Method for calculating and providing a lane route

Abstract

The present invention relates to a method for calculating and providing a lane path of a road using a central computing unit that is connected to vehicles in a vehicle group in a data-technical manner. According to the present invention, the method comprises the following steps: providing a geometric map (K) including geometric lane boundaries (S), applying a grid network (G) having grid cells (G1 to G49) of a predetermined size to the map (K), collecting fleet data of the vehicles in the vehicle group, including a position sequence in which the vehicles in the vehicle group travel, calculating the orientation of the vehicles at the positions of the grid cells (G1 to G49) from the collected position sequence, discretizing the calculated vehicle orientation, generating a histogram (H) for each of the discretized vehicle orientations calculated at the positions of the respective grid cells (G1 to G49) for each individual grid cell position, selecting a map section having a predetermined number of grid cells (G1 to G49), and calculating a lane path in the map section using a learning-based method from the histogram (H) created for the grid cells (G1 to G49) of the map section and the geometric lane boundary (S) in the map section.

Description

Technical Field

[0001] The present invention relates to a method for calculating and providing a lane path of a road using a central computing unit that is technically connected to vehicles in a vehicle group.

Background Art

[0002] From German Patent Application Publication No. 102013211696, a method for completing and / or updating a digital road map of a geographical area, an apparatus for an automobile, and an automobile are known. This method includes steps of: a) detecting a digital image of a road section using a camera of an automobile existing on the road section; b) calculating, using an arithmetic unit of the automobile, a current position of the automobile in the width direction of the road section and / or at least one characteristic of the road section based on the image; c) transmitting a data set including information on the geographical position where the image is detected, and information on at least one calculated characteristic of the road section and / or the position in the width direction to a server device. Steps a) to c) are executed by a plurality of automobiles existing in the area, and the digital road map is completed and / or updated using the server device based on the received data set.

[0003] From German Patent Application Publication No. 102013018315, a method for providing an environmental model for a vehicle is known. In this method, a grid map is used to display the environmental model, and this grid map is adapted to the lane path. For example, when the lane is curved, the cells of the grid map are also arranged in a curved manner.

[0004] From German Patent Application Publication No. 102013208521, a method for creating a road model is known. In this method, trajectory data is obtained from fleet data (vehicle group data), and a road model is created from these trajectory data using a statistical method.

[0005] From German Patent Application Publication No. 102019119002, a method for determining lanes is known. In this method, lane markings are scanned, the scanning direction is determined, and the scan is clustered considering that direction.

[0006] From European Patent No. 2888604, a method for determining the lane path of a vehicle is known. In this method, in a grid map, the position of the vehicle is determined, and the direction and distance of the vehicle with respect to grid cells including lane boundaries are determined.

Summary of the Invention

Problems to be Solved by the Invention

[0007] An object of the present invention is to provide a novel method for calculating the lane path of a road.

Means for Solving the Problems

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

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

[0010] According to the present invention, in a method for calculating and providing a lane path (lane course, lane passage) of a road using a central computing unit that is connected to vehicles in a vehicle group (fleet) technically (from the perspective of data technology), a geometric map including geometric lane boundaries is provided, and a grid network having grid cells of a predetermined size is applied to the geometric map. Fleet data of vehicles in the vehicle group is collected, and the fleet data includes a position sequence in which the vehicles in the vehicle group travel. Further, the orientation of the vehicle at the position of the grid cell is calculated from the collected position sequence, and the calculated orientation of the vehicle is discretized (individualized). Subsequently, for each individual grid cell position, a histogram for the orientation of the vehicle calculated and discretized at the position of each grid cell is generated respectively, a map section having a predetermined quantity (predetermined number) of grid cells is selected, and a lane path in the map section is calculated (determined) using a learning-based method from the histogram created for the grid cells of the map section and the geometric lane boundaries in the map section.

[0011] By applying this method, it becomes possible to perform a process that combines a geometric infrastructure and fleet driving behavior. As a result, it becomes possible to completely automatically derive the lane path. In contrast, in the derivation of the lane path based purely on a geometric infrastructure, manual post-processing (reprocessing, redoing) is always required, and in the derivation of the lane path based on fleet driving behavior, complete coverage cannot be achieved.

[0012] By using geometric information, it is generally possible to ensure that the modeled lane path is directed (conforms) to a predetermined infrastructure.

[0013] In this method, since fleet driving behavior is provided, the driving style of a human, that is, the manual driving operation of the vehicles in the vehicle group is considered. As a result, for example, lane modeling in an unstructured area, for example, a location where there are no lane markings and / or bus lanes is supported.

[0014] Learning-based methods substantially automatically take into account special cases if they are represented in the dataset, i.e., in the fleet data. The method is scalable (extensible) using the fleet data, whereby the calculation (determination) of the lane path is continuously improved as the number of training examples increases.

[0015] Furthermore, learning-based methods can directly generate the geometric characteristics (geometry) of the lane path. Furthermore, old lane paths can be selectively considered in learning-based methods, and thus only the parameters of the lane segment regarding the anchor point, alignment, curvature, and / or lane width need to be predicted. This simplifies the learning process and ensures that geometrically valid lane segments are generated as continuously as possible.

[0016] By abstracting the lane path calculated based on the fleet data of vehicles into a grid network, it becomes possible to consistently process the driving behaviors of vehicles belonging to different groups of vehicles with different driving frequencies.

[0017] The multi-dimensional grid network enables the modeling of multiple driving directions, for example, in the area of two-way roads or intersections.

[0018] Grid networks created depending on (in response to) different points in time enable the detection of changes in fleet driving behavior and thus lead to conclusions about changes in the road infrastructure. Therefore, changes that are not detected by signals from the vehicle sensor system (vehicle sensor system), such as traffic signs prohibiting a direction change, can also be calculated.

[0019] Since the processing is performed at the level of grid cells, the problem of calculating the lane path can be divided (broken down) into partial problems (sub-problems, lower-level problems), thereby enabling parallelization.

[0020] In one embodiment, the calculated lane route is provided so that the vehicles in the vehicle group can acquire it. Based on these provided calculated lane routes, optimized automatic driving operations and autonomous driving operations of the corresponding vehicles become possible. Since each lane route is known, in the automatic driving operation or the autonomous driving operation, for example, due to the lack of lane markings, it is not necessary for the driver of the vehicle to take over the driving task and move the vehicle by manual driving operation.

[0021] In another embodiment, the calculated lane route is provided to the vehicle as data in a digital map and stored in the vehicle, so that the automatic driving operation or the autonomous driving operation of the vehicle can be significantly improved without the intervention of the driver.

[0022] In one possible embodiment of the method, the calculated lane route is supplied to the trajectory planning module of the autonomous driving vehicles in the vehicle group. Subsequently, based on the calculated lane route, a trajectory for each autonomous driving vehicle is planned, so that an optimized autonomous driving operation of the vehicle can be substantially realized without intervention by passing the driving task to the driver of the vehicle.

[0023] In one development form of the method, the training of the learning-based method is performed using the created histogram and the ground truth map created for a predetermined area. Therefore, the learning-based method is trained based on real data and optimized for calculating the lane route based on the transmitted fleet data.

[0024] Hereinafter, embodiments of the present invention will be described in more detail based on the drawings.

Brief Description of the Drawings

[0025]

Figure 1

Figure 2

Figure 3

DETAILED DESCRIPTION OF THE INVENTION

[0026] In all figures, corresponding parts are denoted by the same reference numerals.

[0027] FIG. 1 shows a part of a geometric map K including geometric lane boundaries S to which a grid network G having grid cells G1 to G49 is applied (plotted).

[0028] FIG. 2 shows a grid network G having grid cells G1 to G49 filled (entered) by a data structure, and FIG. 3 shows a plurality of tubular corridors (passages, roads).

[0029] Generally, it is known that an automatic driving operation or an autonomous driving operation of a vehicle requires a high-precision digital map, i.e., a road map. The content of the digital map must be accurate and also needs to be updated. In the application cases of automatic driving operations at level 2+ or higher of the standard SAE J3016, a digital map accurately modeled in terms of lanes by topology is essential. Mapping is respectively performed based on signals detected by vehicle sensors of a fleet of vehicles, particularly those of a vehicle manufacturer. By using the vehicles of the fleet, a relatively wide range can be covered, and the vehicles transmit fleet data (fleet vehicle data) as information regarding static infrastructure such as identified traffic signs and lane markings to a central computing unit that is connected to the vehicles of the fleet in a data-technical manner (from the perspective of data technology).

[0030] In addition to their identified information regarding static infrastructure as peripheral information, the attitude measured by the satellite navigation of each vehicle, particularly the orientation, is also transmitted to the central computing unit in order to convert the detection regarding the vehicle coordinate system into the global coordinate system. Furthermore, the satellite navigation provides information regarding the driving behavior of individual vehicles, particularly information regarding drivable corridor F.

[0031] Hereinafter, a method for calculating and providing a lane route of a road will be described using a central computing unit that is technically connected to the vehicles of a vehicle group. Here, the method is composed of multiple steps.

[0032] First, a geometric map K including lane boundaries S is provided, and a grid network G having grid cells G1 to G49 is applied. Here, the grid cells G1 to G49 have a predetermined size with respect to the map K. In particular, the geometric map K is provided from observations of the vehicles of the vehicle group using a method known from the prior art. The individual steps regarding this are so-called single-trace pose optimization, multi-trace alignment, and geometric aggregation.

[0033] The fleet data of the vehicles of the vehicle group are collected by the central computing unit. Here, the fleet data include the position sequence in which the vehicles in the vehicle group drive. The respective orientation of the vehicle is detected as fleet data, and the position sequence has a predetermined duration (period) or length. In order to anonymize the collected fleet data and make it difficult to draw conclusions regarding the movement profile of each vehicle, the duration or length is selected to be relatively short. Therefore, all orientations with respect to the geometric map K are detected while being localized (positioned) for the exact position of the vehicle in the lane unit. In particular, the orientation of the vehicle at the positions of the grid cells G1 to G49 is calculated (determined) from the collected position sequence.

[0034] The calculated vehicle orientation is discretized (individualized). That is, the calculated vehicle orientation is classified into a predetermined pattern (grid, classification frame), for example, a 60° pattern. In other words, depending on each geographical location, a data structure for the grid network G is generated. The dependence of the grid network G on geographical locations, that is, the dependence on the global coordinate system having coordinates x, y, z, can be discretized (individualized) into 1m×1m squares using a predetermined pattern of the geographical map K.

[0035] Subsequently, for each individual grid cell position, a histogram H of the discretized vehicle orientation calculated at the positions of each grid cell G1 to G49 is generated. Here, for example, in order to be able to map geometric points having a plurality of driving directions in the intersection area, a frequency distribution depending on the driving direction is modeled in multiple dimensions using the grid cells G1 to G49. The frequency distribution can be discretized, for example, using grading in 10° classes.

[0036] Learning-based methods, such as Convolutional Neural Network, Transformer Network, Graph Neural Network, supply a lane model as output information using, as input information, geometric lane boundaries S, particularly lane markings, curbs, etc. from a previously provided geometric map K, and also using a grid network G having a frequency distribution from fleet data.

[0037] The learning-based method is trained using a Ground-Truth map that maps the actual lane route for a specific area. The learning-based method learns, based on examples, how to generate a lane model from the geometric lane boundaries S (lane markings, curbs, etc.) and the histogram H.

[0038] FIG. 1 exemplarily shows a part of a grid network G overlaid with lane boundaries S. The mode of distribution regarding (over) the driving direction is indicated by an arrow P.

[0039] It can be relatively easily seen that in grid cells G1 to G49 through which vehicles pass in both driving directions, two arrows P pointing in opposite directions are represented.

[0040] The grid cells G1 to G49 in the first row Z1 are numbered, and the first corridor F represented by hatching and the second corridor F represented by another hatching are lane routes generated using a learning-based method.

[0041] FIG. 2 exemplarily shows respective histograms H for grid cells G1 to G7 of the grid network G. Discretization (individualization) of the vehicle direction in 60° classes is selected. In particular, FIG. 2 shows respective histograms H in which the number of discretized vehicle orientations is calculated for each of the grid cells G1 to G7 shown in FIG. 1.

[0042] According to grid cell G5 shown in FIG. 1, it can be seen that a vehicle has traveled within this cell in the direction represented by the downward arrow P. Therefore, the mode (maximum value of the distribution) of the histogram H is approximately 200°.

[0043] In grid cells G6 and G7, vehicles travel in the directions of vehicles in the opposite driving direction, so the mode of the histogram H is approximately 20°.

[0044] Respective histograms H based on grid cells G1 to G7 of the grid network G relatively efficiently represent any number of travels in grid cells G1 to G49 of the grid network G applied on a geographical map K, and reproduce (reflect in a unified form) the free-fleet driving behavior (driving characteristics of a vehicle group) of vehicles in a uniform manner.

[0045] Based on such an abstraction process, the histogram H is used as a certain (consistent, coherent) input for a downstream learning-based method that derives a lane path including the geometric surrounding environment.

[0046] Furthermore, by utilizing such uniform representations for different numbers of trips, changes in fleet driving behavior can be detected at a relatively early stage of the mapping process. That is, it is possible to detect changes in fleet driving behavior that are not detected by signals from the vehicle sensor system, such as a special traffic sign that prohibits a direction change.

[0047] The lane path generated using the learning-based method is, from a geometric perspective, a tubular corridor F. For this reason, optionally, an existing old lane path can be considered using the learning-based method, so this learning-based method only needs to predict the parameters of the lane segment such as, for example, the anchor point, orientation, curvature, and width of the lane. This can simplify the training process and ensure that a geometrically valid lane path is continuously generated. An example of such a tubular corridor F resulting from appropriately selecting the input parameters for the learning-based method is shown in Figure 3.

[0048] Furthermore, in this method, the lane path calculated in the shape of the tubular corridor F is provided to the vehicles of the vehicle group as data in the digital map.

[0049] In particular, since the calculated lane path is supplied to the trajectory planning module of an autonomous vehicle or a self-driving vehicle, the trajectory of the vehicle can be planned according to the existing lane path.

Prior Art Documents

Patent Documents

[0050]

Patent Document 1

Patent Document 2

Claims

1. A method for calculating and providing a lane route of a road using a central computing unit that is technically connected to vehicles in a vehicle group, comprising: - providing a geometric map (K) including geometric lane boundaries (S), and applying a grid network (G) having grid cells (G1 to G49) of a predetermined size to the map (K); - collecting fleet data of the vehicles in the vehicle group, including a position sequence in which the vehicles in the vehicle group travel; - calculating the orientation of the vehicle at the positions of the grid cells (G1 to G49) from the collected position sequence; - discretizing the calculated orientation of the vehicle; - generating a histogram (H) for each of the discretized vehicle orientations calculated at the positions of the respective grid cells (G1 to G49) for each individual grid cell position; - selecting a map section having a predetermined amount of the grid cells (G1 to G49), and - calculating the lane route in the map section using a learning-based method from the histogram (H) created for the grid cells (G1 to G49) of the map section and the geometric lane boundary (S) in the map section. A method characterized by the above.

2. The calculated lane route is provided so as to be acquirable by the vehicles in the vehicle group. The method according to claim 1, characterized by the above.

3. The calculated lane route is provided to the vehicle as data in a digital map. The method according to claim 1 or 2, characterized by the above.

4. The calculated lane route is supplied to an orbit planning module of an autonomous vehicle in the vehicle group. The method according to claim 1 or 2, characterized by the above.

5. The training of the learning-based method is performed using the created histogram (H) and a ground-truth map created for a predetermined area. The method according to claim 1 or 2, characterized by the above.

Images