Device and method for fusing reference profiles

Abstract

A device for fusing reference profiles for a roadway section is described. The device is configured to determine a first and a second reference profile for the roadway section. The device is further configured to identify a first sub-section of the roadway section in which the first and second reference profiles satisfy a predefined parallelism criterion, and to determine a fused reference profile for this first sub-section based on the first and second reference profiles, and to use this fused reference profile instead of the first and second reference profiles for the first sub-section of the roadway section.

Description

[0001] The invention relates to methods and corresponding devices which each make it possible to determine information relating to the lanes of a roadway.

[0002] To partially or fully automate the longitudinal and / or lateral control of a vehicle, it is advantageous to have a high-precision geographic map of the surroundings. A standard definition (SD) map, which can be used, for example, to guide the vehicle to a predetermined destination, typically has an accuracy of approximately one to ten meters. A high definition (HD) map should typically deviate from reality by less than approximately one meter. HD and SD maps can differ beyond accuracy in other ways, such as in the information they contain.

[0003] For example, if the vehicle is traveling on a road with multiple lanes (also called traffic lanes), the HD map can indicate lane boundaries to distinguish which lane the vehicle is in. Visual indicators, such as lane markings, can be used to determine lane boundaries. A lane marking typically consists of a line that is directly applied to the road surface (and which can be detected optically, e.g., by a camera, and / or by sensors).

[0004] Creating an HD map using a survey vehicle is complex. Alternatively, an HD map can be created based on sensor-based detection of lane boundaries by a fleet of vehicles that are already using the roadway.

[0005] An HD map can thus specify the lane markings between one or more lanes on a roadway. This information can be used by a driving function to guide a vehicle, at least partially or fully automatically, along the roadway, particularly along a lane, longitudinally and / or laterally. For this purpose, a target trajectory for the vehicle can be determined based on the map data from the HD map. The target trajectory can, for example, run centrally between the lane markings displayed by the map data. The vehicle can then be automatically guided longitudinally and / or laterally along the determined target trajectory.

[0006] Using lane markings to determine a target trajectory can lead to a trajectory that, while running between different lane markings, does not actually follow an available lane (as can happen, for example, on a motorway without physical separation). Furthermore, certain sections of a roadway may lack sufficient or no sensor-detectable lane markings (such as before a toll plaza, within an intersection, or on a rural road). Consequently, no target trajectory, or an unrealistic one, may be determined. Additionally, a target trajectory determined based on sensor-detected lane markings may be perceived as uncomfortable by the user of an automated guided vehicle.

[0007] This document addresses the technical task of efficiently and precisely determining map data from a digital (HD) map for a roadway, enabling reliable and comfortable automated longitudinal and / or lateral guidance of a vehicle.

[0008] The problem is solved by each of the independent claims. Advantageous embodiments are described, inter alia, in the dependent claims. It should be noted that additional features of a claim dependent on an independent claim, without the features of the independent claim itself or only in combination with a subset of the features of the independent claim, can constitute a separate invention independent of the combination of all features of the independent claim, which can be made the subject of an independent claim, a divisional application, or a subsequent application. This applies equally to technical teachings described in the description, which can constitute an invention independent of the features of the independent claims.It should be noted in particular that the features of the devices and methods described in this document can be combined with one another as desired. In particular, the features of a first method or a first device, individually or in combination, are also applicable to another second method or to another second device.

[0009] According to one aspect, a device for fusing reference profiles for a roadway segment is described. The roadway segment can have one or more different lanes. A roadway segment can, for example, have a length between 20 and 50 meters. Furthermore, a roadway segment can have a uniform direction of travel or two opposing directions of travel. In particular, the lanes of the roadway segment can each have a uniform direction of travel or at least partially opposing directions of travel.

[0010] A roadway section can therefore correspond to an area of ​​a (paved or unpaved) roadway that has one or more (possibly marked or unmarked) lanes. The individual lanes may be separated from each other, at least partially, by a structural measure. Furthermore, the individual lanes may have at least partially the same direction of travel and / or at least partially opposite directions of travel.

[0011] The device is configured to determine a first and a second reference profile for the road section. Each reference profile can be determined using one of the methods described in this document. Alternatively, each reference profile can be determined based on measured travel paths (as described in this document).

[0012] The first and second reference paths can be associated with different section sequences. Each of these different section sequences can include the roadway segment and at least one surrounding roadway segment located directly before and / or after the roadway segment in the direction of travel. Furthermore, the different section sequences can differ from each other in at least one surrounding roadway segment. The different section sequences can be determined, for example, based on a digital map (e.g., an SD card) of the road network in the vicinity of the roadway segment. For example, two or more, or five or more different section sequences can be considered.

[0013] The device is further configured to identify at least a first subsection of the roadway section in which the first and second reference paths meet a predefined parallelism criterion (and thus run at least almost parallel to each other).

[0014] The device can be configured to subdivide the first and second reference alignments of the roadway section into a sequence of support point planes. The individual support point planes can each be arranged perpendicular to the respective reference alignment. Furthermore, the sequence of support point planes (of the respective reference alignment) can be arranged at a corresponding sequence of points along the first and second reference alignments. The successive points can be spaced at a uniform interval, approximately between 1 and 3 meters.

[0015] The parallelism criterion can depend on the relative orientation of corresponding support point planes of the first and second reference lines. The device can be configured, in particular, to identify as a first sub-section a sub-sequence of points from the sequence of points for which the orientations of the corresponding support point planes of the first and second reference lines differ from each other by at most a predefined angular threshold. The angular threshold can be, for example, 10° or less.

[0016] The first and second reference paths can be associated with opposite directions of travel. When considering the orientations of the corresponding support point planes of the first and second reference paths, the directions of travel are preferably disregarded.

[0017] In particular, a 180° rotation of orientation (due to an opposite direction of travel) is preferably disregarded.

[0018] Alternatively or additionally, the device can be configured to subdivide (only) the first reference profile of the roadway section into a sequence of support point planes. The individual support point planes can each be arranged perpendicular to the first reference profile. Furthermore, the sequence of support point planes can be arranged at a corresponding sequence of points along the first reference profile.

[0019] The device can be configured to determine, for each support point plane from the sequence of support point planes, an intersection point of the second reference line with the respective support point plane. The parallelism criterion can then be determined in a particularly precise and robust manner based on the orientation of the individual support point planes relative to the orientation of the second reference line at the corresponding intersection points.

[0020] The device can be configured, in particular, to identify as a first sub-section a sub-sequence of points from the sequence of points for which the orientations of the corresponding support point planes and the orientations of the second reference track at the corresponding intersection points differ from each other by no more than a predefined angular threshold value. The angular threshold value can be 10° or less. When determining the orientation, the direction of travel of the respective reference track can be disregarded (in particular, for determining the parallelism criterion, it can be assumed that the two reference tracks each have the same direction of travel).

[0021] The device is further configured to determine a fused reference profile for the first sub-section based on the first and second reference profiles, and to use it instead of the first and second reference profiles for the first sub-section of the roadway section.

[0022] The fused reference path can, for example, be provided as map data for the first segment of the roadway for a digital map. Alternatively or additionally, the fused reference path can be used for automated longitudinal and / or lateral guidance of a vehicle (within the roadway). Alternatively or additionally, one or more lane markings in the first segment of the roadway can be detected based on the fused reference path (e.g., using one of the methods described in this document). Alternatively or additionally, one or more lane boundaries and / or one or more lanes in the first segment of the roadway can be detected based on the fused reference path.

[0023] By merging reference profiles in one or more subsections of a roadway, map data for an HD map can be determined in a particularly resource-efficient manner.

[0024] The device can be configured to identify at least a second subsection of the roadway where the first and second reference paths do not meet the predefined parallelism criterion. The first and second reference paths can then be retained for the second subsection.

[0025] The device can thus be configured to determine a first and a second composite reference profile for the road section, which • show the common merged reference curve for the first subsection; and • for one or more other sub-areas of the roadway (in which the parallelism criterion is not met) show the first or the second reference profile respectively.

[0026] Thus, an initial composite reference history can be provided, which • for the first subsection exhibits the common merged reference curve; and • for one or more other sub-areas of the roadway (in which the parallelism criterion is not met) each has the first reference profile.

[0027] Furthermore, a second composite reference history can be provided, which • for the first subsection exhibits the common merged reference curve; and • for one or more other sub-areas of the roadway (in which the parallelism criterion is not met) each has the second reference profile.

[0028] These composite reference profiles can be provided as map data and / or used within one of the procedures described in this document.

[0029] As explained above, the device can be configured to subdivide the first reference profile in the first subsection into a sequence of support point planes. For the first reference profile and the second reference profile, a sequence of intersection points of the respective reference profile with the corresponding sequence of support point planes can be determined.

[0030] Furthermore, for each sequence of control point planes, a control point can be determined based on the identified intersections with the respective control point plane. The control point for a control point plane can be determined as the mean, or in particular as a weighted mean, of the identified intersections with the respective control point plane.

[0031] The fused reference path can then be determined in a particularly precise manner based on the sequence of control points for the corresponding sequence of control point levels, especially by sequentially connecting the control points with path segments (as exemplified in connection with Fig. 2d described).

[0032] The device can be configured to determine a first set of measured routes, which were used to determine the first reference route, and a second set of measured routes, which were used to determine the second reference route. The weights for determining the weighted average can be determined based on the first set of measured routes and based on the second set. This further increases the accuracy of the fused reference route.

[0033] Alternatively or additionally, the device can be configured to iteratively fuse at least one subsection of each pair of reference profiles from a set of reference profiles for the roadway section until a termination criterion is reached, in order to determine a fused reference profile. The fused reference profile determined in one iteration can then be included in the set of reference profiles for the (directly) subsequent iteration.

[0034] For example, in a first step, a set of three or more reference profiles for the road section can be determined. Then, in each iteration, two reference profiles can be selected to merge at least a portion of them. The resulting merged reference profile can be added to the set of reference profiles for the next iteration. In this way, the reference profiles can be merged iteratively, for example, until the parallelism criterion is no longer met for any possible pair of reference profiles.

[0035] It is possible that during an iteration, a merged reference curve is derived from a merged reference curve that was derived in a previous iteration. In particular, due to the iterative approach, it is possible that a merged reference curve is derived from more than two original reference curves.

[0036] The device can be configured to determine the number of (original, unfused) reference curves from which the reference curves to be merged in each iteration are selected. These determined numbers can then be used as weights during the merging process. This further increases the accuracy of the merged reference curves.

[0037] According to another aspect, a device for adjusting reference profiles at a transition between two successive roadway sections is described. The device is configured to determine a first reference profile for a first roadway section and a second reference profile for a second roadway section that directly follows the first (e.g., using one of the methods described in this document).

[0038] The device is further configured to adjust the first reference profile and / or the second reference profile such that the first adjusted reference profile and the second adjusted reference profile have a uniform orientation at the transition.

[0039] The device can be configured, in particular, to determine a first orientation of the first reference curve at the transition and a second orientation of the second reference curve at the transition. A target orientation can then be determined based on the first and second orientations. The target orientation can be determined with particular precision by averaging, in particular by calculating the mean value, of the first and second orientations.

[0040] Furthermore, the first reference curve and / or the second reference curve can be adjusted such that the first adjusted reference curve and the second adjusted reference curve each have the target orientation at the transition.

[0041] The device can be configured to provide the first adapted reference path as map data for the first lane segment and the second adapted reference path as map data for the second lane segment for a digital map. Alternatively or additionally, the device can be configured to use the first adapted reference path and / or the second adapted reference path for automated longitudinal and / or lateral guidance of a vehicle (in the respective lane segment). Alternatively or additionally, the device can be configured to detect and / or determine one or more lane markings in the first lane segment based on the first adapted reference path and / or in the second lane segment based on the second adapted reference path (e.g., using one of the methods described in this document).Alternatively or additionally, the device can be configured to detect one or more lane boundaries and / or one or more lanes in the first carriageway section based on the first adapted reference track and / or in the second carriageway section based on the second adapted reference track.

[0042] By aligning the orientation of the reference profiles at transitions between road sections, the quality of the determined map data can be further improved.

[0043] The device can be configured to determine the first orientation of the first reference profile at an endpoint of the first reference profile and the second orientation of the second reference profile at a starting point of the second reference profile. A transition point can then be determined based on the endpoint and the starting point, in particular as the midpoint of the endpoint and the starting point (or as a midpoint determined from the endpoint and the starting point). Furthermore, the first reference profile and / or the second reference profile can be adjusted such that the first adjusted reference profile at the endpoint and the second adjusted reference profile at the starting point are each perpendicular to a plane passing through the transition point, with the normal vector of the plane having the target orientation. This further improves the accuracy of the fit of the reference profiles.

[0044] It should be noted that the terms "endpoint" and "starting point" are preferably used independently of the direction of travel of the respective reference path. Rather, the endpoint of the first reference path is the final point of the first reference path that faces the starting point (i.e., the final point) of the second reference path.

[0045] The device can be configured to adjust the first reference profile and / or the second reference profile such that the first adjusted reference profile and the second adjusted reference profile lie in a uniform plane at the transition, which has a uniform orientation (in particular the target orientation).

[0046] The device can be configured, in particular, to adjust (e.g., shift) the endpoint of the first reference curve and the starting point of the second reference curve such that the endpoint of the first adjusted reference curve and the starting point of the second adjusted reference curve are located in the plane passing through the transition point. This further improves the accuracy of the fit between the reference curves.

[0047] The device can be configured to determine multiple first reference trajectories for the first roadway section and / or multiple second reference trajectories for the second roadway section. The multiple first reference trajectories can be associated with different section sequences, each comprising the first roadway section and at least one surrounding roadway section located directly in front of and / or directly behind the roadway section in the direction of travel. The different section sequences can differ from each other in at least one surrounding roadway section.

[0048] The device can be configured to adjust the multiple first reference profiles and / or the multiple second reference profiles such that all adjusted reference profiles have a uniform orientation at the transition point. Alternatively or additionally, the multiple first reference profiles and / or the multiple second reference profiles can be adjusted such that all adjusted reference profiles lie in a uniform plane (which has the target orientation) at a common transition point. This further improves the accuracy of the reference profile adjustment.

[0049] The device can further be configured to adjust the first reference curve and / or the second reference curve such that the first adjusted reference curve and the second adjusted reference curve transition smoothly, and in particular continuously, into one another at the transition point. This further improves the quality of the fit between the reference curves.

[0050] Alternatively or additionally, the device can be configured to adjust the first reference curve and / or the second reference curve such that the first adjusted reference curve and the second adjusted reference curve have a matching second derivative and / or a matching curvature at the (common) transition point. This further improves the accuracy of the fit between the reference curves.

[0051] According to another aspect, a device for determining lane boundaries on a roadway section is described. Based on one or more determined lane boundaries, it is possible to infer the existence of one or more lanes. A lane is typically defined by two lane boundaries. Furthermore, groups of one or more lanes (so-called lane groups) can be formed, which, for example, run parallel to each other. Additionally, the connectivity between multiple lanes can be determined. This connectivity indicates whether or not a change between lanes is possible.

[0052] The device is set up to determine a reference profile for the roadway section (e.g., using one of the methods described in this document).

[0053] The device is further configured to detect lane markings in the roadway section based on sensor data (in particular, image data). The sensor data may have been acquired by one or more sensors (e.g., one or more cameras, one or more lidar sensors, one or more radar sensors, etc.). The image data may have been acquired by at least one camera on a vehicle and / or by a camera on an aircraft. A (potential) lane marking can be identified from the image data using an image recognition algorithm. In general, a (potential) lane marking can be identified from the sensor data using a recognition algorithm. The lane marking can extend along the direction of travel of the roadway section. The lane marking can be represented by a sequence of pixels along the direction of travel of the roadway section.The reference route (with a sequence of points) and the road marking (with a sequence of pixels) can be represented in a common coordinate system and compared with each other.

[0054] The device is configured to determine arrangement information regarding the position of the road marking relative to the reference alignment. In particular, arrangement information can be determined to ascertain whether the road marking runs (substantially) parallel to the reference alignment or not. Alternatively or additionally, the distance between the road marking and the reference alignment can be determined as arrangement information. In this context, the described parallelism criterion can be used to determine whether the road marking and the reference alignment run (substantially) parallel or not.

[0055] The layout information can specify how the lane marking is positioned and / or oriented relative to the reference alignment. In particular, the layout information can specify the orientation of the lane marking relative to the reference alignment.

[0056] The device is further configured to determine, based on the arrangement information, whether the road marking constitutes a lane boundary (relevant to the reference path) or not. In particular, it can determine that the road marking is a lane boundary (relevant to the reference path) if it has been determined that the road marking runs (essentially) parallel to the reference path. Conversely, it can determine that the road marking is not a lane boundary (relevant to the reference path) if it has been determined that the road marking does not run (essentially) parallel to the reference path. Alternatively or additionally, it can be determined whether the road marking is a lane boundary (relevant to the reference path) or not based on the distance between the road marking and the reference path.

[0057] The reference profile for a road section (e.g., determined on the basis of measured driving paths) can thus be used to determine in a robust and precise manner whether a road marking (detected by sensors, especially optically) corresponds to a lane boundary or not.

[0058] The detected lane markings can be provided as map data for a digital map of the road segment. Alternatively or additionally, the detected lane markings can be used for automated longitudinal and / or lateral guidance of a vehicle (within the road segment).

[0059] As explained above, the device can be configured to divide the reference path into a sequence of support point planes at a corresponding sequence of points along the reference path. Each support point plane can be perpendicular to the reference path at its respective point. A sequence of intersection points of the road markings with the corresponding sequence of support point planes can then be determined. In particular, the intersection point with the road markings can be determined for each support point plane, resulting in a corresponding sequence of intersection points for the sequence of support point planes. Furthermore, the arrangement information can be determined with exceptional precision based on this sequence of intersection points.

[0060] In particular, arrangement information can be determined for each intersection point (e.g., the distance between the lane marking and the reference path at the respective intersection point, and / or whether the lane marking and the reference path are essentially parallel to each other at the respective intersection point). It can then be determined for each intersection point whether the lane marking at that intersection point constitutes a lane boundary (relevant to that reference path) or not. In this way, it can be sequentially checked whether a (sensor-detected) lane marking corresponds to a lane boundary or not.

[0061] The device can be configured to interpolate the lane markings detected based on the sequence of intersection points and / or, in particular, to extrapolate them beyond the lane segment. This further improves the quality of the map data obtained.

[0062] The device can be configured to determine the average distance of the intersection points of the sequence of intersection points from the corresponding points of the reference line. The distance of each intersection point to the corresponding points is determined within the respective support plane. The device can further be configured to compare this average distance with a width value representing the (typical) width of a traffic lane and, based on this comparison, determine whether the road marking constitutes a lane boundary or not. This allows for particularly robust detection of a lane boundary.

[0063] The device can be configured to determine, for each support point level of the sequence of support point levels of the reference path, an intersection point with a multitude of (sensor-detected) road markings, in order to determine a multitude of sequences of intersection points for the corresponding multitude of road markings. Thus, the intersection points of the individual support point levels with multiple (possibly parallel) road markings can be determined.

[0064] Based on the multitude of intersection sequences, it can be robustly determined whether the multitude of lane markings corresponds to a corresponding multitude of lane boundaries for a (contiguous) group of several parallel lanes. In this way, lane groups can be detected efficiently and robustly.

[0065] The device can be configured to determine a multitude of measured travel paths of one or more vehicles for a corresponding multitude of passages through the road section, and to determine a reference travel path for the road section based on the multitude of measured travel paths (e.g., based on the data associated with Fig. 3 described procedure). It can then be determined in a particularly robust manner, based on the reference route, in particular on the arrangement of the reference route relative to the road marking, whether the road marking is a lane boundary (relevant for the relevance sequence) or not.

[0066] The device can be configured to detect a first lane marking and a second lane marking in the road section based on sensor data (in particular, image data), and to determine arrangement information regarding the arrangement of the first and second lane markings relative to the reference path and / or relative to each other. It can then be determined, based on this arrangement information, whether • the first and second lane markings are lane boundaries; and • the first and second lane markings are part of a group of several parallel lanes.

[0067] This allows multiple lane markings to be analyzed in order to robustly identify groups of lanes (so-called lane groups).

[0068] According to another aspect, a device for detecting a traffic lane on a roadway section is described. The device can be specifically designed to determine the width of the traffic lane. Alternatively or additionally, the device can be designed to determine one or both lane boundaries (i.e., the course and / or position of one or both lane boundaries) of the traffic lane.

[0069] The device is designed to determine a large number of measured travel paths from one or more vehicles for a corresponding number of passages along the road section. For each passage, a measured travel path can be sent from the respective vehicle to the device and received by the device via a (wireless) communication link. For example, 5 or more, 10 or more, or 20 or more measured travel paths can be determined.

[0070] A measured vehicle path can comprise a sequence of measurement points indicating the vehicle's position, particularly the position of a specific reference point (e.g., the center of an axis), as it traverses a road segment. These measurement points can be provided with a specific spatial resolution, such as one measurement point per meter or more. Alternatively or additionally, a measured vehicle path can depict the vehicle's actual trajectory as it travels through the road segment. Thus, measured paths can be determined that show the trajectory along which individual vehicles (manually) traveled through the road segment. The measured paths can therefore describe the actual driving behavior of vehicles within that road segment.

[0071] Furthermore, the device is configured to determine a vertical dispersion metric (e.g., the standard deviation) for the dispersion of the multitude of measured driving paths in the vertical direction, perpendicular to the surface of the roadway section. The lane on the roadway section can then be determined in a particularly precise manner based on the multitude of measured driving paths and taking the vertical dispersion metric into account. The device can, in particular, be configured to determine the width of the lane, transverse to the direction of travel of the roadway section, based on the vertical dispersion metric, especially such that • the width of the lane decreases with increasing dispersion of the multitude of measured driving paths in the vertical direction; and / or • The width of the lane increases as the dispersion of the multitude of measured driving paths decreases in the vertical direction.

[0072] The determined lane (with its determined width) can be provided as map data for a digital map in relation to the road segment. Alternatively or additionally, the determined lane (with its determined width) can be used for automated longitudinal and / or lateral guidance of a vehicle.

[0073] This allows the vertical dispersion of the measured driving paths to be taken into account when determining a reference route and / or a lane and / or a lane boundary in order to increase the quality of the determined map data.

[0074] The device can be configured to determine a lateral dispersion metric (e.g., the standard deviation) for the dispersion of the multitude of measured driving paths in the lateral direction, horizontal to the surface of the roadway section and transverse to the direction of travel of the roadway section. The lane, in particular the lane width, on the roadway section can then be determined with particular precision, also taking the lateral dispersion metric into account. The device can be configured, in particular, to determine the lane width, transverse to the direction of travel of the roadway section, based on the vertical dispersion metric and on the lateral dispersion metric, in particular based on a quotient of the vertical dispersion metric and the lateral dispersion metric. The quotient can have the vertical control metric in the numerator and the lateral dispersion metric in the denominator.In this particular case, the width of the lane can be determined in such a way that... • The width of the lane decreases as the quotient increases; and / or • The width of the lane increases as the quotient decreases.

[0075] Alternatively, the quotient can have the vertical control metric in the denominator and the lateral dispersion metric in the numerator. In this case, in particular, the lane width can be determined such that... • The width of the lane increases with increasing quotient; and / or • The width of the lane decreases as the quotient decreases.

[0076] As explained above, the device can be configured to divide a reference path of the roadway section into a sequence of support point planes for a corresponding sequence of points along the reference path. The sequence of support point planes can be determined such that, along the reference path of the roadway section, directly successive support point planes each have a predefined distance from one another, for example, between 1 and 3 meters, and / or that the individual support point planes are arranged perpendicular to the reference path of the roadway section (at the respective point).

[0077] For each of the multitude of measured routes, a sequence of intersection points of the respective measured route with the corresponding sequence of support point planes can then be determined, so that for each support point plane a multitude of intersection points results for the corresponding multitude of measured routes (as, for example, in connection with the method from Fig. 3 described).

[0078] For each individual control point level, a vertical dispersion metric can be determined for the dispersion of the multitude of intersection points in the vertical direction. Furthermore, for each of the sequence of control point levels, a control point can be determined based on the multitude of intersection points with the respective control point level and taking into account the vertical dispersion metric for the respective control point level.

[0079] Based on the sequence of control points for the corresponding sequence of control point levels, a reference path for the lane on the road segment can be determined. The reference path can precisely describe the center line (in the direction of travel) of the lane.

[0080] The device can thus be configured (in particular using a method described in this document) to determine a reference path for the lane based on the multitude of measured driving paths. The reference driving path can, for example, be used as the center line of the driving lane. Furthermore, based on the multitude of measured driving paths (in particular based on the lateral dispersion metric of the multitude of measured driving paths) and taking into account the vertical dispersion metric, the width of the driving lane can be determined (with 50% of the width being located on each side of the reference driving path). In this way, a driving lane, and in particular the lane boundaries, can be determined in a particularly precise manner.

[0081] According to another aspect, a device for detecting lane boundaries on a roadway section is described. The device can be designed, in particular, to detect a lane that is bounded by two lane markings.

[0082] The device is configured to determine a reference profile for the roadway section (e.g., using one of the methods described in this document). The reference profile of the roadway section can be subdivided (as described) into a sequence of support point planes for a corresponding sequence of points along the reference profile.

[0083] The device is further configured to determine at least one measurement profile for the lane markings. This measurement profile can include at least one measured path taken by a vehicle traversing the lane segment. Alternatively or additionally, the measurement profile can include at least one lane marking detected in the lane segment based on sensor data (in particular, image data). Alternatively or additionally, the measurement profile can include at least one reference path for the lane segment (which, for example, was determined using one of the methods described in this document, in particular based on a multitude of measured paths).

[0084] The device is configured to determine a sequence of intersection points of the measurement process with the corresponding sequence of support point planes, and to determine input data (in particular a feature vector) for a machine-learned detection unit based on this sequence of intersection points. The detection unit preferably comprises at least one artificial neural network. The detection unit (in particular the at least one neural network) can be trained in advance using training data and a machine learning algorithm (e.g., a backpropagation algorithm). The training data can comprise a large number of training datasets (e.g., 1000 or more or 10000 or more training datasets), each with training input data and training output data. The training data can (statistically) describe the objective function of the trained detection unit.

[0085] The device is further configured to determine output data for the investigation unit based on the input data (transmitted to the investigation unit). The lane markings on the road section can then be determined precisely and robustly using these output data.

[0086] The determined lane markings can be provided as map data for a digital map of the road segment. Alternatively or additionally, the determined lane markings can be used for automated longitudinal and / or lateral guidance of a vehicle (within the road segment).

[0087] The device is preferably configured to sequentially determine input data for each reference plane from the sequence of support point planes. This input data includes the intersection point, and in particular at least one coordinate of the intersection point, of the measurement path with the respective reference plane. Output data can then be determined using the detection unit. This output data includes an intersection point, and in particular at least one coordinate of the intersection point, of the lane boundary with the respective reference plane. The coordinates can be efficiently specified relative to the coordinate system of the respective reference plane.

[0088] Thus, the intersection points of the lane markings with the individual support point levels can be determined step by step, level by level. These intersection points represent points of the lane markings along the reference alignment and / or along the direction of travel of the road segment. In this way, the lane markings can be determined with exceptional precision.

[0089] The device can be configured to determine input data that includes one or more intersection points of the measurement path with one or more further support point planes from a defined neighborhood of the respective reference plane. The defined neighborhood of the respective reference plane can include, • one or more control point planes that are positioned directly in front of the respective reference plane in the sequence of control point planes; and / or • one or more control point planes that are arranged directly after the respective reference plane in the sequence of control point planes.

[0090] This allows additional intersection points from the vicinity of the reference plane to be considered as additional input data. This further improves the accuracy of the determined lane markings.

[0091] The device can be configured to determine input data that includes the intersection of the lane boundary with the reference plane located (particularly directly) before the respective reference plane in the sequence of reference planes. The output data for a previous iteration (i.e., for a previous reference plane) can thus be used as input data for the current reference plane. This further improves the accuracy of the determined lane boundary.

[0092] The device can thus be configured to sequentially select a reference level from the sequence of reference levels, starting with the first reference level, in order to sequentially determine a sequence of output data for the corresponding sequence of reference levels. The lane markings on the roadway section can then be determined with particular precision based on the sequence of output data.

[0093] The device can be configured to determine a multitude of measurement profiles for one or more lane boundaries on the road section and, for each of these multiple profiles, to determine a sequence of intersection points of the measurement profile with the corresponding sequence of reference points. The input data for the detection unit can then be determined based on the multitude of intersection point sequences. By considering multiple measurement profiles, the accuracy of the one or more determined lane boundaries can be further improved.

[0094] The detection unit can be configured to accept a predefined (fixed) maximum number of intersection points as input data for each support point level. The device can also be configured to include one or more placeholder intersection points for a specific support point level in the input data if the number of intersection points with that level is less than the predefined maximum. This allows for particularly robust detection of one or more lane boundaries.

[0095] According to another aspect, a procedure for merging reference alignments for a roadway segment is described. The procedure includes determining a first and a second reference alignment for the roadway segment (e.g., each using one of the methods described in this document). Furthermore, the procedure includes identifying a first sub-segment of the roadway segment in which the first and second reference alignments satisfy a predefined parallelism criterion. The procedure also includes determining, based on the first and second reference alignments, a merged reference alignment for the first sub-segment, which is then used in place of the first and second reference alignments for the first sub-segment of the roadway segment.

[0096] According to another aspect, a procedure for adjusting reference profiles at a transition between two consecutive roadway sections is described. The procedure includes determining a first reference profile for the first roadway section and determining a second reference profile for the second roadway section that directly follows the first (e.g., using one of the methods described in this document). Furthermore, the procedure includes adjusting the first reference profile and / or the second reference profile such that the first adjusted reference profile and the second adjusted reference profile have a uniform orientation at the transition.

[0097] According to another aspect, a method for determining a lane boundary on a roadway segment is described. The method includes determining a reference alignment for the roadway segment (e.g., using one of the methods described in this document). Furthermore, the method includes detecting, based on sensor data (especially image data), a lane marking within the roadway segment. The method also includes determining arrangement information regarding the arrangement of the lane marking relative to the reference alignment and determining, based on this arrangement information, whether the lane marking constitutes a lane boundary (relevant to the relevance alignment) or not.

[0098] According to another aspect, a method for detecting a lane on a roadway segment is described. The method includes determining a multitude of measured paths from one or more vehicles for a corresponding multitude of traverses of the roadway segment, as well as determining a vertical dispersion metric for the dispersion of the multitude of measured paths in a vertical direction, perpendicular to the surface of the roadway segment. Furthermore, the method includes determining a lane on the roadway segment based on the multitude of measured paths and taking the vertical dispersion metric into account.

[0099] According to another aspect, a procedure for determining a lane boundary on a roadway segment is described. The procedure includes determining a reference alignment for the roadway segment (e.g., using one of the methods described in this document). Furthermore, the procedure includes subdividing the reference alignment of the roadway segment into a sequence of reference point planes (e.g., using one of the methods described in this document).

[0100] The procedure further includes determining a measurement profile for the lane markings. This profile can comprise a measured path taken by a vehicle traversing the lane segment and / or lane markings detected within the lane segment based on sensor data (particularly image data). Alternatively or additionally, the measurement profile can include at least one reference path for the lane segment.

[0101] Furthermore, the procedure includes determining a sequence of intersection points of the measurement path with the corresponding sequence of reference points, as well as determining, based on the sequence of intersection points, input data for a machine-learned detection unit. The procedure also includes determining output data for the detection unit based on the input data and determining the lane markings on the road section based on the output data.

[0102] Another aspect described is a software (SW) program. The SW program can be configured to run on a processor and thereby execute one or more of the procedures described in this document.

[0103] Another aspect describes a storage medium. This storage medium can include a software program configured to run on a processor and thereby execute one or more of the procedures described in this document.

[0104] It should be noted that the methods, devices, and systems described in this document can be used both alone and in combination with other methods, devices, and systems described in this document. Furthermore, any aspect of the methods, devices, and systems described in this document can be combined with one another in a variety of ways. In particular, the features of the claims can be combined with one another in a variety of ways. Features listed in parentheses are to be understood as optional features.

[0105] The invention will now be described in more detail using exemplary embodiments. Fig. 1a an exemplary system for determining reference routes for a road section; Fig. 1b an exemplary road section with several effective lanes; Fig. 2a Exemplary measured routes for the in Fig. 1b section of roadway shown; Fig. 2b Exemplary intersections and support points within a plane of a roadway section; Fig. 2c Exemplary support points for the sequence of levels of a roadway section; Fig. 2D exemplary reference routes for the in Fig. 1b section of roadway shown; Fig. 3 a flowchart of an exemplary procedure for determining a reference route for a road section; Fig. 4a Exemplary measured travel paths at a roadway checkpoint; Fig. 4b an exemplary sequence of reference routes for successive road sections; Fig. 5a a flowchart of an exemplary procedure for determining a reference route for a road section; Fig. 5b a flowchart of an exemplary procedure for determining a target trajectory for a vehicle; Fig. 6 exemplary reference points for reference routes in successive road sections; Fig. 7 an exemplary sequence of planes along a reference path of a roadway section; Fig. 8a a flowchart of an exemplary procedure for determining a reference route for a road section; Fig. 8b a flowchart of an exemplary procedure for determining a reference profile of a roadway section; Fig. 9a Exemplary reference profiles in a roadway section; Fig. 9b an exemplary fused reference curve; Fig. 9c a flowchart of an exemplary procedure for merging reference curves; Fig. 10a Exemplary reference profiles in successive road sections; Fig. 10b a flowchart of an exemplary procedure for fitting reference curves; Fig. 11a Exemplary road markings; Fig. 11b a flowchart of an exemplary procedure for detecting a lane boundary; Fig. 12 a flowchart of an exemplary procedure for lane detection; Fig. 13a an exemplary machine learning-based investigation unit; and Fig. 13b a flowchart of an exemplary procedure for determining a lane boundary.

[0106] As stated at the outset, this document deals with the efficient and reliable determination of one or more reference paths and / or reference trajectories and / or lane boundaries for a section of a road network. For example, different reference paths and / or lane boundaries can be determined for different lanes within the road section. The determined reference paths and / or lane boundaries can be provided as map data in a high-definition map of the road network. In particular, the one or more determined reference paths and / or lane boundaries can be used to determine target trajectories for the automated longitudinal and / or lateral guidance of vehicles.

[0107] Fig. Figure 1a shows an exemplary system 100 for determining at least one reference route for at least one lane section of a road network. The system 100 comprises a central unit 120, which is configured to acquire measurement data 115 from a multitude of vehicles 110 and / or for a multitude of passages through a lane section. The measurement data 115 for a passage and / or from a vehicle 110 can be received by the respective vehicle 110 via a (wireless) communication link 101. The measurement data 115 of a passage by a vehicle 110 can display a measured route of the vehicle 110 through the lane section (e.g., as a sequence of measurement points along the route traveled by the vehicle 110).

[0108] The central unit 120 (also referred to generally as the device in this document) can be configured to determine at least one reference route for the road section based on measurement data 115 from a large number of vehicles 110 and / or for a large number of passages through the road section. The reference route can describe the typical route of vehicles 110 when passing through the road section. Different reference routes can be determined for different lanes of the road section. The one or more determined reference routes for the road section can be provided by the central unit 120 as map data 125 of a digital (HD) map for the road section (e.g., sent via a communication link 101 to one or more vehicles 110).

[0109] A vehicle 110 may include a position sensor 114 configured to acquire data relating to the position of the vehicle 110 using a global navigation system (GNSS). Furthermore, the vehicle 110 may include one or more sensors 112 (such as a speed sensor, an inertial measurement unit (IMU), one or more cameras, a wheel speed sensor, a steering sensor, etc.) configured to acquire sensor data that can be used for odometry-based determination of the vehicle 110's position.

[0110] A control unit 111 of the vehicle 110 can be configured to determine the position of the vehicle 110 as it travels along a road section, based on data from the position sensor 114 and / or on sensor data from one or more sensors 112. In particular, a sequence of measurement points representing the position of the vehicle 110 can be determined along the road section, with the sequence of measurement points describing the actual path traveled by the vehicle 110. This path is referred to in this document as a "measured path". The sequence of measurement points of the measured path can be provided by the vehicle 110 as measurement data 115 (e.g., via a communication unit 113 of the vehicle 110). The sequence of measurement points can, for example, have a sampling rate or spatial resolution of 1 measurement point per meter or higher.

[0111] Fig. Figure 1b shows exemplary measured traffic paths 160 for a roadway 150 (which can be subdivided into one or more roadway sections). Each measured traffic path 160 comprises a sequence of measuring points 161. The roadway 150 may have lane markings 152 that indicate the edge of the roadway and / or a subdivision of the roadway 150 into several lanes 151, which can be detected optically and / or by sensors. The lane markings 152 may each represent an actual lane boundary.

[0112] The actual travel path 160 of a vehicle 110 can roughly correspond to the center line of a lane 151 of the roadway 150, especially if the roadway 150 has a straight course. On the other hand, substantial deviations between the center line of a lane 151 and the actual travel path 160 of a vehicle 110 can occur, for example, before a curve. Furthermore, certain roadways 150 do not have lane markings 152 for dividing them into lanes 151, such as before a toll plaza or at an intersection. In this case, no lane center lines 151 can be determined as reference travel paths based on sensor-detected lane markings 152.

[0113] As explained at the beginning, the center line of a lane 151 can be used as a target trajectory for the automated longitudinal and / or lateral guidance of a vehicle 110. As in Fig. As shown in Figure 1b, in certain road sections there may be no lane markings 152 for determining lanes 151, and therefore no target trajectory can be determined. Furthermore, in certain road sections (e.g., before or after a curve), the center line of a lane 151 may be unsuitable as a target trajectory to achieve a driving behavior typical of manually driven vehicles 110 and correspondingly comfortable during automated driving.

[0114] Fig. 2a shows the section of road from Fig. 1b determined multitude of measured routes 160. It is assumed that the in Fig. The roadway 150 shown in Figure 1b corresponds to a (single) roadway section. The central unit 120 can be configured to determine a route 270 of the roadway section based on a digital (SD) map. This route 270 can be referred to as the map route of the roadway section. The map route 270 can be used as a reference for subdividing the roadway section into a sequence of (reference) levels 271. In particular, levels 271 can be arranged perpendicular to the map route 270 at a specific distance (e.g., between 1 and 3 meters, approximately 2 meters), which are used to subdivide the measured travel paths 160 into a corresponding sequence of intersection points.

[0115] The individual levels 271 intersect the different measured driving paths 160. Thus, a measured driving path 160 can be subdivided into a corresponding sequence of intersection points by the sequence of levels 271 along the map course 270 of the road section.

[0116] Fig. Figure 2b shows the multitude of intersection points 272 for a corresponding multitude of measured driving paths 160 within an exemplary level 271 of the sequence of levels 271 for a road section. The central unit 120 can be configured to transform the multitude of intersection points 272 of a level 271 into a set of vertices 200 for a corresponding set of reference driving paths for the road section using a clustering algorithm. The DBSCAN algorithm, for example, can be used as the clustering algorithm.

[0117] The central unit 120 can be configured to determine, in a corresponding manner, a plurality of intersection points 272 for each sequence of levels 271 of the roadway section and, based on these, a set of support points 200, as exemplified in Fig. 2c is shown. The number of support points 200 in the different levels 271 can differ, at least partially. The change in the number of support points 200 in different levels 271 can be due, for example, to traffic paths merging along the roadway section or to a traffic path splitting into several traffic paths along the roadway section (as can be the case, for example, when approaching a toll plaza).

[0118] As from Fig. As can be seen in Figure 2c, based on the measured travel paths 160, a set of one or more support points 200 can be determined for each of the individual levels 271 of the sequence of levels 271. The central unit 120 can be configured to determine one or more reference travel paths 210 for the road section based on the sequence of sets of support points 200 (as exemplified in Figure 2c). Fig. (2d shown). For this purpose, along the sequence of levels 271 (in the direction of travel of the roadway section), starting from the first level 271, a support point 200 from the respective level 271 can be sequentially connected by a path segment 211 to a support point 200 from the immediately following level 271. Thus, at least one sequence of support points 20 for the corresponding sequence of levels 271 can be determined and sequentially connected with path segments 211 in order to determine a reference travel path 210. In the Fig. In the example shown in 2d, a total of 5 different reference routes 210 can be determined, whereby the reference routes 210 overlap in sub-segments of the road section.

[0119] The procedure described in this document processes discrete measurement points 161 along the measured trajectories 160 (i.e., the measured driving paths) of actual traverses of a road segment. This allows a reference driving path 210 to be determined for each lane 151, regardless of the existence of lane markings 152. This reference driving path, representing natural driver behavior, can replace the lane centerline as the target value for trajectory planning.

[0120] For each passage, there is a single estimated and / or measured sequence 160 of vehicle positions 161 (i.e., a measured route 160). A clustering algorithm can be used to assign the intersection point 272 of a measured route 160 with a plane 271 perpendicular to the road alignment 270 to the corresponding intersection points 272 of the other measured routes 160.

[0121] Individual traverses of a roadway or road segment typically differ due to natural driver behavior and a variety of environmental influences. To improve the reliability and accuracy of determining a reference driving path 210 along a lane, measures can be implemented to reduce the influence of interfering factors. In particular, when summarizing the intersection points 272 of the measured trajectories 160 with the planes 271 perpendicular to the road alignment 270 (possibly for each coordinate), an outlier-free mean value (especially a so-called trimmed mean value) can be calculated. The central unit 120 can thus be configured to identify and disregard outliers when determining the reference points 200, thereby increasing the accuracy of the determined reference points 200.

[0122] Fig. Figure 3 shows a flowchart of a (possibly computer-implemented) procedure 300 for determining a reference route 210 for a road section 410 (in a road network). A road section 410 is exemplified in Fig. 4a. The method 300 comprises determining 301 a plurality of measured travel paths 160 of one or more vehicles 110 for a corresponding plurality of crossings of the roadway section 410. The measured travel paths 160 can each indicate the trajectory that the individual vehicles 110 actually traveled when crossing the roadway section 410 in each instance.

[0123] The procedure 300 further includes the subdivision 302 of a course 270, 700 (e.g. the map course 270 or one in connection with Fig. The reference path 700 described in section 7 of the roadway section 410 is transformed into a sequence of support point planes 271. In other words, support point planes 271 (each perpendicular to the path 270, 700) can be placed along the path 270, 700 at regular intervals. The support point planes 271 can each be two-dimensional. The path 270, 700 of the roadway section 270 and / or the measured travel paths 160 can be described in three-dimensional space.

[0124] Furthermore, the procedure 300 includes determining, for each of the multitude of measured routes 160, a sequence of intersection points 272 of the respective measured route 160 with the corresponding sequence of support point planes 271. In other words, it can be determined where the individual measured routes 160 intersect the support point planes 271, and thus form intersection points 272.

[0125] The procedure 300 further comprises determining, 304, for each of the sequence of support point levels 271, a set of support points 200 based on the determined intersection points 272 with the respective support point level 271. For this purpose, one or more clusters of intersection points 272 can be determined in each individual support point level 271 using a clustering algorithm. Furthermore, one support point 200 can be determined based on each cluster (e.g., as the (possibly trimmed) mean of the intersection points 272 in the cluster).

[0126] Furthermore, the procedure 300 includes determining 305, based on the sequence of sets of control points 200 for the corresponding sequence of control point levels 271, at least one reference travel path 210 for the roadway section 410. For this purpose, one or more contiguous sequences of control points 200 along the sequence of control point levels 271 can be determined and each connected to the others via path segments 211. The individual contiguous sequences of control points 200 then each form a reference travel path 210.

[0127] When a roadway 150 branches off (e.g., at an exit or intersection) into two different connecting roadways 150 with different orientations, the measured travel paths 160 of vehicles 110 within a roadway segment can differ, depending on whether the vehicle 110 enters the first connecting roadway 150 or the second connecting roadway 150 after the roadway segment. For example, at a turning, the measured travel paths 160 of vehicles 110 traveling straight ahead can (statistically) differ from the measured travel paths 160 of vehicles 110 turning. This is an example of a turning situation 400 in Fig. 4a is shown. In particular, it shows Fig. 4a a (core) carriageway section 410 followed (in straight-ahead driving) by a subsequent first (surrounding) carriageway section 410 and followed (in right-turning driving) by a subsequent second (surrounding) carriageway section 410.

[0128] Out of Fig. 4a It is particularly evident that the measured travel paths 160 in the (core) carriageway section 410 differ before the turn, depending on whether a vehicle 110 continues straight ahead or turns after this (core) carriageway section 410. The measured travel paths 160 in a (core) carriageway section 410 can therefore depend on in which one or more subsequent (surrounding) carriageway sections 410 the respective measured travel path 160 runs.

[0129] Similarly, the measured travel paths 160 for a (core) carriageway section 410 can depend on one or more preceding (surrounding) carriageway sections 410 through which the respective measured travel path 160 has passed.

[0130] The central unit 120 can be configured to determine several different reference routes 210 for different preceding and / or subsequent (surrounding) road sections 410 for a (core) road section 410. In particular, for a given core road section 410 (e.g., based on data from an (SD) card), all possible section sequences of directly consecutive road sections 410 can be determined that have a certain number N of surrounding road sections 410 before the given core road section 410 and / or a certain number M of surrounding road sections 410 after the given core road section 410. N and / or M can each be 1 or more, or 2 or more. For each possible sequence of roadway sections 410 (each comprising the specified core roadway section 410), a reference route 210 can then be determined in the specified core roadway section 410.If different possible section sequences of road sections 410 exist for the specific core roadway section 410 Q, then Q different reference driving paths 210 can be determined (e.g. Q greater than 1, or Q greater than 2).

[0131] For this purpose, the central unit 120 can be configured to assign the multitude of available measured travel paths 160 for the specific core roadway section 410 to one of the Q sequences of roadway sections 410. In the Fig. In the example shown in 4a, the measured driving paths 160 in straight-ahead direction would thus be assigned to a first section sequence of road sections 410 and the measured driving paths 160 in right-turn direction to a second section sequence of road sections 410.

[0132] Thus, for each segment sequence of roadway sections 410, a subset of measured travel paths 160 can be determined. Based on the respective subset of measured travel paths 160, a reference travel path 210 can then be determined (using the procedure 300 described in this document). The Q reference travel paths 210 can then be provided as map data 125 for the specific roadway section 410. During the operation of an (automated) vehicle 110, the segment sequence of roadway sections 410 of the vehicle 110's route can then be determined. For traversing the specific core roadway section 410, the corresponding reference travel path 210 can then be selected (and used to determine the target trajectory of the vehicle 110 for traversing the core roadway section 410).

[0133] For the linking of reference routes 210 of successive roadway sections 410, it may be advantageous (as exemplified in Fig. (as shown in Figure 4b), a distance 420 may be provided between the endpoint of the reference path 210 of a carriageway section 410 and the endpoint 412 of the carriageway section 410. Similarly, a distance 420 may also be provided between the beginning 411 of the carriageway section 410 and the beginning of the reference path 210 of the carriageway section 410. The distance 420 may, for example, correspond to half the distance between two successive levels 271. The beginning point of a reference path 210 may correspond to the support point 200 of the first support point level 271 and / or the endpoint of a reference path 210 may correspond to the support point 200 of the last support point level 271 (in each case with respect to the direction of travel through the carriageway section 410).

[0134] As in Fig. As shown in Figure 4b, the last support point 200 of the reference path 210 of a roadway section 410 has a specific distance 420 from the end 412 of the roadway section 410. Furthermore, the first support point 200 of the reference path 210 of the subsequent roadway section 410 has a specific distance 420 from the beginning 411 of the subsequent roadway section 410. A (possibly linear) connecting segment 421 can then be inserted between these two support points 200 of the reference paths 210 to determine a linked reference path for the concatenation of the roadway sections 410.

[0135] Fig. Figure 5a shows a flowchart of a (possibly computer-implemented) procedure 500 for determining a reference route 210 for a core roadway section 410 (i.e., for a roadway section 410, which is referred to as the core roadway section for the unambiguous description of the procedure 500). The procedure 500 comprises determining 501 a plurality of measured routes 160 of one or more vehicles 110 for a corresponding plurality of crossings of the core roadway section 410.

[0136] Furthermore, the method 500 comprises assigning 502 each of the plurality of measured travel paths 160 to a section sequence from a set of different section sequences in order to determine, for each section sequence, a subset of measured travel paths 160 from the set of section sequences. The different section sequences can each include the core roadway section 410 and at least one surrounding roadway section 410 located directly in front of and / or directly behind the core roadway section 410 in the direction of travel. The section sequences can thus each include one or more roadway sections 410.

[0137] Furthermore, procedure 500 includes determining, for each section sequence from the set of section sequences, a reference driving path 210 for the core roadway section 410 based on the respective subset of measured driving paths 160. For this purpose, procedure 300 can be used, for example.

[0138] Fig. Figure 5b shows a flowchart of a (possibly computer-implemented) procedure 510 for the at least partially automated longitudinal and / or lateral guidance of a vehicle 110 along a route through a road network. The procedure 510 can be executed by a control unit 111 of the vehicle 110.

[0139] Method 510 comprises determining 511, in particular based on map data 125 of a digital map of the road network (such as an HD map), for a future traverse of a core roadway segment 410 of the road network along the route of a set of different reference routes 210 for a corresponding set of different segment sequences. Method 510 also comprises identifying 512, based on the route, a segment sequence from the set of different segment sequences. In particular, it can determine the segment sequence that corresponds to the route, i.e., along which the route passes.

[0140] Furthermore, the procedure 510 includes determining 513 a target trajectory of the vehicle 110 for at least partially automated longitudinal and / or lateral guidance of the vehicle 110 when passing through the core roadway section 410 based on the reference route 210 for the identified section sequence.

[0141] Determining reference routes 210 for a sequence of road segments 410 that exhibit a relatively large curvature (e.g., due to a turning situation) can lead to significant discontinuities at the transitions between the reference routes 210 for directly consecutive road segments 410. This is exemplified in Fig. Figure 6 shows in particular Fig. 6. The support points 200 of the reference path 210 for a first carriageway section 410 (each shown as white circles) and the support points 200 of the reference path 210 for a subsequent second carriageway section 410 (each shown as hatched circles) are shown. The alignment 270 of the second carriageway section 410 is essentially perpendicular to the alignment 270 of the first carriageway section 410. This results in the support point planes 271 for the second carriageway section 410 being arranged essentially perpendicular to the support point planes 271 for the first carriageway section 410.

[0142] As from Fig. As can be seen in Figure 6, due to the relatively strong divergence between the orientation of the support point planes 271 and the course of the measured travel paths 160, overlaps and / or discontinuities can occur between one or more support points 200 at the end 412 of the first carriageway section 410 and one or more support points 200 at the beginning 411 of the subsequent second carriageway section 410. This, in turn, can lead to discontinuities at the transition between the reference travel paths 210 of the two carriageway sections 410.

[0143] As exemplified in Fig. As shown in Figure 7, the central unit 120 can be configured to determine a reference route 700 for a road section 410 based on the measured travel paths 160. For example, the measured travel paths 160 can be averaged (possibly without considering one or more outliers) to determine the reference route 700. The reference route 700 can then be used instead of the map route 270 of the road section 410 to determine the sequence of control point levels 200 for the road section 410. The individual control point levels 200 can each be arranged perpendicular to the reference route 700.

[0144] This in connection with Fig. 3 and the Fig. The procedure 300 described in 2a to 2d can then be used to determine the support points 200 for a reference route 210. This increases the accuracy of the determined reference route 210, particularly with regard to the transition to the reference route 210 for the subsequent road section 410.

[0145] As from Fig. As can be seen in Figure 7, the beginning 411 (i.e., the starting point or position of the starting point) of the map route 270 of the road section 410, for which a driving path 210 is to be determined, may differ from the beginning 701 (i.e., from the starting point or position of the starting point) of the corresponding reference route 700. Alternatively or additionally, the end 412 (i.e., the endpoint or position of the endpoint) of the map route 270 of the road section 410 may differ from the end 702 (i.e., from the endpoint or position of the endpoint) of the corresponding reference route 700. The central unit 120 may be configured to determine the beginning 701 and / or the end 702 of the reference route 700 in such a way that a certain distance criterion or distance measure is reduced, in particular minimized.The distance criterion can depend on the distance of the respective point 701, 702 of the reference route 700 from the corresponding point 411, 412 of the map route 270. Alternatively or additionally, the distance criterion can depend on the deviation of the orientation 720 of the beginning 701 or the end 702 of the reference route 700 from the orientation 720 of the beginning 411 or the end 412 of the map route 270. The beginning 701 and / or the end 702 of the reference route 700 can thus be determined in such a way that the distance to the beginning 411 and / or the end 412 of the map route 270 is as small as possible and / or that the orientation 720 of the beginning 701 and / or the end 702 of the reference route 700 is as similar as possible to the orientation 720 of the beginning 411 and / or the end 412 of the map route 270.

[0146] Fig. Figure 8a shows a flowchart of a (possibly computer-implemented) procedure 800 for determining a reference route 210 for a roadway section 410. The procedure 800 comprises determining 801 a plurality of measured routes 160 of one or more vehicles 110 for a corresponding plurality of crossings of the roadway section 410. Furthermore, the procedure 800 comprises determining 802, based on the plurality of measured routes 160, a reference route 700 of the roadway section 410. This reference route 700 can then be used as route 270, 700 in the procedure 300 to determine at least one reference route 210 for the roadway section 410.

[0147] The procedure 800 can therefore, in particular, include arranging 803 a sequence of support point planes 271 along the reference route 700, and determining 804, for each of the plurality of measured driving paths 160, a sequence of intersection points 272 of the respective measured driving path 160 with the corresponding sequence of support point planes 271. Furthermore, the procedure 800 can include determining 805, based on the plurality of determined sequences of intersection points 272 for the corresponding plurality of measured driving paths 160, at least one reference driving path 210 for the road section 410.

[0148] Fig. Figure 8b shows a flowchart of a (possibly computer-implemented) procedure 810 for determining a reference path 700 of a core roadway section 410 (i.e., a roadway section referred to as the core roadway section). The procedure 810 comprises determining 811 a plurality of measured travel paths 160 of one or more vehicles 110 for a corresponding plurality of crossings of the core roadway section 410.

[0149] Furthermore, the method 810 comprises assigning 812 a first subset of measured travel paths 160 from the multitude of measured travel paths 160 to a first section sequence from a set of different section sequences. The different section sequences can each include the core roadway section 410 and at least one surrounding roadway section 410 located directly in front of and / or directly behind the core roadway section 410 in the direction of travel.

[0150] Procedure 810 further comprises determining 813, (solely) on the basis of the first subset of measured travel paths 160 for the first section sequence, a first reference profile 700 of the core roadway section 410 for the first section sequence. The first reference profile 700 can be determined, as in this document, (solely) on the basis of the first subset of measured travel paths 160.

[0151] Thus, when determining the reference route 700 for a core lane section 700, it is already possible to consider from which surrounding lane section 700 a vehicle 100 enters the core lane section 700 and / or into which surrounding lane section 700 a vehicle exits the core lane section 700. A direction-dependent reference route 700 can therefore be determined. In this way, the quality of the determined reference routes 210 can be further improved.

[0152] In particular, the different reference routes 210 for the different section sequences can each be determined using the specific reference profile 700 for the respective section sequence. This significantly improves the quality of the determined reference routes 210.

[0153] The reference profile 700 of the core lane section 410 for a specific section sequence can be provided as map data for a digital map. The reference profile 700 can then be displayed, for example, in a head-up display of a vehicle 110 when driving through the core lane section 410 (e.g., as an augmented reality display).

[0154] It can happen that for a (core) carriageway section 410, a large number of different reference profiles 700 are determined, e.g., different section sequences for Q. This is exemplified in Fig. 9a is shown. In particular, it shows Fig. 9a a first reference path 901 (e.g. for driving straight ahead) and a second reference path 902 (e.g. for turning maneuver) for the (core) carriageway section 410 (each with a beginning 701 and an end 702).

[0155] The use of a large number of different reference profiles 901, 902 can involve relatively high resource expenditure (e.g., for storage and / or computing resources) during further data processing. The device 120 can be configured as shown by way of example in Fig. Figure 9b shows how to identify at least one subsection 904 of the carriageway section 410 in which the reference profiles 901 and 902 are essentially oriented in the same way (and consequently provide essentially identical support point planes 271). In the Fig. In the example shown in 9b, this is the case for the subsection 904 between the beginning 701 and an intermediate point 903.

[0156] A merged reference trajectory 911 can be determined for the identified subsection 904, for example, by averaging the reference trajectories 901 and 902 within the identified subsection 904. The originally determined reference trajectories 901 and 902 can then be connected at the end 903 of subsection 904. A spatial offset between the merged reference trajectory 911 and the subsequent reference trajectories 901 and 902 is typically negligible (for determining the support point levels 271). On the other hand, when determining the merged reference trajectory 911, it can be taken into account and / or ensured that the merged reference trajectory 911 and the subsequent reference trajectories 901 and 902 each have the same orientation at the transition point 903.

[0157] The separate treatment of individual driving and turning directions during the generation of the road geometry (i.e., the reference trajectories 700) can thus lead to situations where road trajectories of different driving directions run almost parallel before junctions and after mergers being stored separately in the digital map 125. The multiple representation of the nearly identical road geometry (i.e., the nearly identical reference trajectory 700) increases the data volume required for storing and transmitting the map data 125. Furthermore, the localization of a vehicle 110 on a digital map of road trajectories or associated road geometry 700 can be made more difficult due to the relatively high number of possible assignments. Path planning based on the map (i.e.,Determining a target trajectory for a vehicle (110) typically requires evaluating a relatively large number of possible paths when determining the optimal path, thereby increasing the required amount of computing resources.

[0158] When using the road alignment geometry 700 to generate cross-sectional planes 271 for determining localization-relevant landmarks, the multiple representation of the same road alignment geometry 700 can lead to landmarks being determined multiple times and subsequently having to be sorted out.

[0159] After generating the road alignment geometries (i.e., the reference alignments) 901, 902, the road alignment geometries 901, 902 can be divided into groups of potentially fusionable road alignment geometries 901, 902. Criteria for fusion include, in particular, belonging to the same map tile (an SD card), belonging to one of its neighbors, spatial proximity in general, and / or belonging to the same SD road segment 410 or the same sequence of road segments.

[0160] Within a group, it can be iteratively checked for each pair of road alignment geometries 901, 902 whether they run approximately parallel to each other in one or more subsections 904 and can therefore be merged. If this is the case, the original road alignment geometries 901, 902 are replaced by the merged result 911. The (iterative) procedure can then be continued with the merged reference alignment 911 and the remaining road alignment geometries. The merged result 911 may include the combined, merged representation of the road alignment geometry of the subsection 904 where the two original road alignment geometries 901, 902 run parallel to each other, and the (now duplicate-free) subsections before and / or after it.

[0161] The procedure can be continued until no more pairs of fusionable road alignment geometries 901, 902 can be found. The testing and fusion of a first and second road alignment geometry 901, 902 is carried out, for example, by first determining the distances along the first road alignment geometry 901 from and to which the second road alignment geometry 902 runs approximately parallel to the first road alignment geometry 901, i.e., while adhering to a parallelism criterion. If this is the case for a non-empty interval 904, two or more cross-sectional planes 271 (in particular at regular intervals) can be constructed within this interval 904 orthogonally to the first road alignment geometry 901.For each cross-sectional plane 271, a support point 200 of the merged road alignment 911 can be determined from the intersection points 272 of the first and second road alignment geometries 901, 902 with the respective cross-sectional plane 271.

[0162] The determination of the support point 200 can advantageously be carried out by a weighted averaging of the intersection points 272 with the respective cross-sectional plane 271. The number of road alignment geometries 901, 902 that have already been averaged to generate the first and second road alignment geometries 901, 902, respectively, can advantageously be used as the weighting factor.

[0163] Fig. Figure 9c shows a flowchart of a (possibly computer-implemented) procedure 930 for merging reference profiles 901, 902 for a roadway section 410. The procedure 930 comprises determining 931 a first and a second reference profile 901, 902 for the roadway section 410. The two reference profiles 901, 902 may, for example, have been determined for different section sequences (e.g., using procedure 810).

[0164] The procedure 930 further includes the identification 932 of a first subsection 904 of the roadway section 410 in which the first and the second reference alignment 901, 902 meet a predefined parallelism criterion.

[0165] Furthermore, procedure 930 includes determining 933 for the first subsection 904, based on the first and second reference profiles 901, 902, a merged reference profile 911, which is used instead of the first and second reference profiles 901, 902 for the first subsection 904 of the carriageway section 410. The merged reference profile 911 can be used, for example, to determine a lane boundary 152 and / or to determine a reference driving path 210 (e.g., within the framework of procedure 300).

[0166] Determining different reference profiles 700 for different, consecutive roadway sections 410 can lead to a discontinuity at the transition between the reference profiles 700 for two directly consecutive roadway sections 410, in particular a discontinuity with respect to the orientation of the reference profiles 700. This is exemplified in Fig. 10a is shown. In particular, it shows Fig. 10a a first reference alignment 1001 for a first carriageway section 410 and a second reference alignment 1011 for a second carriageway section 410 directly following the first carriageway section. At the transition at the end 702 of the first reference alignment 1001 and the beginning 701 of the second reference alignment 1011, both reference alignments 1001, 1011 have different orientations 1002, 1012.

[0167] The device 120 can be configured to adjust the reference profiles 1001, 1011 such that the two reference profiles 1001, 1011 have the same orientation 1002, 1012 at the transition.

[0168] As explained above, for determining one or more reference routes 210 and / or localization-relevant landmarks, it is typically not the road alignment geometry 1001, 1011 itself that is relevant, but rather the position and orientation of the cross-sectional planes 271 constructed by it. Therefore, all road alignment geometries 1001, 1011 entering and / or exiting a transition point can be adjusted so that they have the same orientation 1002, 1012 at the transition point itself, without necessarily merging into one another. With a continuous representation of the road alignment geometries 1001, 1011, i.e., a spline, the continuity of the first derivative can thus be ensured, but not necessarily the zeroth derivative.

[0169] The adjustment of the road alignment geometries 1001, 1011 can be achieved by determining a common adjustment plane, the point of reference of which advantageously corresponds to the centroid of the starting or ending points 701, 702 of the road alignment geometries involved 1001, 1011, and whose normal vector is advantageously determined as the mean of the associated orientations, which may be mirrored if the direction is opposite.

[0170] Subsequently, the road alignment geometries involved, 1001 and 1011, can be adjusted so that their starting and ending points, 701 and 702 respectively, lie in the determined plane and their tangent vector at this point is parallel to the normal vector of the determined plane. The adjustment of the geometries can be achieved by modifying their support points, 200, for example, using spline interpolation.

[0171] Optionally, an average curvature can be determined that all road alignment geometries 1001, 1011 should exhibit on the determined plane. Optionally, it can be ensured that all road alignment geometries 1001, 1011 pass through the point of reference of the determined plane, thus also achieving continuity in the zeroth derivative, even if this results in the direction of travel of the determined road alignment geometries 1001, 1011 differing more significantly locally from the actual routes 160.

[0172] Fig. Figure 10b shows a flowchart of a (possibly computer-implemented) procedure 1020 for fitting reference profiles 1001, 1011 at a transition between two consecutive carriageway sections 410. The procedure 1020 comprises determining 1021 a first reference profile 1001 for a first carriageway section 410, and determining 1022 a second reference profile 1011 for a second carriageway section 410 directly following the first carriageway section 410. The reference profiles 1001, 1002 can each be determined using procedure 810.

[0173] Procedure 1020 further comprises adapting 1023 the first reference profile 1001 and / or the second reference profile 1011 such that the first adapted reference profile and the second adapted reference profile have a uniform orientation at the transition. The first adapted reference profile and / or the second adapted reference profile can be used, for example, to determine a lane boundary 152 and / or to determine a reference driving path 210 (e.g., within the framework of procedure 300).

[0174] Using camera data from one or more vehicle cameras 110 and / or based on satellite data, i.e., generally based on image data, possible lane markings can be detected that, for example, separate different lanes 151 of a roadway 150 from one another. The reference profiles 700 for individual roadway sections 410 described in this document can be used to reliably determine whether a possible lane marking (detected based on sensor data, especially image data) is also an actual lane boundary 152.

[0175] For this purpose, as exemplified in Fig. As shown in Figure 11a, the reference profile 700 for a roadway section 410 can be subdivided into a sequence of control point levels 271 (for a corresponding sequence of points along the reference profile 700). Directly consecutive control point levels 271 can have a constant distance (e.g., 1 to 3 meters) between them.

[0176] The intersection points 272 of the sequence of support point planes 271 with one or more possible road markings 1101, 1102 can then be determined. For each possible road marking 1101, 1102, this results in a corresponding sequence of intersection points 272.

[0177] The sequence of intersection points 272 for a possible lane marking 1101, 1102 can be analyzed to determine whether the possible lane marking 1101, 1102 is an actual lane boundary 152 or not. In particular, it can be analyzed how the individual intersection points 272 are arranged relative to the reference line 700. For example, the (perpendicular) distance 1103 of the individual intersection points 272 to the reference line 700 can be considered. If this distance 1103 is substantially constant for the sequence of intersection points 272 and / or has a certain value (which is typical for the width of a lane 151), then it can be reliably concluded that the possible lane marking 1101 is an actual lane boundary 152.

[0178] On the other hand, if the distance 1103 of the intersection points 272 changes and / or the distance 1103 has a value that is unsuitable for a lane 151, it can be concluded that the possible road marking 1102 is not an actual lane boundary 152.

[0179] Lanes 151, which can be used to travel locally from one point in the road network to another, e.g., all left-turn lanes 151 or all straight-ahead lanes 151 at an intersection, generally run parallel to each other. Since vehicles 110 primarily move along lanes 151, the road geometry determined from the measurement data 115 (i.e., the reference alignment 700) also runs parallel to all lanes 151 of a contiguous group of lanes 151. In the coordinate system of a given road geometry 700, the corresponding lanes 151 therefore run parallel to the coordinate axis in the direction of travel, provided that the configuration of the lanes 151, e.g., the width of the lanes 151, does not change.

[0180] This can be used to determine, from one or more possible lane markings 1101, 1102 (e.g., detected on the basis of sensor data, especially image data), those that are relevant for the currently considered lane section 410 (e.g., for a specific turning direction at an intersection), in order to determine a common lane group 152 for these actual lane boundaries 152, e.g., in the sense of the NDS (Navigation Data Standard) data format.

[0181] The NDS data format groups adjacent lanes 151 on a roadway section 410 as a lane group. In combination with the type of lane marking 152 (e.g., solid line or dashed line), this results in a compact representation of the lateral connectivity of the lanes 151, i.e., the information on which roadway sections 410 it is possible to change from one lane 151 to another.

[0182] With the method described in this document, a subsequent determination of the lateral connectivity of adjacent lanes 151 is not necessary, since the lateral connectivity is already given by the sorting of the coordinates of the intersection points 272 of the road markings 1101, 1102 in the cross-sectional plane 271 along the coordinate axis running perpendicular to the direction of travel.

[0183] This also means that the intersection points 272 of the road markings 1101 in the coordinate system of the cross-sectional planes 271 perpendicular to the road geometry 700 typically change only minimally between the planes 271, thus enabling the tracing of individual markings 1101 from one plane 271 to a successor plane 271. This allows gaps in the available observations of the road markings 1101 (e.g., in the interior of the intersection) to be closed. This enables a complete determination of the lanes 151, even in the interior of the intersection.

[0184] Because only road markings 1101 that run parallel to the corresponding road geometry 700 are used for the individual roadway sections 410, the described method is robust against incorrectly detected road markings 1102 that run transversely or obliquely to the direction of travel. Furthermore, considering several parallel road markings 1101 together allows for additional plausibility checks, for example, regarding the assumption that the lane width is the same for all lanes 151. This can further increase the robustness against incorrect road markings 1102.

[0185] To further reduce the risk of errors in the determined road model, observations of actual traffic patterns 160 in a lane 151, for example in the form of one or more previously determined reference traffic paths 210, can be used for plausibility checks. If necessary, only those lanes 151 for which such a reference traffic path 210 or such an observation 160 exists, or for which certain other criteria apply, can be included in the road model. Such criteria can relate, for example, to whether an adjacent lane 151 has been confirmed by a corresponding observation and what type of lane marking 152 the current lane 151 has.

[0186] Fig. Figure 11b shows a flowchart of a (possibly computer-implemented) procedure 1110 for determining a lane boundary 152 on a roadway section 410. The procedure 1110 includes determining 1111 a reference profile 700 for the roadway section 410 (e.g., using procedure 810). Furthermore, the procedure 1110 includes detecting 1112 a road marking 1101, 1102 in the roadway section 410, based on sensor data, in particular image data.

[0187] Procedure 1110 further includes determining 1113 arrangement information regarding the arrangement of the road markings 1101, 1102 relative to the reference course 700. Furthermore, procedure 1110 includes determining 1114, based on the arrangement information, whether the road markings 1101, 1102 constitute a lane boundary 152 or not.

[0188] As in connection with Fig. As explained in section 1b, measurement data 115 relating to the travel path 160 of each passage can be recorded for a large number of traverses of a roadway section 410. In other words, a large number of measured travel paths 160 can be recorded for the roadway section 410. Based on the measured travel paths 160, one or more reference travel paths 210 for the roadway section 410 can be determined. Alternatively or additionally, at least one reference route 700 of the roadway section 410 can be determined.

[0189] The accuracy of the individual measured driving paths 160 can vary. In particular, the positional accuracy of the individual measurement points 161 of measured driving paths 160 can vary. The positional accuracy of the position sensor 114 of a vehicle 110 is typically direction-independent. Specifically, the position sensor 114 typically exhibits the same positional accuracy in the horizontal direction (i.e., parallel to the surface of the roadway 150) as in the vertical direction (i.e., perpendicular to the surface of the roadway 150). The control of the measurement points 161 of a measured driving path 160 in the vertical direction (i.e., the vertical dispersion) can thus be used as an indicator of the positional accuracy in the horizontal direction.

[0190] The vertical dispersion of the measurement points 161 of a measured driving path 160 can be used, in particular, to determine how strongly (e.g., with what weighting) the respective measured driving path 160 is taken into account when determining a reference driving path 210 and / or when determining a lane boundary 152. Typically, the weighting increases with decreasing vertical dispersion. Conversely, the weighting typically decreases with increasing vertical dispersion.

[0191] A reference route 210 determined from measurement data 115 can contain information regarding the variance of the individual traverses used to generate the reference route 210, in particular the scatter of the trajectories 160 perpendicular to the direction of travel. This scatter can advantageously be used to determine the width of a (possibly virtual) lane 151 and thus to determine the position of the determined lane boundaries 152.

[0192] Since the localization accuracy of the determined trajectories 160 can vary due to the availability and / or uniqueness of localization-relevant landmarks, it is not always possible to determine with certainty whether a high dispersion of the trajectories 160 results from reduced localization accuracy or from the actual driving behavior during the individual passes. By utilizing the observation that reduced localization accuracy affects all spatial directions, or that a correlation exists between vertical and lateral localization accuracy, the vertical dispersion of the trajectories 160 can be used to determine whether a relatively high lateral dispersion is due to reduced localization accuracy or to the actual driving behavior during the individual passes.

[0193] For example, the ratio of the lateral and vertical dispersion metrics of the measured driving paths 160 belonging to a reference driving path 210 can be used to determine the width of the (possibly virtual) driving lane 151. A driving lane 151 is only widened if a relatively low vertical dispersion indicates that the actual driving behavior varies, and not if a relatively low localization accuracy results in inaccurate trajectories 160.

[0194] In addition to the variance of the measured driving paths 160 during the individual passages, further information can be used to determine the position of the lane boundaries 152, such as the position and variance of the measured driving paths 160 for one or more adjacent reference driving paths 210 and / or the position of lane boundaries, which were detected, for example, on the basis of sensor data, in particular image data, before or after the considered roadway section 410 along the reference driving path 210.

[0195] Alternatively or additionally, other metrics of lateral dispersion can be used, such as the minimum and maximum and / or a predetermined quantile of lateral dispersion.

[0196] Fig. Figure 12 shows a flowchart of a (possibly computer-implemented) procedure 1200 for detecting and / or determining a lane 151 on a roadway section 410. In particular, a reference driving path 210 and / or a lane boundary 152 for the lane 151 can be detected and / or determined.

[0197] Method 1200 comprises determining 1201 a plurality of measured paths 160 of one or more vehicles 110 for a corresponding plurality of passages over the roadway section 410. Furthermore, method 1200 comprises determining 1202 a vertical dispersion metric for the dispersion of the plurality of measured paths 160 in the vertical direction, perpendicular to the surface of the roadway section 410. Method 1200 also comprises determining 1203 a lane 151 (in particular the width of the lane 151) on the roadway section 410 based on the plurality of measured paths 160 and taking into account the vertical dispersion metric.

[0198] As explained above, the reference profile 700 for a roadway section 410 can be used to determine lane boundaries 152 for one or more lanes 151 of the roadway section 410. In particular, the reference profile 700 can be used to divide the roadway section 410 into a sequence of support point levels 271, with each support point level 271 being arranged perpendicular to the reference profile 700. The individual support point levels 271 can then be used to determine the intersection points 272 of the respective support point level 271 with (image-recorded) road markings 1101, 1102 and / or measured traffic paths 160. These intersection points 272 (per support point level 271) can be analyzed to determine a reference driving path 210 and / or a lane boundary 152 for a lane 151 of the roadway section 410.

[0199] The determination of a reference driving path 210 and / or a lane boundary 152 can be achieved in a particularly efficient and precise manner using a previously trained neural network. Fig. Figure 13a shows an exemplary (artificial intelligence and / or machine learning-based) investigation unit 1300, which is set up to determine output data 1302 (which describe the reference route 210 and / or the lane boundary 152) on the basis of input data 1301 (which is based on the intersections 272 with one or more support point levels 271).

[0200] Investigation unit 1300 may have been pre-trained using training data (e.g., using a learning algorithm). Investigation unit 1300 may comprise one or more pre-trained neural networks.

[0201] The sequence of vertex levels 271 can, for example, have K consecutive vertex levels 271 (e.g., K ≥ 10 or K ≥ 20). A window can be viewed at a time containing a subset of vertex levels 271 with a specific number of directly consecutive vertex levels 271 (e.g., one or more, or two or more, or three or more vertex levels 271). This window can be moved successively from the beginning 701 of the reference path 700 to the end 702 of the reference path 700 in order to gradually determine the reference route 210 and / or the lane boundary 152.

[0202] The input data 1301 can be determined based on the intersection points 272 for a window of one or more control point levels 271. Based on this, output data 1302 for this window of one or more control point levels 271 (or for a reference level 271 from the window) can then be determined using the determination unit 1300. The window can then be repeatedly shifted to the end 702 of the reference path 700 with a specific step size, and input data 1301 and output data 1302 can be determined for the shifted window. The step size can, for example, correspond to one control point level 271 each time.

[0203] This describes a method for determining the road model of a roadway section 410 using machine learning methods, in particular deep neural networks. For this purpose, the available information can be transformed such that the trained machine learning model (i.e., the determination unit 1300), represented by the function f, can map a set X of possible feature vectors of fixed length (i.e., possible input data 1301) to a set Y of output vectors of fixed length (i.e., output data 1302), from which the road model to be determined (in particular, the course of a lane boundary 152) can be ascertained.

[0204] Based on the road alignment geometry 700, the information relevant for the road model can be represented in the cross-sectional planes 271 perpendicular to the road alignment geometry 700. To define the road model, features can be represented as attributes of the individual cross-sectional planes 271 and / or associated coordinates can be transformed into the coordinate systems of the individual cross-sectional planes 271.

[0205] The input data 1301 (in particular the input vector) can comprise intersection points 272 of one or more road markings 1101, 1102, which were detected, for example, on the basis of sensor data, in particular on the basis of image data, with a reference cross-sectional plane 271 (as well as with one or more directly adjacent cross-sectional planes 271). A fixed set and / or number of intersection points 272 can be used per plane 271.

[0206] The individual intersection points 272 can each be represented by their coordinates. Advantageously, the coordinates of the intersection points 272 contained in the input data 1301 are transformed into the coordinate system of the respective plane 271 and / or normalized to a predefined range of values ​​(e.g., minus 10 to plus 10 meters).

[0207] Typically, the actual number of intersection points 272 is not fixed, but variable. To nevertheless transform this into a feature vector of fixed length, the intersection points 272 can be grouped according to their associated cross-sectional plane 271, ordered per plane 271 according to their respective coordinate perpendicular to the direction of travel, or according to their respective coordinate in the coordinate system of the respective plane 271, and then supplemented by a number of placeholder intersection points, where the number of placeholder intersection points corresponds to the difference between the number of actually existing intersection points 272 and the defined constant maximum number of intersection points 272 per plane 271.

[0208] The machine learning model, i.e., the detection unit 1300, may have been pre-trained such that the output data 1302, in particular the output vector, contains a set of coordinates for each cross-sectional plane 271, representing the one or more intersection points 272 of the one or more lane boundaries 152 of a consistent road model with the reference cross-sectional plane 271. In this process, incorrectly detected lane markings 1102 may have been eliminated by the detection unit 1300.

[0209] The output data 1302, in particular the output vector, may (analogous to the input data 1302) contain placeholder intersection points if the number of intersection points actually specified by the consistent road model 272 does not correspond to the specified maximum number of intersection points 272.

[0210] For training the machine learning model, i.e., the detection unit 1300, a variety of training datasets can be used, with each training dataset containing training input data and corresponding (desired) training output data. The training output data can be obtained through actual measurements of lane markings 152 (where the measurements were carried out, for example, with a measuring vehicle). Thus, supervised training of the detection unit 130 can have been performed beforehand.

[0211] The trained machine learning model, i.e., the detection unit 1300, can be applied to feature vectors (i.e., input data 1301) for a sequence of successive reference cross-sectional planes 271. In this way, the intersection points 272 of the lane boundaries 152 of a consistent road model with the individual reference cross-sectional planes 271 can be determined. Any errors and / or gaps in the road markings 1101, 1102 used for the input data 1301 can be corrected by the machine learning model, so that the individual intersection or support points 272, 200 can be precisely assembled into a coherent road model.

[0212] Alternatively or additionally to the use of placeholder intersection points, in the event that the defined maximum number of intersection points is not exhausted, a grid with a fixed number of cells can be placed over the respective level 271, whereby each cell represents either no or a maximum of one intersection point 272.

[0213] Alternatively or additionally, the type of lane markings 152 can be taken into account in the input data 1301 and / or in the output data 1302, e.g. in the form of probabilities for the different possible types of lane markings 152.

[0214] Alternatively or additionally, the intersection points 272 of measured routes 160 with the respective levels 271 can be taken into account in the input data 1301.

[0215] Fig.Figure 13b shows a flowchart of a (possibly computer-implemented) procedure 1310 for determining a lane boundary 152 on a roadway section 410. The procedure 1310 includes determining 1311 a reference profile 700 for the roadway section 410 (e.g., using procedure 810). Furthermore, the procedure 1310 includes subdividing 1312 the reference profile 700 of the roadway section 410 into a sequence of support point planes 271 for a corresponding sequence of points along the reference profile 700.

[0216] Furthermore, procedure 1310 includes determining 1313 at least one measurement path 160, 1101, 1102 for the lane boundary 152. The measurement path 160, 1101, 1102 can, for example, be a measured travel path 160 of a vehicle 110 for traversing the roadway section 410. Alternatively or additionally, the measurement path 160, 1101, 1102 can be a road marking 1101, 1102 detected in the roadway section 410 based on sensor data, in particular image data. Alternatively or additionally, the measurement path 160, 1101, 1102 can be a reference travel path 210 for the roadway section 410. The procedure 1310 further includes determining 1314 a sequence of intersection points 272 of the measurement sequence 160, 1101, 1102 with the corresponding sequence of support point planes 271.

[0217] Furthermore, the procedure 1310 includes the determination 1315, based on the sequence of intersection points 272, of input data 1301 for a machine-learned determination unit 1300, the determination 1316 of output data 1302 of the determination unit 1300 for the input data 1301, and the determination 1317 of the lane boundary 152 on the road section 410 based on the output data 1302.

[0218] The measures described in this document enable the efficient and precise determination of reference trajectories 700 and / or lanes 151 and / or lane boundaries 152 for road sections 410 of a road network, which can be provided as map data 125 for an HD map. The determined reference trajectories 700 and / or lanes 151 and / or lane boundaries 152 can be used by a partially autonomous vehicle 110 to determine target trajectories for the vehicle 110. This can improve the performance of autonomous vehicles 110.

[0219] The present invention is not limited to the embodiments shown. In particular, it should be noted that the description and the figures are intended only to illustrate the principle of the proposed methods, devices, and systems by way of example.

Claims

[1] Device (120) for merging reference profiles (901, 902) for a road section (410); wherein the device (120) is configured - to determine a first and a second reference profile (901, 902) for the road section (410); - to identify a first subsection (904) of the roadway section (410) in which the first and second reference profiles (901, 902) satisfy a predefined parallelism criterion; and - to determine a fused reference profile (911) for the first sub-section (904) on the basis of the first and second reference profiles (901, 902), and to use it instead of the first and second reference profiles (901, 902) for the first sub-section (904) of the roadway section (410). [2] Device (120) according to claim 1, wherein the device (120) is arranged - to identify a second subsection (905) of the roadway section (410) in which the first and second reference profiles (901, 902) do not meet the predefined parallelism criterion; and - to maintain the first and second reference profiles (901, 902) for the second subsection (905). [3] Device (120) according to one of the preceding claims, wherein the device (120) is configured to determine a first and a second composite reference profile for the road section (410), which - for the first subsection (904) have the common merged reference profile (911); and - for one or more other sub-areas (905) of the roadway section (410) each have the first or the second reference profile (901, 902). [4] Device (120) according to one of the preceding claims, wherein the device (120) is arranged - to divide the first and second reference profiles (901, 902) of the roadway section (410) into a sequence of support point planes (271); wherein the individual support point planes (271) are each arranged perpendicular to the respective reference profile (901, 902); wherein the sequence of support point planes (271) is arranged at a corresponding sequence of points along the first and second reference profiles (901, 902); wherein successive points follow one another, in particular at a uniform distance, approximately between 1 and 3 meters; and - the parallelism criterion depends on a relative orientation of corresponding support point planes (271) of the first reference profile (901) and the second reference profile (902) to one another. [5] Device (120) according to claim 4, wherein - the device (120) is configured to identify, as a first subsection (904), a subsequence of points from the sequence of points for which the orientations of the corresponding support point planes (271) of the first and second reference profiles (901, 902) deviate from each other by at most a predefined angle threshold value; and - the angle threshold is in particular 10° or less. [6] Device (120) according to one of the preceding claims, wherein the device (120) is arranged - to divide the first reference profile (901) of the roadway section (410) into a sequence of support point planes (271); wherein the individual support point planes (271) are each arranged perpendicular to the first reference profile (901); wherein the sequence of support point planes (271) is arranged at a corresponding sequence of points along the first reference profile (901); - for each support point plane (271) from the sequence of support point planes (271), to determine a respective intersection point (272) of the second reference profile (920) with the respective support point plane (271); and - the parallelism criterion depends on an orientation of the individual support point planes (271) relative to an orientation of the second reference profile (902) at the corresponding intersection points (272). [7] Device (120) according to claim 6, wherein - the device (120) is configured to identify, as a first subsection (904), a subsequence of points from the sequence of points for which the orientations of the corresponding support point planes (271) and the orientations of the second reference profile (902) at the corresponding intersection points (272) deviate from each other by at most a predefined angle threshold value; and - the angle threshold is in particular 10° or less. [8] Device (120) according to one of the preceding claims, wherein the device (120) is arranged - to subdivide the first reference profile (901) in the first subsection (904) by a sequence of support point planes (271); - to determine for the first reference profile (901) and the second reference profile (902) a sequence of intersection points (272) of the respective reference profile (901, 902) with the corresponding sequence of support point planes (271); - for each support point plane (271) from the sequence of support point planes (271) to determine a respective support point (200) on the basis of the determined intersection points (272) with the respective support point plane (271); and - to determine the fused reference profile (911) on the basis of the sequence of support points (200) for the corresponding sequence of support point levels (271), in particular by sequentially connecting the support points (200) by path segments (211). [9] Device (120) according to claim 8, wherein the device (120) is configured to determine the support point (200) for a support point plane (271) as a mean value, in particular as a weighted mean value, of the determined intersection points (272) with the respective support point plane (271). [10] Device (120) according to claim 9, wherein the device (120) is arranged - to determine a first number of measured travel paths (160) which were used to determine the first reference profile (901); - to determine a second number of measured travel paths (160) which was used to determine the second reference profile (902); and - To determine weights for the calculation of the weighted mean based on the first number and based on the second number. [11] Device (120) according to one of the preceding claims, wherein the device (120) is arranged - until a termination criterion is reached, iteratively merging at least one subsection of two reference profiles (901, 902) from a set of reference profiles (901, 902) for the roadway section (420) in order to determine a fused reference profile in each case; and - to include the fused reference curve determined during an iteration in the set of reference curves (901, 902) for the subsequent iteration. [12] Device (120) according to claim 11, wherein the device (120) is arranged - to determine the number of reference histories from which the reference histories to be merged in an iteration were determined; and - to use the determined numbers as weights when merging the reference histories to be merged. [13] Device (120) according to one of the preceding claims, wherein the device (120) is arranged - to provide the merged reference course (911) as map data (125) relating to the first subsection (904) of the roadway section (410) for a digital map; and / or - to use the merged reference profile (911) for automated longitudinal and / or lateral guidance of a vehicle (110); and / or - to detect one or more lane markings (152) in the first subsection (904) of the lane section (401) on the basis of the fused reference profile (911); and / or - to detect one or more lane boundaries (152) and / or one or more lanes (151) in the first subsection (904) of the roadway section (401) on the basis of the fused reference profile (911). [14] Device (120) according to one of the preceding claims, wherein - the first and second reference courses (901, 902) are associated with different section sequences; - the different section sequences each comprise the roadway section (410) and at least one surrounding roadway section (410) arranged in the direction of travel directly in front of and / or directly behind the roadway section (410); and - the different section sequences differ from one another in at least one surrounding roadway section (410). [15] Method (930) for merging reference profiles (901, 902) for a roadway section (410); the method (930) comprising - determining (931) a first and a second reference profile (901, 902) for the road section (410); - identifying (932) a first subsection (904) of the roadway section (410) in which the first and second reference profiles (901, 902) satisfy a predefined parallelism criterion; and - determining (933) for the first subsection (904), on the basis of the first and the second reference profile (901, 902), a fused reference profile (911) which is used instead of the first and the second reference profile (901, 902) for the first subsection (904) of the roadway section (410).

Images