Patent Application
20210190534
The present application claims the benefit under 35 U.S.C. § 119 of German Patent Application No. DE 102019220510.0 filed on Dec. 23, 2019, which is expressly incorporated herein by reference in its entirety.
The present invention relates to a method and a device for providing a digital localization map for a vehicle. The present invention also relates to a computer program. In addition, the present invention relates to a machine-readable storage medium.
European Patent Application No. EP 3 293 489 A1 describes a method and a device for providing trajectory bundles for a map data analysis. It proposes an approach for generating trajectory bundles for map data analysis. The approach involves the receiving of data associated with the bounded geographic area. The data is collected from sensors of a plurality of devices traveling in the bounded geographic area, and includes checkpoints that indicate a position, a heading, a speed, a time or a combination thereof.
PCT Application No. WO 2018/126228 A1 describes a method for generating signs and lanes for high-resolution maps for autonomous vehicles.
Localization systems with feature-based, digital localization maps to determine vehicle position and orientation are central system components of automated driving functions.
A grid-based approach or an approach based on road sections are familiar for creating a digital localization map. In the case of the grid-based approach (approach A), the mapping area is subdivided into uniform rectangles, the data within each rectangle constituting a partition, also known as a spatial cluster. This data is then processed independently of data outside of the rectangle.
Another conventional approach (approach B) involves grouping the data which is recorded along the traffic lanes for the purpose of map production. In this case, a partition is an assemblage of successive road sections.
Both conventional approaches A and B allow mapping on a large scale. While the grids in the case of approach A are able to be calculated quickly and with little expenditure, the partitions are not considered for the map density. In creating maps for a larger area (e.g., an entire country), partitions with a great deal of data (e.g., large cities) and partitions with very little data (e.g., rural regions) may exist, which can make a map production process problematic.
A scalability of processes is described in, for example: “Tectonic SAM: Exact, Out-of-Core, Submap-Based SLAM”, Proceedings 2007 IEEE International Conference on Robotics and Automation, 2007.
Various methods for local clustering are also available.
An object of the present invention is to make available an improved method for providing a digital localization map for a vehicle.
According to a first aspect of the present invention, the objective may be achieved with a method for providing a digital localization map. In accordance with an example embodiment of the present invention, the method includes the steps:
It is advantageous that because data is processed together, a highly accurate digital localization map is able to be provided. Computing power for creating the digital localization map may thus advantageously be very well scaled or parallelized.
According to a second aspect of the present invention, the objective may be achieved with a device that is equipped to carry out the example method for providing a digital localization map, in accordance with example embodiments of the present invention.
According to a third aspect of the present invention, the objective may be achieved with a computer program including commands which, upon execution of the computer program by a computer, cause it to carry out the method in accordance with example embodiments of the present invention.
According to a fourth aspect of the present invention, the objective may be achieved with a machine-readable storage medium, on which the computer program is stored.
Advantageous further developments of the method in accordance with the present invention are described herein.
One advantageous further development of the method provides that in step d), a road section is assigned to a cluster. A defined allocation of road sections to clusters is thus carried out, so that parallel processing of road sections advantageously does not occur. This beneficially fosters high systematization in creating the digital localization map.
Another advantageous further development of the method provides that in step d), the clusters are aggregated with the road sections. In doing so, starting from starting points, the clusters are aggregated until they have a defined quantity of data. Good scalability in creating the digital localization map is thus advantageously promoted.
According to another advantageous further development of the method, in step d), the clusters are made as equal as possible in terms of the amount of data. Good parallelization for the computational processing of the clusters may thereby be realized.
Another advantageous further development of the method provides that in step d), upon reaching a defined size, a cluster is no longer further aggregated with road sections. As a consequence, computing power for processing the individual clusters is able to be apportioned very well among the computers available.
One advantageous further development of the method provides that in step d), similar road sections are combined to form clusters. Thus, for example, objects of the road-traffic infrastructure such as all tunnels, all roads with/without crossroad, etc., may be combined, which advantageously simplifies processing of the digital map.
According to another advantageous further development of the method, defined numerical values are assigned to the road sections in steps b) and c). This beneficially promotes easy classification of the road sections.
Another advantageous further development of the method provides that a volume of data of one or more clusters is adapted to a computing power of a map-production device. This advantageously fosters good scalability of the computers for creating the digital localization map.
According to another advantageous further development of the method, features that are defined for defined road sections for the digital localization map and that were ascertained during the classifying of the road sections in steps b) and c), are usable in the vehicle. This advantageously supports the provision of map qualities whose level of quality varies regionally, so that, for example, fully autonomous driving by use of the digital localization map is not possible in defined regions because of reduced quality.
Further measures improving the present invention are presented in greater detail below together with the description of the preferred exemplary embodiments of the present invention on the basis of figures.
In accordance with an example embodiment of the present invention, it is provided that the pertinent data of the individual road sections acquired in each case using sensors be processed in its entirety in clusters C1 . . . Cn. In this way, the computational effort for creating and providing the digital localization map may be very well scaled and parallelized.
Computing power of the various computers provided for this purpose may thereby be very well equalized and utilized in balanced fashion.
Preferably, the road sections are aggregated starting from so-called points of origin, until a certain defined volume of data is reached. For example, such points of origin may be crossroads, tunnels or other similar infrastructure elements.
The approach in accordance with the present invention facilitates an efficient and balanced map-production process, resulting in high-quality localization maps. Balanced map production is thus ultimately facilitated, which leads to higher efficiency in the creation of the digital localization map.
Similar to the above-indicated convention approach B, the example method in accordance with the present inventions suggests an assignment of road sections in relevant areas in the form of clusters C1 . . . Cn. In addition, however, qualitative parameters and complex critical scenarios are taken into account like, for example, crossroads, tunnels, etc., permitting them to be mapped consistently.
As a result, a data-driven clustering approach is thus realized for creating the digital localization map.
The provided approach is based on HAD mapping, with due regard to localization-relevant parameters such as road infrastructure (e.g., tunnels, critical road routings, crossroads, exits from turnpikes, etc.), connection quality and localization quality. As a result, anticipated probability values are assigned to road sections of a road network to be mapped. These probability parameters are used to classify the road sections.
Based on the classifications, clusters C1 . . . Cn are formed, which determines the data that is processed together in creating the digital localization map. A balanced map production with complex scenarios relevant for the HAD map production is thereby made possible.
One advantage of the provided approach is the combining of a data-driven, classifying or categorizing approach, with a clustering process.
Efficient processing of a large volume of data as well as high map quality are very important for producing digital localization maps (HAD mapping, highly autonomous driving) on a large scale. The proposed approach allows scalable map production, resulting in high-quality maps.
The scaling may be achieved by grouping the road network to be mapped into individual road sections C11 . . . Cnn and the data-oriented processing of them in groups (clusters). High quality of the digital localization map may be achieved by combining the road sections based on a qualitative classification of them.
One advantage of such “soft clustering” (utilizing “soft” criteria) is that the process of producing the digital localization map permits good scalability, which is necessary for map production on a large scale. As a result, the digital localization map may be produced in a computationally balanced production procedure. Furthermore, the digital localization map produced in this way advantageously exhibits high quality, after relevant influences have been taken into account in the map production by the clustering.
In the following, the example process provided for creating and providing a feature-based digital localization map in accordance with an example embodiment of the present invention is explained in greater detail:
First, road sections C11 . . . Cnn are calculated along a road network relevant for the digital localization map. In this context, a road section C11 . . . Cnn represents an approximation of possible trajectories of vehicles which are on the road in the road section of the road network, a size of road section C11 . . . Cnn being configurable
A first classification or categorization of road sections C11 . . . Cnn is carried out for an allocation of characteristics specific to the road network. These characteristics encode a complexity and a tendency toward a partitioning of road sections C11 . . . Cnn. For example, a low number could be assigned to a road section C11 . . . Cnn which is dividable (e.g., because of an exceeded data limit), whereas a high number may be assigned to a road section C11 . . . Cnn which is out of the question for a division.
Generally, challenging scenarios for creating the digital localization map, such as tunnels, crossroads, exits from turnpikes, etc., receive a higher number, while country roads more likely receive a low number.
The second classification carried out is based on driving-environment sensor data acquired using sensors and provided by vehicles. In this case, availability and quality of the driving-environment sensor data are taken into account. For example, a road section C11 . . . Cnn on which GPS localization is hardly or not at all possible (as may be the case in tunnels, for example), may be assigned a high number. On the other hand, a road section for which high-quality driving-environment sensor data is available and for which a high localization quality is possible may be assigned a low number.
After the indicated classifications, road sections C11 . . . Cnn are combined utilizing a standard clustering approach, in which starting with a road section C11 . . . Cnn having the highest classification, adjacent road sections C11 . . . Cnn are then added or aggregated iteratively to form cluster C1 . . . Cn, until a further separated road section C11 . . . Cnn has the highest classification, so that an additional cluster C1 . . . Cn is generated. The aim is for clusters to have data volumes that are as similar as possible, so that a computer load for calculating the digital localization map is equalized. Concrete numerical values cannot be indicated here, of course, and are dependent on a wide variety of parameters such as, e.g., number of available computers, size of the road network to be mapped, etc. Characteristic or similar scenarios thereby advantageously remain combined in one cluster, as it were, thus facilitating their processing.
By keeping track of the cluster sizes, it is thus possible “to run” mathematical algorithms for similar cluster sizes, the probability of a further road section being added to a cluster, notwithstanding “fulfilled size”, being reduced.
In a step 200, road sections C11 . . . Cnn are determined by evaluating data of a road network 100 to be mapped.
In a step 210, road sections C11 . . . Cnn are classified, taking defined conditions of road network 100 into account.
In a step 220, road sections C11 . . . Cnn are classified, taking driving-environment sensor data into account.
In a step 230, road sections C11 . . . Cnn are combined to form clusters C1 . . . Cn, taking the classifications carried out in steps b) and c) into account.
In a step 240, in each case joint computational processing of clusters C1 . . . Cn is carried out.
Finally, in a step 250, the digital localization map created is transmitted to a vehicle.
The map-production device for creating the digital localization map is preferably a server device in the cloud, on which the proposed method runs as software. Easy adaptability of the method is thereby facilitated. Preferably, the respective clusters are adapted in terms of the amount of data to the computing power of the individual computers of the map-production device, thus promoting a balanced usage of the computer infrastructure for creating the digital localization map.
In implementing the present invention, one skilled in the art will also realize specific embodiments not specifically explained above.
| Number | Date | Country | Kind |
|---|---|---|---|
| 102019220510.0 | Dec 2019 | DE | national |
