., in this case, for each grid data point). This distance corresponds to the distance parameter ε of the smallest possible ε-neighborhood containing the specified minimum number of minPts data points (or, in this case, weighted grid data points). A so-called "reachability distance" is determined from the maximum of the core distance and the real Euclidean distance between two points to be compared (i.e., in this case, two grid data points). The data points are then sorted, starting from a starting point, according to their next smallest reachability distance. Due to the consideration of the ε-neighborhoods and the nature of the initial sensor data, both methods are particularly suitable for density-based clustering of the available grid data points.

[0036] It proves to be further advantageous if, in order to determine a number of data points within a reference distance of the clustering method around a given data point, the grid data points are each weighted based on and in particular with the said number of primary data points as a multiplicity. This means in particular that the information on the density of the primary data points, which is attached to each of the grid data points, is given by the number of primary data points or by a preferably linear function of this number, and that, in order to determine the minimum number of data points within the reference distance (i.e. the ε-neighborhood), this number or function of the number is counted as the multiplicity for each grid data point. If, for example, a grid data point pk is based on three primary data points dk in the grid cell k, the grid data point with the multiplicity Nk = 3 can be taken into account in the corresponding clustering.

[0037] In an advantageous embodiment, at least for some grid data points, information on the SNR of at least one primary data point in the relevant grid cell is used in the density-based clustering method. This can be achieved, in particular, by weighting the number of primary data points for the density information of the grid data point depending on the SNR (for example, by a factor dependent on the SNR).

[0038] Conveniently, a grid SNR is assigned to each grid data point based on the average SNR of the primary data points in the respective grid cell and / or based on the maximum SNR across the primary data points in the respective grid cell. While the individual primary data points dk of a grid cell k can also be weighted according to their respective SNR (where a better SNR leads to a higher weighting, e.g., by a factor such as SNR (dk) / (SNR (dk) + 1)), using the average or maximum SNR across all primary data points of the grid cell is less complex.

[0039] Preferably, the information on the density of the primary data points is determined for the relevant grid data points by weighting the number Nk of primary data points dk in the respective grid cell k based on the grid SNR. This can be achieved, in particular, by a multiplicative factor Fsnr(k) for the number Nk, which depends monotonically increasing on the grid SNR and approaches 1 for increasing grid SNR, i.e., in particular, Fsnr(k) Nk as the "multiplicity" of the grid data point pk.

[0040] It has proven further advantageous if the method is iterated by adding at least one new primary data point to a grid cell in the digital map based on the first sensor data, generating or updating a or the associated grid data point based on the new primary data point, and identifying the cluster of grid data points that corresponds to an object in the surroundings of the motor vehicle based on an updated set of all grid data points. In other words, the first sensor data is repeatedly and, in particular, continuously collected, so that primary data points are generated in the same area of ​​the digital map at different points in time. Accordingly, primary data points that are generated at a time at which primary data points, and thus also grid data points, already existed in the digital map orpresent in the grid are added to the existing data set of (older) primary data points and corresponding grid data points.

[0041] In particular, a new grid data point is generated when a newly added primary data point falls into a previously empty grid cell. Preferably, an existing grid data point is updated using the new primary data point, for example by recalculating the centroid (as information on the position of the grid data point in question in the digital map) and the information on the density of the primary data points (by increasing the multiplicity and, if necessary, changing the weighting depending on the SNR of new primary data points). The clustering can then be repeated using the updated (or newly generated) grid data points, or new grid data points from the update can be added to an existing cluster depending on cluster characteristics (such as core distances, etc.), which are calculated using the respective new grid data point.

[0042] Advantageously, the first sensor data is generated using a radar and / or a lidar as the at least one environmental sensor. These sensors exhibit high accuracy (especially with regard to distance and angular direction) in detecting the data points and are cost-effective.

[0043] The invention further relates to a motor vehicle comprising at least one first environmental sensor and a control unit, wherein the motor vehicle is configured to carry out the method described above. In particular, the control unit is configured to carry out the calculations performed in the method by means of a processor and a working memory addressable by the processor, as well as by means of corresponding program instructions in the working memory or in a non-volatile memory.

[0044] The motor vehicle according to the invention shares the advantages of the method according to the invention. The advantages stated for the method and its further developments can be applied analogously to the motor vehicle.

[0045] An embodiment of the invention is explained in more detail below with reference to the accompanying drawings, each of which shows schematically: Fig. 1 a motor vehicle with a vehicle sensor system comprising several radar sensors for detecting data points, and Fig. 2 using various processing steps in a digital map, a method for detecting objects the vehicle according to Fig. 1.

[0046] Corresponding parts and sizes are provided with the same reference numerals in all figures.

[0047] Fig. 1 schematically shows a motor vehicle 2 with a vehicle sensor system 4. The vehicle sensor system 4 is designed as a vehicle sensor system, by means of which a surroundings of the vehicle 2 (not shown in detail) are monitored. The vehicle sensor system 4 has an environment sensor 7 designed as a front radar 6 and two further environment sensors 9 each designed as rear radars 8. In the Fig. 1, the radar areas 10, 12 of the front radar 6 and the rear radar 8 are also schematically shown. In addition, the vehicle sensor system 4 has further environmental sensors 18, with two such environmental sensors 18 being arranged at the front and two at the rear. The environmental sensors 18 are also designed as radar sensors, but can also be designed as lidar sensors. The vehicle sensor system 4 can also have further Fig. 1 may have sensors not shown, which may be provided, for example, by one or more vehicle cameras, ultrasonic sensors, etc., which, however, do not play a direct role in the present embodiment for the method described above.

[0048] The individual environmental sensors 7, 9, 18 (and possibly the additional sensors not shown in detail) of the vehicle sensor system 4 are controlled by a controller 14, which in particular controls the transmission of radar signals by the front radar 6 and the rear radars 8 (and of additional radar signals and / or laser pulses from the environmental sensors 18) and receives the associated sensor signals as first sensor data 19. The sensor signals or the first sensor data 19 are forwarded, possibly after low-threshold preprocessing such as analog preamplification and / or digitization in the controller 14, to an evaluation unit 20, which is connected to the controller 14. However, the evaluation unit 20 can also be connected directly to the radar and the additional environmental sensors 7, 9, 18, and receive the respective sensor signals from them for evaluation.

[0049] The evaluation unit 20 thus records the sensor signals of the environment sensors 7, 9, 18 and, when driving through the environment, examines the same for the presence of relevant objects (in particular walls, pillars and vehicle). Navigation data 17 of a navigation system 16 are also used in order to be able to assign an absolute position within a digital map, in particular to the motor vehicle 2, and based on this, to the data points generated by the environment sensors 7, 9, 18 (in Fig. 1 not shown) to also be able to assign an absolute position in the digital map.

[0050] In Fig. 2 schematically shows a method for detecting objects for a motor vehicle using various processing steps S1-S6 in a respective digital map 30, for which purpose the first sensor data 19 are preferably Fig. 1. From the first sensor data 19, individual primary data points dm are generated in a first processing step S1 and entered into the digital map 30, which represents the surroundings of the vehicle 2 according to Fig. 1. A grid 32 is placed over the digital map 30, so that the primary data points dm each lie in a grid cell ak (the index ak of the grid cells is preferably two-dimensional, i.e. vector-valued).

[0051] In the next processing step S2, for each of the grid cells ak in which at least one primary data point dm is located, a corresponding grid data point pk for the respective grid cell ak (in Fig.2 only shown for grid cell k), wherein the position of the grid data point pk is determined in this case by the center of gravity of the primary data points pk in the grid cell k (see enlarged section of grid cell k in the digital map 30 in processing step S2). In grid cells ak, in which only one primary data point dm is located, this center of gravity is trivially identical to the position of the respective primary data point dm. In the subsequent processing step S3, for each grid data point pk (see enlarged section of grid cells k-1, k in the digital map 30 in processing step S3), information I(k) (or for grid cell k-1, information I(k-1)) is determined, which provides information about the density of the underlying primary data points dk in the grid cell k.

[0052] This information I (k) can in particular be chosen as the number of primary data points dk in grid cell k, i.e. in this case I (k) = 4 (and correspondingly I (k-1) = 1 for grid cell k-1 with only one primary data point). However, said number can also be weighted based on an SNR of the primary data points dk in grid cell k, i.e. with a factor 0 ≤ Fk ≤ 1, which is the larger the larger the average SNR (across all primary data points dk of grid cell k) or the maximum SNR in grid cell k, i.e. I (k) = Fk (SNR(k)) Nk with the number Nk of primary data points dk.

[0053] In the next processing step S4, the ε-neighborhoods are defined with the reference distance ε (shown only for some grid data points pk) to determine which of the grid data points pk are core points in the sense of the DBSCAN algorithm. For the corresponding counting according to equation (i), each grid data point pk is preferably counted with its associated information I(k) as a "multiplicity" to determine the number of data points within the ε-neighborhood and thus to identify which grid data points pk are "dense" in the sense of the DBSCAN algorithm.

[0054] In the next processing step S5, those grid data points pd, pj (full points in processing step S5) that were not identified as "dense" in the sense of the DBSCAN algorithm in processing step S4 can now be removed or marked as negligible. The other grid data points (ring-shaped points in processing step S5) remain and are now used in processing step S6 to identify a principal direction 34, for example, using a principal component analysis or similar. This can be used to perform further analyses regarding the detection of a specific object, such as a wall (in particular, a dimension 36 of the object to be identified along the principal direction 34).

[0055] Although the invention has been illustrated and described in detail by the preferred embodiment, the invention is not limited to the disclosed examples and other variations may be derived therefrom by those skilled in the art without departing from the scope of the invention. List of reference symbols 2 Motor vehicle (motor vehicle) 4 Vehicle sensor system 6 front radar 7 Environment sensor 8 rear radar 9 Environment sensor 10 radar range 12 radar range 14 controllers 16 Navigation system 17 Navigation data 18 Environment sensor 19 first sensor data 20 Evaluation unit 30 digital cards 32 grids 34 Main direction 36 Dimensions ak grid cell dm, dk primary data point I (k) Information (about density) pk grid data point S1-S6 processing step ε reference distance

Claims

[1] Method for detecting objects for a motor vehicle (2), - wherein at least one environmental sensor (7, 9, 18) of the motor vehicle (2) detects at least a partial area of ​​the environment of the motor vehicle (2), and first sensor data (19) are generated in the process, - wherein, based on the first sensor data (19), a plurality of primary data points (dk, dm) are generated in a digital map (30), each of which corresponds to a detection of at least one partial object at a specific location in said environment, - wherein a grid (32) is generated in the digital map (30), which grid comprises a plurality of grid cells (ak), and the primary data points (dk, dm) are each assigned to a grid cell (ak), - wherein at least for a subset of the plurality of grid cells (ak) a grid data point (pd, pj, pk) is created, for which purpose at least one piece of information (I(k)) is determined on the basis of the primary data points (dk, dm) contained in the respective grid cell (ak). -- a position in the digital map (30) and -- to a density of the primary data points (dk, dm) in the respective grid cell (ak), - wherein a density-based clustering method is applied to the grid data points (pd, pj, pk), which uses the respective said information (I(k)) on the density of the primary data points (dk, dm), and from this a cluster of grid data points (pd, pj, pk) is recognized which corresponds to an object in the said environment, and - wherein at least for some grid data points (pd, pj, pk) the respective information (I(k)) on the position in the digital map (30) is determined on the basis of a geometric center of gravity of the primary data points (dk, dm) in the respective grid cell (ak). [2] Method according to claim 1, wherein at least for some grid data points (pd, pj, pk) the number of primary data points (dk, dm) is used for the respective information (I(k)) for the density of the primary data points (dk, dm) in the respective grid cell (ak). [3] Method according to claim 1 or claim 2, wherein as the density-based clustering method a DBSCAN algorithm and / or an OPTICS algorithm is applied to the grid data points (pd, pj, pk). [4] Method according to claim 3 in conjunction with claim 2, wherein for determining a number of data points within a reference distance (ε) of the clustering method around a given data point, the grid data points (pd, pj, pk) are each weighted by said number of primary data points (dk, dm) as a multiplicity. [5] Method according to one of the preceding claims, wherein at least for some grid data points (pd, pj, pk) information on a signal-to-noise ratio of at least one primary data point (dk, dm) in the relevant grid cell (az) is used in the density-based clustering method. [6] Method according to claim 5, wherein the respective grid data points (pd, pj, pk) are each assigned a grid signal-to-noise ratio based on the average signal-to-noise ratio of the primary data points (dk, dm) in the respective grid cell (ak) and / or based on the maximum signal-to-noise ratio across the primary data points (dk, dm) in the respective grid cell (ak). [7] Method according to claim 6 in conjunction with claim 2, wherein for the respective grid data points (pd, pj, pk) the information (I(k)) on the density of the primary data points (dk, dm) is determined by weighting the number of primary data points (dk, dm) in the respective grid cell (ak) based on the grid signal-to-noise ratio. [8] Method according to one of the preceding claims, wherein the method is iterated by - based on the first sensor data (19), at least one new primary data point (dk, dm) is added to a grid cell (ak) in the digital map (30), - one or the corresponding grid data point (pd, pj, pk) is created or updated based on the new primary data point (dk, dm), and - the cluster of grid data points (pd, pj, pk) corresponding to an object in the surroundings of the motor vehicle (2) is recognized on the basis of an updated set of all grid data points (pd, pj, pk). [9] Method according to one of the preceding claims, wherein the first sensor data (19) are generated using a radar sensor (6, 8) and / or a lidar sensor as the at least one environment sensor (7, 9, 18). [10] Motor vehicle (2) comprising at least one first sensor and one control unit (14, 20), wherein the motor vehicle (2) is configured to carry out the method according to one of the preceding claims.

Images