Method for determining and providing lane course

Abstract

The invention relates to a method for determining and providing a lane course of a road by means of a central computer unit coupled to fleet vehicle data. According to the invention, the method comprises the following steps:-providing a geometric map (K) with geometric lane boundaries (S) and drawing a grid (G) of grid cells (G1 to G49) of predetermined dimensions on the map (K); collecting fleet data from the fleet vehicles, wherein the fleet data comprises a sequence of positions traveled by the fleet vehicles; -determining a vehicle orientation of the vehicle at the positions of the grid cells (G1 to G49) from the collected sequence of positions; discretizing the determined orientation of the vehicle; generating a histogram (H) for each grid cell position for the discretized vehicle orientation determined at each grid cell (G1 to G49) position; -selecting a map portion having a predetermined number of grid cells (G1 to G49); and-determining a lane course on the map section by means of a learning-based method from a histogram (H) created for the grid cells (G1 to G49) of the map section and from geometric lane boundaries (S) on the map section.

Description

Technical Field

[0001] The invention relates to a method for determining and providing the course of lanes of a road by means of a central computer unit coupled to vehicle data of a fleet of vehicles. Background Art

[0002] DE 10 2013 211 696 A1 discloses a method for completing and / or updating a digital road map of a geographical area, a device for a motor vehicle, and a motor vehicle. The method comprises the following steps: a) capturing digital images of a road section by means of a camera of a motor vehicle located on the road section; b) determining the current position of the motor vehicle in the width direction of the road section and / or at least one property of the road section by means of a computing device of the motor vehicle based on these images; c) transmitting a data record to a server device, the data record containing information about the geographical location of the capture of the image and information about at least one determined property of the road section and / or information about the position in the width direction. Steps a) to c) are performed by a plurality of motor vehicles located in the area, and the digital road map is completed and / or updated by the server device based on the received data records.

[0003] DE 10 2013 018 315 A1 discloses a method for providing an environment model for a vehicle, wherein it is proposed to use a grid map to represent the environment model and to adapt the grid map to the lane course, for example if the lane is curved, the cells of the grid map are also arranged in a curved manner.

[0004] A method for creating a road model is known from DE 10 2013 208 521 A1, in which it is proposed to determine trajectory data from fleet data and to create a road model from the trajectory data by means of statistical methods.

[0005] A method for determining lanes is known from DE 10 2019 119 002 A1, in which it is proposed to scan lane markings, determine the scanning direction and cluster the scans taking into account the direction.

[0006] EP 2 888 604 B2 discloses a method for determining a lane course for a vehicle, wherein it is provided to determine the position of the vehicle in a grid map and to determine the direction of the vehicle and the distance to the grid cells having the road edge lines. Summary of the invention

[0007] The object of the present invention is to provide a new method for determining the course of a lane of a road.

[0008] According to the invention, this object is achieved by a method having the features of claim 1 .

[0009] Advantageous embodiments of the invention are the subject matter of the dependent claims.

[0010] According to the present invention, a method for determining and providing the lane orientation of a road by means of a central computer unit coupled to fleet vehicle data is proposed, providing a geometric map with geometric lane boundaries, and plotting a grid of grid cells of a predetermined size on the map. Fleet data is collected from the fleet vehicles, wherein the fleet data includes a sequence of positions in which the fleet vehicles travel. In addition, a vehicle orientation of the vehicle at the grid cell position is determined based on the collected position sequence, and the determined vehicle orientation is discretized. Then, for the discretized vehicle orientation determined at each grid cell position, a histogram is generated for each grid cell position, a map portion with a predetermined number of grid cells is selected, and the lane orientation on the map portion is determined based on the histogram created for the grid cells of the map portion and the geometric lane boundaries on the map portion by a learning-based method.

[0011] By applying this method, geometric infrastructure and convoy driving behavior can be processed comprehensively. This allows for fully automated lane derivation, whereas derivation based purely on geometric infrastructure always requires manual rework, and lane derivation based on convoy driving behavior cannot achieve full coverage.

[0012] By using geometric information, it can be ensured to a large extent that the modeled lane alignment is based on the given infrastructure.

[0013] Since the present method proposes the use of platoon driving behavior, human driving style is taken into account, i.e. manual driving of platoon vehicles. This supports lane modeling in unstructured areas, e.g. where lane markings and / or bus lanes are missing.

[0014] Learning-based methods essentially automatically take special cases into account if they are reflected in the data set, ie the fleet data. The method can be extended with fleet data, so that the determination of the lane course improves continuously with more training examples.

[0015] Furthermore, learning-based methods can directly generate the geometry of the lane alignment. In addition, older lane alignments can optionally be taken into account in learning-based methods, so that only the parameters of the lane segment related to the anchor point, orientation, curvature and / or lane width need to be predicted. This simplifies the training process and can largely ensure that geometrically valid lane segments are always generated.

[0016] By abstracting the lane orientations that can be determined based on the vehicle convoy data into a grid, the driving behavior of vehicles belonging to a convoy can be processed consistently even when the number of trips is different.

[0017] Multidimensional grids enable modeling of multiple directions of travel, such as on two-way roads or in intersection areas.

[0018] The grids created at different points in time make it possible to detect changes in the convoy’s driving behavior and thus draw conclusions about changes in the road infrastructure. This also allows changes to be identified that would not be detectable based on the signals recorded by the vehicle’s sensor systems, such as traffic signs prohibiting turns.

[0019] Since the processing is done at the grid cell level, it can be broken down into sub-problems related to determining the lane orientation, allowing for parallel processing.

[0020] In one embodiment, the determined lane orientation is made available to the convoy vehicles for retrieval. Based on the provision of the determined lane orientation, the automated driving operation and autonomous driving operation of the respective vehicle can be optimized. The respective lane orientation is known, so that in the automated driving operation or autonomous driving operation, it is not necessary for the vehicle driver to have to take over the driving task and move the vehicle under manual driving operation, for example due to a lack of lane markings.

[0021] In another specific embodiment, the determined lane course is provided to the vehicle as data of a digital map and stored in the vehicle, whereby automated or autonomous driving of the vehicle can be significantly improved without driver intervention.

[0022] One possible embodiment of the method provides that the determined lane course is fed to a trajectory planning module of an autonomous vehicle in the convoy. Based on the determined lane course, a trajectory is then planned for the respective autonomous vehicle, so that an optimized autonomous driving operation of the vehicle can be substantially achieved without interruption due to the transfer of the driving task to the vehicle driver.

[0023] In a refinement of the method, the learning-based method is trained using the created histogram and a ground truth map created for the specified area. Thus, the learning-based method is trained on real data and optimized to determine the lane orientation based on the transmitted fleet data. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Embodiments of the present invention are explained in more detail below with reference to the accompanying drawings.

[0025] in:

[0026] Figure 1 schematically shows a portion of a geometric map with geometric lane boundaries and a drawn grid with grid cells,

[0027] Figure 2 Schematically shows the frequency distribution histograms generated for the grid cell positions, and

[0028] Figure 3 A plurality of tubular channels are schematically shown.

[0029] In all the figures, parts which correspond to one another are provided with the same reference numerals. DETAILED DESCRIPTION

[0030] Figure 1 A section of a geometric map K with geometric lane boundaries S and a drawn grid G ​​with grid cells G1 to G49 are shown.

[0031] Figure 2 shows a grid G ​​and grid cells G1 to G49 populated by a data structure, and Figure 3 A plurality of tubular channels F are shown as being identified.

[0032] In general, it is known that automated and autonomous driving of vehicles requires high-precision digital maps, i.e. road maps, wherein the content of the digital maps must be correct and constantly updated. Such digital maps with lane-accurate topological modeling are absolutely necessary for applications of level 2+ or higher automated driving in accordance with the SAE J3016 standard. Mapping is performed perspectively based on signals acquired by vehicle sensor systems of vehicles in a fleet (especially of vehicle manufacturers). With the fleet vehicles, relatively large coverage areas are achieved, and the vehicles send the fleet data as information related to the static infrastructure (e.g. recognized traffic signs and lane markings) to a central computer unit, which is coupled to the fleet vehicle data.

[0033] In addition to the identification information related to the static infrastructure as environmental information, the attitude, in particular the orientation, based on the satellite navigation positioning of the respective vehicle is also transmitted to the central computer unit in order to transform the detection results with respect to the vehicle coordinate system into the global coordinate system. In addition, the satellite navigation also provides information about the driving behavior of the respective vehicle, in particular the driving behavior on the traversable corridor F.

[0034] The following describes a method for determining and providing the course of lanes of a road by means of a central computer unit coupled to vehicle data of a fleet of vehicles. The method is designed to have multiple stages.

[0035] First, a geometric map K with lane boundaries S is provided, and a grid G ​​with grid cells G1 to G49 is drawn. The grid cells G1 to G49 have a predetermined size with respect to the map K. In particular, the geometric map K is provided based on observations of the convoy vehicles using methods known in the prior art. The specific steps include so-called single track pose optimization, multi-track alignment, and geometric aggregation.

[0036] Fleet data of the convoy vehicles are collected in a central computer unit. The fleet data include position sequences of the convoy vehicles. The orientation of each vehicle is recorded as fleet data, wherein the position sequence has a specified duration or length. A relatively short duration or length is selected in order to anonymize the collected fleet data and to make it difficult to draw conclusions about the movement profile of the individual vehicles. In this case, all orientations are located and recorded with respect to the geometric map K in order to ensure lane-accurate vehicle positions. In particular, the orientation of the vehicle at the position of the grid cells G1 to G49 is determined based on the collected position sequence.

[0037] The determined vehicle orientation is discretized, i.e., the determined vehicle orientation is classified into a predetermined grid (e.g., a 60° grid). In other words: the data structure of the grid G ​​is generated according to the corresponding geographical location. The relationship of the grid G ​​to the geographical location, i.e., to the global coordinate system with x, y, z coordinates, can be discretized into 1m x 1m squares via a specified grid of the geographical map K.

[0038] A histogram H is then generated for each grid cell position for the discretized vehicle orientation determined at the position of each grid cell G1 to G49. In this case, the frequency distribution is multidimensionally modeled with respect to the driving direction with the aid of the grid cells G1 to G49 so that geographical points with multiple driving directions (e.g., in the area of ​​intersections) can be mapped. The frequency distribution can be discretized, for example, by 10° levels.

[0039] Learning-based methods (e.g., convolutional neural networks, Transformer networks, graph neural networks) use the geometric lane boundaries S (especially lane markings, curbs, etc.) in a previously provided geometric map K and a grid G ​​with a frequency distribution derived from fleet data as input information, and provide a lane model as output information.

[0040] Learning-based methods are trained using a ground truth map that depicts the actual lane orientation in a particular area. Learning-based methods learn from examples how to generate lane models based on geometric lane boundaries S (lane markings, curbs, etc.) and histograms H.

[0041] Figure 1 By way of example, a section of a grid G ​​is shown with a lane boundary S superimposed thereon. The arrows P indicate the distribution pattern with respect to the direction of travel.

[0042] It is relatively easy to see that in the grid cells G1 to G49 where vehicles travel in both directions, two arrows P pointing in opposite directions are shown.

[0043] The grid cells G1 to G49 of the first row Z1 are numbered, wherein a first channel F shown with hatching and a second channel F shown with different hatching are lane courses generated using a learning-based method.

[0044] Figure 2 The corresponding histogram H of the grid cells G1 to G7 of the grid G ​​is shown by way of example. The vehicle direction is discretized in 60° steps. In particular, Figure 2 The corresponding histogram H is shown, where Figure 1 The individual grid cells G1 to G7 shown quantitatively define discrete vehicle orientations.

[0045] according to Figure 1 As shown in the grid cell G5, it can be seen that the vehicle is traveling in the orientation shown by the downward arrow P. Therefore, the mode (maximum value of the distribution) in the histogram H is around 200°.

[0046] The vehicle travels through grid cells G6 and G7 in opposite directions, where the mode in histogram H is around 20°.

[0047] The corresponding histogram H based on the grid cells G1 to G7 of the grid G ​​relatively effectively represents any number of trips on the grid cells G1 to G49 of the grid G ​​drawn on the geographic map K, and reflects the convoy driving behavior of the vehicles in a unified form.

[0048] Based on such an abstraction process, the histogram H is used as a consistent input to a downstream learning-based method, which derives the lane orientation by adding geometric context.

[0049] This unified representation of variable trip counts can also be used to detect changes in fleet behavior at a relatively early stage in map development. This allows changes in fleet behavior that would not be detectable based on signals collected by the vehicle sensor systems, such as special traffic signs prohibiting turns.

[0050] From a geometric point of view, the lane orientation to be generated using the learning-based method is a tubular channel F. Therefore, the learning-based method can optionally take into account existing old lane orientations, so that the learning-based method only needs to predict the parameters of the lane segments, such as anchor points, orientation, curvature, lane width. This simplifies the training process and ensures that geometrically valid lane orientations are always generated. An example of such a tubular channel F is Figure 3 As shown, it is produced by selecting appropriate input parameters for the learning-based method.

[0051] Furthermore, the method provides that the determined lane course in the form of the tubular corridor F is made available to the vehicles of the motorcade as data in a digital map.

[0052] In particular, the determined lane course is fed to a trajectory planning module of the automated or autonomous vehicle, so that the trajectory of the vehicle can be planned based on the existing lane course.

Claims

1. A method for determining and providing a lane course of a road by means of a central computer unit coupled to vehicle data of a fleet of vehicles, It is characterized by the following steps: - providing a geometric map (K) having geometric lane boundaries (S) and drawing a grid (G) having grid cells (G1 to G49) of predetermined sizes on said map (K); - collecting fleet data from fleet vehicles, wherein the fleet data comprises a sequence of locations travelled by the fleet vehicles; - determining a vehicle orientation of the vehicle at the position of the grid cell (G1 to G49) based on the collected position sequence; - discretizing the determined vehicle orientation; - for the discretized vehicle orientation determined at each of the grid cell positions (G1 to G49), generating a histogram (H) for each of the grid cell positions; - selecting a map portion having a predetermined number of grid cells (G1 to G49); as well as - determining the lane orientation on the map portion by a learning-based method based on the histogram (H) created for the grid cells (G1 to G49) of the map portion and the geometric lane boundaries (S) on the map portion.

2. The method according to claim 1, It is characterized in that The determined lane direction is provided to the convoy vehicles for retrieval.

3. The method according to claim 1 or 2, It is characterized in that The determined lane course is made available to the vehicle as data in a digital map.

4. The method according to any one of the preceding claims, It is characterized in that The determined lane courses are fed to trajectory planning modules of the autonomous vehicles in the convoy.

5. The method according to any one of the preceding claims, It is characterized in that The learning based method is trained using the created histogram (H) and the ground truth map created for the specified area.

Images