Methods for segmenting trajectories of recorded journeys

Abstract

Method for segmenting a trajectory (1) of a recording drive (2) for swarm data-based HD map creation, wherein sensor data are acquired along the trajectory (1) during the recording drive (2), and wherein the sensor data may include GNSS measurement data, comprising the following steps: a) Identifying points (4) on the trajectory (1) where a position determination is not possible using only GNSS measurement data or where the accuracy of the position determined using only GNSS measurement data does not reach a specified positioning accuracy, and b) Segmenting the trajectory (1) into several segments (3) such that each segment (3) reaches a predetermined minimum length (5) and that neither in the starting area (6) nor in the ending area (7) of each segment (3) is there at least one point (4) identified in step a).

Description

State of the art

[0001] The procedure described here concerns the generation of digital high-resolution map data (so-called HD maps) that can be used for applications of highly automated and possibly autonomous driving.

[0002] Digital maps play a crucial role in the implementation of assisted and automated driving functions in vehicles. Digital maps are data sets that describe the situation along a vehicle's route and, as a supplement to real-time environmental data, enable significantly improved implementation of assisted and automated driving functions than would be possible based solely on real-time environmental data.

[0003] Such digital maps (unlike simpler digital navigation maps used by road navigation systems) contain data about the situation in the immediate vicinity of the route. This allows assisted and automated driving functions to, in a sense, see beyond the visibility horizon of the vehicle's environmental sensors and to better assess and evaluate the environmental data acquired by these sensors. For example, if the environmental sensors detect a specific traffic situation in the immediate vicinity of the vehicle, this situation can be better assessed using data from a digital map that describes the further course of the road. For instance, conclusions can be drawn about how the traffic situation is likely to develop as the journey continues along the road.

[0004] Such digital maps regularly contain a significant amount of data describing the roadway and its immediate surroundings. This includes roadway and marking data, data on objects in the vicinity of the roadway, road topology and lane-level geometry, as well as semantic contextual information such as traffic signs and traffic light positions. The accuracy and depth of information provided by such maps regularly exceeds the amount of information that can be acquired using sensors alone. Extracting comprehensive semantic information directly from raw sensor data in real time presents a major challenge, especially in complex traffic scenarios. By using digital maps that incorporate the aforementioned data, the environmental data acquired by sensors can be enriched and analyzed more efficiently.In particular, the use of digital maps in many situations or for many types of information means that it is only necessary to correctly evaluate the information contained in the digital maps using the sensor data.

[0005] Such digital maps (unlike simple digital navigation maps used by road navigation systems) are not typically available in a uniform format for the entire road network. These digital maps are provided by a wide variety of service providers. Unlike simple digital navigation maps, they are also often not available in a fixed, uniform global coordinate system. The data underlying these digital maps is frequently acquired through test drives of vehicles along routes / roads. Highly individualized coordinate systems are often used to describe the position of the data on the map, referring, for example, to the position and, if applicable, the orientation of the vehicle that performed the test drive. In this case, a coordinate system exists that is referenced to the route.Often, it is not possible to precisely convert digital maps, or the positional data contained within them, into a uniform global coordinate system. This is frequently because, during the acquisition of the data underlying the digital map, no precise positions are available that would be necessary for converting positional data into a global coordinate system. This might be because no GNSS signal is available at all, or the GNSS signal quality during acquisition is insufficient to determine an exact position. A typical example would be a digital map describing the surroundings of a vehicle in a tunnel. In such a situation, no GNSS signal is available. Nevertheless, digital maps are required for assisted and automated driving functions.

[0006] The process described here concerns the processing and preparation of swarm-collected data for the creation and updating of high-resolution digital maps—specifically, so-called HD maps. The initial creation (production) of an HD map and its subsequent updating are distinct tasks that nevertheless share many similarities, particularly regarding the initial collection of the necessary data for creation and / or updating. When the creation of an HD map is mentioned below, this generally also refers to its updating, and vice versa.

[0007] Traditionally, high-resolution maps for applications of highly automated and possibly autonomous driving are created by having manufacturer-specific fleets record data for mapping or by using other information sources to obtain data.

[0008] This traditional approach presents classic map providers with two fundamental challenges today: the ability to provide map updates and the difficulty of collecting sufficient data for very high map accuracy (granularity). Both challenges are related to the size of the vehicle fleet that collects the data used for the map. The data collected so far by classic map providers is typically only sufficient for mapping with the granularity common to conventional in-vehicle navigation systems. Such maps are usually referred to as SD cards. For highly automated and potentially even autonomous driving functions, maps are required that contain data for every single lane of a road and also describe the surrounding area in detail.Such cards are usually grouped under the term "HD card" introduced above.

[0009] To address the two problems of sufficiently frequent map updates and adequate map granularity, approaches to generating maps based on swarm data are currently being discussed. Swarm data is preferably collected indirectly by the customers of vehicle manufacturers – that is, by every single driver who operates a vehicle. During everyday driving, sensor data from the vehicles is sent to the vehicle manufacturer and made available to the map service provider. The map service provider processes this data to create highly accurate HD maps.

[0010] Creating high-definition maps from such swarm data is a complex process that typically involves several steps. The data processing effort required to create high-definition maps from swarm data is regularly very high. Therefore, measures that reduce this data processing effort are generally desirable.

[0011] An important step is the so-called segmentation of the swarm-collected data. This segmentation creates processable data sections from the swarm-collected data, which can be used to create and / or update an HD map.

[0012] The purpose of the method described here is to disclose a particularly advantageous method for segmenting recorded journeys. This purpose is achieved by the method according to the features of the independent claims. Further advantageous embodiments are specified in the dependent claims, as well as in the description and, in particular, in the description of the figures. It should be noted that a person skilled in the art can combine the individual features in a technologically meaningful way and thereby arrive at further embodiments of the invention.

[0013] This section describes a novel method for segmenting or dividing swarm-based collected data or recording drives, which is particularly suitable for the subsequent creation of HD maps.

[0014] The following section presents a novel algorithm for segmenting / dividing journeys into recorded journeys ("drives"). It also highlights the particular advantages of certain segmentation / division methods for specific data processing applications (initial map creation and / or updates of HD maps).

[0015] Herein, a method for segmenting a trajectory of a recording drive for the swarm data-based creation of HD maps is described, wherein sensor data are acquired along the trajectory during the recording drive, and wherein the sensor data can include GNSS measurement data, comprising the following steps: a) Identifying points on the trajectory where position determination is not possible using only GNSS measurement data, or where the accuracy of the position determined solely with GNSS measurement data does not reach a specified positioning accuracy, and b) Segmenting the trajectory into several segments such that each segment reaches a predetermined minimum length, and that neither in the start area nor in the end area of ​​each segment is there at least one point identified in step a).

[0016] One step in using swarm-based data collected during recording runs to generate HD maps is aligning, referencing, or mapping individual recording runs to each other. Referencing or aligning means that identical sensor data acquired during different recording runs are aligned. Alignment is necessary because all sensor data is subject to measurement errors, and the data acquisition during recording runs (as described above) occurs in different coordinate systems. Typically, all objects detected during a recording run (e.g., a traffic light) are assigned specific vehicle-specific coordinates, which are based on global coordinates determined during the recording run. These global coordinates are determined, for example, using GNSS. The accuracy of GNSS is limited.It can be accurate to within centimeters, but especially for vehicles in operation that are not specifically equipped for generating map data, this accuracy is often lower. Furthermore, there are situations where GNSS signals are blocked, such as in a tunnel, as described above. In such situations, odometry data is typically used to determine a global coordinate—a vehicle's current position. Odometry data comes, for example, from wheel speed sensors and steering sensors on vehicles. With such odometry data, vehicle positions can be updated even when GNSS signals are (temporarily) unavailable, thus enabling the determination of global positions. The use of odometry data in addition to GNSS to determine a vehicle's global position is necessary, for example, in tunnels.However, this results in position shifts because the odometry data must be regularly integrated in order to be used for determining global positions.

[0017] Measurement errors that can arise from inaccurate GNSS and / or the use of odometry data when using swarm data to create HD maps can potentially be at least partially corrected by processing a large volume of swarm data. In principle, with suitable approaches, it is possible to generate high-quality HD maps using swarm data. The redundancy of the information available in swarm data can compensate for individual measurement errors.

[0018] Approaches to processing swarm data for creating highly accurate maps are often based on the so-called SLAM method. SLAM stands for Simultaneous Localization and Mapping. In the SLAM method, all available additional information about a vehicle's surroundings, acquired by sensors, is compared with information already available in the form of a map, and the map is improved or corrected with this additional information. This involves a continuous supplementation or correction of map data based on additional sensor data. The SLAM approach is comparable to the way the human brain typically perceives its environment. A person, while observing a room, knows what the room looks like and what is in it. A kind of virtual map of the room exists in the person's mind.Through the perception of space, humans automatically supplement their image of their surroundings, or the virtual map of the environment that they have in mind.

[0019] Regarding the processing of swarm data for the production of highly accurate maps, the SLAM method is preferably used to obtain aligned driving trajectories or route profiles of the individual journeys from which the swarm data was collected. The sensor measurements of all sensors of a vehicle are recorded relative to the vehicle's position when determining the respective data. Errors in determining the vehicle's position are reflected in the driving trajectory or route profile of the vehicle's journey. Aligning the driving trajectories or route profiles of the individual recording journeys from the data collection process can typically be achieved using the SLAM method. Preferably, landmarks detected by the vehicle's sensors from different recording journeys are matched and used to reference / align the recording journeys relative to each other.In particular, such a step makes it possible to use swarm data to create HD maps.

[0020] However, the approach of using swarm data to generate HD maps also presents further new challenges. To align journeys, journeys that have covered a certain distance are required. Only then are there sufficient features (especially enough landmarks in the vicinity of the journey trajectory or route) to allow referencing and alignment of the individual recorded journeys. An upper limit on journey length is not strictly necessary. However, a limit does exist to restrict the volume of data packets associated with a recorded journey that are transmitted from the vehicle to the central instance.

[0021] In preferred implementations, recording journeys, or the data packets processed as a recording journey, each represent segments of routes with a specific length. This length can be defined in some implementations by a unit of measurement (e.g., meters or kilometers). It can also be defined by time, so that, for example, a defined time interval of a recording vehicle's journey is interpreted as a recording journey. This division serves the purpose of keeping the data packets processed for recording journeys within a specific size range. The division of an actual journey of a recording vehicle along a specific route (e.g., from Frankfurt to Munich) into individual recording journeys or data packets for processing to create map data can also be referred to as splitting or segmenting the recording journeys or data packets.The use of the SLAM method is significantly simplified by creating segments of the recorded journeys within specific length ranges. By forming segments with suitable characteristics, the data that can be fed into a SLAM algorithm is standardized. This enables a more efficient execution of the SLAM algorithm.

[0022] The formation of segments for recorded journeys is preferably carried out in such a way that the start and end areas of adjacent segments partially overlap, or segment overlap areas are formed in which data is present in both adjacent segments. This makes it possible to reconnect or process the segments together in subsequent data processing procedures.

[0023] A suitable, particularly advantageous cutting, division, and / or segmentation of recorded journeys and data packets enables highly efficient processing of these recordings and data packets for map production. In particular, the parallel processing of a large number of recorded journeys can be facilitated by a suitable cutting, division, or segmentation. Such a cutting or division can also be chosen differently depending on the situation. For example, a distinction can be made between whether the journey is on a highway, a rural road, or in an urban area.

[0024] There is a strong interest in keeping the number of recorded trips used to generate map data within a specific range. At the same time, the frequency with which swarm-based data collection or recorded trips are carried out for specific road sections varies considerably. For example, while a large number of vehicles travel on a highway, resulting in a large number of (potential) recorded trips for generating data packets, trips are conducted much less frequently on secondary roads. Consequently, there is also a significantly smaller amount of data available for secondary roads.

[0025] By appropriately editing or dividing / segmenting recorded journeys, it is possible to ensure that the recorded journeys, or rather the corresponding data packages, are well-suited for generating HD maps. While it may make sense to use more data for main roads than for secondary roads when creating HD maps, suitable segmentation makes it possible to select from the total amount of collected data for HD map creation and achieve a tailored data coverage.

[0026] As described above, when creating HD maps from swarm data, a fundamental distinction must be made between the initial creation of the HD map and its regular updates (required due to changes in travel routes). During the initial creation of an HD map, no map yet exists, and data packets from a large number of recorded trips must be combined to generate a first (initial) HD map. HD map updates are regularly necessary because HD maps typically contain a multitude of features that can change frequently. Such features might include, for example, construction sites. The high level of detail in HD maps alone means that changes occur more often. Updates may require replacing small sections of the map.Generally, when updating HD cards, it's preferable to modify the card as little as possible. This can be advantageous, for example, if the card is security-relevant and needs to be secured through a review or other measures.

[0027] The two different phases or types of map creation (initial map creation and updating / revision of HD maps) present different challenges for data processing, particularly when processing data acquired from swarm data. These different challenges can be addressed and resolved through appropriate segmentation / division of the data or recording drives.

[0028] The approach described here for segmenting recorded journeys and their trajectories comprises two fundamental steps: First, points with poor GNSS accuracy are identified. Second, segments are created such that no such points exist in either the start or end regions of the segments.

[0029] The first step, a) of identifying sections of the journey with poor GNSS measurement data (e.g., in tunnels), is important because segments should preferably not be generated during which no GNSS measurement data or only poor GNSS measurement data was acquired. This applies particularly to the start and end regions of the segments. Good GNSS measurement data should be available for the start and end regions of the segments whenever possible. For example, if a segment had no GNSS measurement data only at the beginning but none at the end, odometry drift during SLAM optimization could potentially produce invalid results. The GNSS measurement data is available in recording journeys for specific segments.The drive data packages belonging to these recording drives are important for describing a route that is used when recording the data in the data package with the sensors on the recording vehicle to spatially reference all recorded data. In areas where no GNSS measurement data is available, this route must be determined solely based on odometry data, which is obtained, for example, from the vehicle's wheel speed sensors and / or steering angle sensors. Such odometry data can typically be subject to drift. To limit the effect of drift in the odometry data, it is a significant advantage if segments are generated in such a way that at the beginning and end of each segment, respectively,GNSS measurement data is always available for each recording run, which can be used to describe the route, and a description of the route is only provided between the beginning and the end of each segment, exclusively based on odometry data.

[0030] When the process described here refers to segmenting trajectories, it means that all data collected during the recording drive is segmented along the trajectory. During segmentation, data packages are created from the data collected during the recording drive, each package being assigned to a section (segment) of the trajectory. These packages are large enough to allow for efficient processing in the creation and / or updating of HD maps.

[0031] The method is particularly advantageous if the minimum length of the segments is defined as greater than 200 meters.

[0032] The minimum segment length can also be greater, e.g., 500 meters. The minimum length is preferably chosen to ensure efficient data storage of the segments and to allow algorithms to process the data contained within them efficiently.

[0033] This minimum length corresponds to the distance traveled by the recording vehicle while collecting data. Segments must be long enough to contain sufficient information. At the same time, the length should not be so long that unnecessarily large amounts of data have to be considered for updating and processing the data.

[0034] It is particularly preferred if in step b) the trajectory has a starting point (8) and an endpoint, such that the trajectory from the starting point to the endpoint is segmented into several consecutive segments according to the minimum length, wherein a segment is combined with an immediately preceding and / or following adjacent segment (3) to form a new segment (3) if a point identified in step a) is located in the starting area and / or in the end area.

[0035] A swarm-based data package typically begins with data recorded during a (complete) vehicle journey. For example, a customer's vehicle travels from a driver's home to their workplace. The starting point is at the home, and the endpoint is at the workplace. If this route is 10 kilometers long in total, it can be divided into 20 segments, each 500 meters long, using the described method.

[0036] This document describes in particular a procedure for providing an HD map from swarm data collected from customer vehicles, comprising the following steps: A) Receiving data relating to a large number of recording trips with customer vehicles such that a trajectory and corresponding sensor data are recorded for each recording trip, whereby the sensor data may include GNSS measurement data, B) Segmenting the respective trajectories recorded in step a) into several segments (3) according to the procedure with steps a) and b), C) Assigning segments to predefined map sections, so that several groups of segments are formed, each corresponding to a map section, so that several map sections can be generated from several groups and combined to form a complete localization map, and D) Providing an HD map from the formed groups.

[0037] The goal of the procedure according to steps A) to D) is to generate the most uniform possible data density of the collected swarm data for the subsequent creation and / or updating of an HD map. The segmentation described in steps a) and b) creates the prerequisites for this.

[0038] Step A) involves receiving data from recording runs, preferably captured by a swarm of customer vehicles. This data includes, in particular, a trajectory describing the route taken by the recording vehicle during the recording run, as well as other data acquired by sensors during the recording run. The trajectory in this data is preferably generated using GNSS measurement data / signals and is most preferably in global coordinates. A global position can preferably be assigned to all data acquired during the recording run (e.g., data concerning objects on the roadway) via the trajectory. However, this global position is subject to errors for the reasons mentioned and due to the uncertainties mentioned.

[0039] The segmentation of the received recorded journeys is carried out according to steps a) and b) in step B).

[0040] In step C), each created segment of a recording trip is then assigned to a map section. The map sections are preferably defined by a fixed grid that corresponds to the area to be mapped. All generated segments are added to this grid. This grid of map sections is primarily used to later identify recording trips for a specific map section.

[0041] Step D) then concerns the actual production of the HD map. According to step D), due to the preparatory work carried out in steps A) to C), the HD map can be created map section by map section, and map sections can also be selectively updated.

[0042] By assigning or grouping segments of recorded journeys according to map sections, the processability of the segments for creating HD maps is significantly improved. Segments from individual map sections, or potentially adjacent map sections, can be selectively fed into a SLAM algorithm for aligning segments with each other.

[0043] Assigning segments to map sections significantly increases the potential for parallelizing data processing to create HD maps. SLAM optimization for aligning segments of recorded journeys can be performed in parallel for various locations defined by the map section grid.

[0044] By assigning segments of recorded journeys to map sections that form a grid, a search structure can be created that allows for the efficient retrieval of these segments. This can be used, for example, to determine data for updates.

[0045] The procedure is particularly advantageous if, between step A) and step D), it is analyzed whether the number of segments in each group exceeds an upper threshold or falls below a lower threshold.

[0046] In this context, it is preferred to remove excess segments in a group if the number of segments in that group exceeds the upper threshold.

[0047] It is also preferred to insert new segments into a group if the number of segments in that group falls below the lower threshold.

[0048] An upper threshold could, for example, be ten segments per group or for a map section. A lower threshold could, for example, be five. Ideally, the entire area or road system to be mapped should be covered as uniformly as possible in the form of recorded drive segments. This allows for the creation of a high-definition map of consistent quality. The upper and lower thresholds enable this. A system for operating the described method is particularly advantageous if it is also configured to request additional, targeted recording drives as needed, for example, if insufficient swarm-based data is available in certain areas to be mapped.

[0049] Furthermore, the method is preferred if step D) providing a localization map includes generating an initial localization map.

[0050] Furthermore, the method is preferred if step D) providing a localization map includes updating the initial localization map, updating only a selected map section.

[0051] The described method is applicable and advantageous for both use cases, especially due to the grouping of segments according to map sections.

[0052] Furthermore, it is advantageous if the selected map section (especially according to step C) is updated so that new segments that can be assigned to this map section are added to the corresponding group, and existing segments are removed from this group.

[0053] In particular, it is possible to ensure that for each map section, there are always sufficient segments of recorded journeys with a specific date of issue (e.g., no older than one year, or no older than one week if the map section contains a construction site). This allows a data storage system to always provide recorded journeys with a predetermined date of issue for the segments of recorded journeys.

[0054] Furthermore, it is preferred if in step D) the segments in each group are aligned to each other in order to form an aligned pose graph.

[0055] A pose graph included all poses of the recording vehicles during the various recording runs, thus enabling the correlation of data collected in different recording runs with each other.

[0056] Aligning segments of recorded journeys is a crucial step in processing data from different segments to create or update an HD map. This proposal suggests first generating the recorded journey segments and then storing them unaligned (as segmented raw data) in the data storage system. This allows access to these segmented raw data for creation and / or updates. This approach prevents information loss that might occur during data alignment from affecting the data storage, ensuring that the HD map is essentially created using the original recorded data.

[0057] Furthermore, it is preferred if all segments obtained in step B) are combined into a data structure and stored in a central data storage system, the data structure being designed so that new segments can be inserted and already stored segments can be deleted.

[0058] It is particularly preferred if, in step C), the assignment is carried out using a kD tree by inserting the midpoints of the respective segments into the kD tree.

[0059] A kD tree, or "k-dimensional tree," is a balanced search tree for storing spatial data. It offers the possibility to efficiently search for stored data.

[0060] This creates an efficient data management structure, which allows the captured raw data to be advantageously provided in a data provision for map generation, in order to later create and / or update HD maps from it (at any time).

[0061] This section will also describe a mapping system for producing HD maps with - a data collection facility for collecting data, wherein segments of recording journeys are formed and collected in the data collection facility according to the described procedure in accordance with steps a) and b), - a central data storage device for storing segments of recorded journeys, wherein the central data storage device is designed in such a way that new segments can be inserted into the central data storage device and existing segments can be removed from the central data storage device, and - a data alignment device for generating an aligned pose graph based on the segments stored in the central data storage.

[0062] It should be noted that the special advantages and design features described above are also applicable and transferable to the mapping system.

[0063] The mapping system is preferably operated at a central instance and serves to provide users with constantly up-to-date HD maps.

[0064] The mapping system is specifically designed to manage, store and process segments of recorded journeys according to steps A) to D).

[0065] The invention and its technical context are explained in more detail below with reference to the figures. The figures show preferred embodiments, to which the invention is not limited. It should be noted in particular that the figures, and especially the size relationships shown in the figures, are only schematic. They show: Fig. 1: A road system consisting of a main road with side streets, with schematically represented recorded journeys; Fig. 2a, Fig. 2b: Examples of recording runs in the area of ​​a route through a tunnel; Fig: 3a, 3b: Examples of the segmentation of a recording vehicle's journey into recording journeys in the area of ​​tunnels; Fig. 4a, Fig. 4b: Examples of which segmented recording trips are used to update a map section; and Fig. 5: A flowchart and a device for creating HD maps using the method described here.

[0066] The Fig. Figure 1 shows a road system with a main road 11 and side roads 12 branching off from the main road 11. This is a simple example of a road system to be mapped, or from which an HD map is to be created. Regular vehicle traffic takes place on the road system. These vehicles are preferably ordinary customer vehicles carrying out normal journeys that are not primarily intended for data collection to create a map. However, during these journeys, sensors record trips, so that the trips serve as recording trips 2. Each recording trip 2 has a trajectory 1 that describes the path the respective vehicle travels during the recording trip 2. Fig. It can be seen that, in principle, more trips take place on main roads (11) that can be used as recording trips (2) than on secondary roads (12). This is primarily due to the fact that main roads (11) carry more traffic than secondary roads. It can also be seen that recording trips (2) carried out by ordinary customer vehicles (2) are not subject to any central planning. From a data acquisition perspective, the routes / roads driven are more or less random. The natural distribution of trips regularly creates somewhat unfavorable concentrations of recording trips for the creation of HD maps. This results in conventional algorithms for processing the swarm data collected in this way being supplied with an unnecessarily large amount of data. This costs computing time and increases the effort required for map production without creating any added value.The proposed approach of segmenting recording trips according to steps a) and b) and further processing the segmented recording trips according to steps A), B), C), and D) reduces this problem. Overall, these approaches reduce the computational effort required to generate, provide, and update HD maps.

[0067] The Fig. 2a and Fig. Figure 2b shows examples of recording runs 2, some of which run through tunnel 13. Fig. 2a and Fig. Figure 2b shows that, under the strict application of the minimum length 5 for dividing a recording run 2 into segments 3, it is possible that start areas 6 and / or end areas 7 of segments 3 terminate in tunnels 13. GNSS reception is poor in tunnels 13, so all points on the trajectories 1 of the recording runs in tunnels 13 are points 4 where precise GNSS positioning is not possible. An example of such a point 4, located in an end area 7 of a segment 3, is shown here. This concerns a segment 3 that terminates in a tunnel 13.

[0068] In such an area with poor GNSS measurement data, determining the trajectory 1 of recording run 3 is possible (only) using odometry data. Fig. Figure 2b illustrates (in an enhanced manner) how drift in odometry data can affect the results. Segment 3 of the recorded journey 2, or rather its trajectory 1, exhibits a highly erroneous path, which could be caused, for example, by drift in the odometry data. To avoid such effects, it is advantageous that, particularly in the start area 6 and the end area 7 of segment 3, no points 4 with poor GNSS accuracy are located. Start areas 6 and end areas 7 of the segments regularly form overlapping areas 19 with other segments 3. Here, segments 3 must be related to other segments 3 for the execution of a SLAM algorithm. High positional accuracy determined by GNSS, which cannot be influenced by drift in the odometry data, is essential here.

[0069] Figures 3a and 3b show, along a longer route of a recording run 2, how the recording run 2 can be divided into segments 3 to avoid starting areas 6 or ending areas 7 of the segments terminating in tunnels 13. According to the Fig. 3a and Fig. 3b shows in particular the execution of the procedural steps a) and b), by carrying them out it can be avoided that segments 3 with points 4 in the start area 6 or end area 7 are created.

[0070] The following is shown in the Fig. 3a and Fig. 3b each a route that runs through a tunnel 13. The in the Fig. 3a and Fig. The route shown in 3b has a starting point 8 and an end point 9 and was subdivided into segments 3 or segmented recording journeys 2 for better processing in order to produce HD maps / mapping.

[0071] According to the Fig. 3a only applied a minimum length of 5 to divide the route into segments 3. It can be seen that for individual segments, the starting area 6 and / or the ending area 7 then terminate in tunnels 13 and thus have points 4 in the starting area 6 and / or the ending area 7 where there is poor positional accuracy based on GNSS measurement data.

[0072] According to Fig. 3b such segments 3 with a start area 6 or an end area 7 were now joined with other segments 3 in such a way that no segments 3 remain whose start area 6 or whose end area 7 lies in a tunnel 13 and thus in an area where there is poor position accuracy based on GNSS measurement data.

[0073] The Fig. 4a and Fig. 4b relates in particular to the described procedure according to steps A), B), C) and D), in which the procedure according to steps a) and b) is embedded. According to Fig. 4a and Fig. Figure 4b shows a multitude of segments 3, which contain data relating to a specific route section 20 and on which an HD map can be created and / or updated. The segments 3 are (as described) sections of recording journeys 2. Each segment 3 is assigned the data acquired by that segment 3 during the recording journey 2.

[0074] According to The Fig. 4a and Fig. Section 4b now presents the problem that a specific map section 10 of an existing, already created HD map is to be updated. The in Fig. 4a and Fig. The segments 3 shown in section 4b are stored in a database containing numerous segments 3 from recorded journeys 2, along with their associated data. These segments 3 were typically collected using a swarm approach. The segments 3 stored in the database are usually time-stamped, indicating when they were recorded. To update the map section 10, the most recent segments 3 are used whenever possible. Furthermore, it is important to use segments 3 that contain information relevant to the map section 10 being updated.

[0075] Fig. Figure 4a shows an example of a total set of segments 3 that may be stored in a database in the vicinity of a map section 10 to be updated.

[0076] Fig. Figure 4b shows, as an example, new segments 21 that form a subset of the total set of segments 3. A distinction is made between these new segments 21, shown as dashed lines, which lie entirely outside the map area 10 to be updated, and those new segments 21 that do affect the map area 10. Only the new segments 21 shown as solid lines contain additional information suitable for updating the map area 10. Preferably, only these new segments 21 are used to update the map area 10.

[0077] According to Fig. Section 5 schematically illustrates the procedure for segmenting recording trips, comprising steps a) and b), and the (overarching) procedure for providing the HD map, comprising steps A) to D). While steps A) to D) describe the overarching process for producing the HD map, steps a) and b) relate to the execution of step B) or are substeps of step B).

[0078] Both processes are executed using a data processing system 22, which is schematically divided into components here. This division is not mandatory. The data processing system 22 can also be structured differently. The representation is only exemplary. The components of the data processing system 22 shown here are a data collection unit 14, a data storage unit 18, which is further subdivided into a data update unit 15 and a data reduction unit 16, and a data processing unit 17.

[0079] The data collection device 14 serves first and foremost to collect raw data that can be used to create HD maps. Raw data is collected, for example, from recording trips 2 with fleet vehicles in the field. Such vehicles preferably record raw data during their regular operation (swarm-based), which can then be used with the data processing system described here to create HD maps. Step A) shown here therefore preferably takes place outside the data processing system 22. The recorded data is preferably transmitted from vehicles in the field to the data processing system 22, which is operated by a central instance, via a transmission interface 23 (e.g., a mobile network connection). The raw data initially contains recording trips in non-segmented form. In the raw data, the collected data is typically simply represented along the routes that the vehicles regularly traveled in the field. This can, for example,The route taken by drivers of vehicles in the field is to work and / or shopping. Process step B), with sub-steps a) and b), serves to generate segments from this raw data that are suitable for further processing to create and / or update HD maps. The segmentation is described in detail above and illustrated in the figure. Fig. 5 of the data collection facility 14 assigned.

[0080] The data collection process initially takes place independently of the subsequent generation and / or updating of the HD map. The segments of recorded journeys generated by the data collection unit 14 are stored in the data storage unit 18 to be used later for the generation and / or updating of HD maps. The data storage unit 18 is simplified here with a data update 15 and a data reduction 16. The purpose of the data storage unit 18 is to store and provide only data or segments 3 that offer relevant added value for the generation and / or updating of an HD map. For example, if several thousand segments of recorded journeys along a main road are collected by the data collection unit 14, it is generally not practical to store all of these segments in the data storage unit for later use in generating HD maps.The information in the individual segments is then largely redundant. Rather, the task of the data storage facility 18 is to provide the most up-to-date segments 3 possible, ensuring uniform coverage of the roads and routes to be mapped. With data update 15, newly added relevant segments are preferentially added. With data reduction 16, segments that are old or have been replaced by better, more up-to-date segments are preferentially removed. The data storage facility 18 is therefore preferably configured to always provide the most up-to-date raw data set possible, exhibiting the most uniform quality possible across all roads and routes to be mapped. This is essentially achieved by carrying out process step C), which is shown in the representation according to... Fig. 5 is assigned to the data storage facility 18.

[0081] By appropriate segmenting according to steps a) and b), a granularity of data stored in data storage facility 18 is achieved. This granularity increases the efficiency of data storage facility 18. If complete real-world vehicle journeys in the field (e.g., a vehicle user's commute to work or shopping) were stored in data storage facility 18, these complete real-world journeys would also have to be added or removed. Segmentation makes it possible to add only particularly relevant segments of a real-world journey, which offer added value for the creation and / or updating of the HD map, to the data set provided in data storage facility 18.Because individual segments are always generated with overlapping areas to neighboring segments in the start and end areas, a complete image of each individual map section of a system of roads and routes to be mapped can be generated at any time using the segments.

[0082] The data processing unit 17 accesses the data stored in the data storage unit 18 and is responsible for the actual generation and provision of the HD map based on the raw data. The data processing unit 17 can preferably be used both for generating a (new) HD map and for updating map sections 10 of an existing HD map.

Claims

[1] Method for segmenting a trajectory (1) of a recording drive (2) for swarm data-based HD map creation, wherein sensor data are acquired along the trajectory (1) during the recording drive (2), and wherein the sensor data may include GNSS measurement data, comprising the following steps: a) Identifying points (4) on the trajectory (1) where a position determination is not possible using only GNSS measurement data or where the accuracy of the position determined using only GNSS measurement data does not reach a specified positioning accuracy, and b) Segmenting the trajectory (1) into several segments (3) such that each segment (3) reaches a predetermined minimum length (5) and that neither in the starting area (6) nor in the ending area (7) of each segment (3) is there at least one point (4) identified in step a). [2] Method according to claim 1, wherein the minimum length (5) is defined as greater than 200 meters. [3] Method according to claim 1 or 2, wherein in step b) the trajectory (1) has a starting point (8) and an end point (9), such that the trajectory (1) is segmented successively into several segments (3) from the starting point (8) to the end point (9) according to the minimum length (5), wherein a segment (3) is combined with an adjacent segment (3) immediately before and / or after it to form a new segment (3) if a point (4) detected in step a) is located in the starting area (6) and / or in the end area (7). [4] Method for providing an HD map from swarm data collected from customer vehicles, comprising the following steps: A) Receiving data relating to a multitude of recording journeys (2) with customer vehicles such that a trajectory (1) and sensor data corresponding to the trajectory (1) are recorded for each recording journey (2), wherein the sensor data may include GNSS measurement data, B) Segmenting the respective trajectories recorded in step a) into several segments (3) according to the method according to one of claims 1 to 3, C) Assigning segments (3) to predefined map sections (10) such that several groups of segments (3) are formed, each of which corresponds to a map section (10), so that several map sections (10) can be generated from several groups and joined together to form a complete localization map, and D) Providing an HD map from the formed groups. [5] Method according to claim 4, wherein between step A) and step D) it is analyzed whether the number of segments (3) in each group exceeds an upper threshold or falls below a lower threshold. [6] Method according to claim 5, wherein excess segments (3) in a group are removed when the number of segments (3) in that group exceeds the upper threshold. [7] Method according to claim 5, wherein new segments (3) are inserted into a group when the number of segments (3) in that group falls below the lower threshold. [8] Method according to any one of claims 4 to 7, wherein in step C) providing a localization map comprises generating an initial localization map. [9] Method according to claim 8, wherein in step D) providing a localization map comprises updating the initial localization map, with only a selected map section (10) being updated. [10] Method according to claim 9, wherein in step D) the selected map section (10) is updated such that new segments (3) that can be assigned to this map section (10) are added to the corresponding group and existing segments (3) are removed from this group. [11] Method according to any one of claims 4 to 10, wherein in step D) the segments (3) in each group are aligned to each other to form an aligned pose graph. [12] Method according to any one of claims 4 to 11, wherein all segments (3) obtained in step B) are combined in a data structure and stored in a central data storage, wherein the data structure is designed such that new segments (3) can be inserted and already stored segments (3) can be deleted. [13] Method according to any one of claims 4 to 12, wherein in step C) the assignment is carried out using a kD tree by inserting the midpoints of the respective segments (3) into the kD tree. [14] Mapping system for producing HD maps with - a data collection device (14) for collecting data, wherein segments (3) of recording journeys (2) are formed and collected in the data collection device (14) according to the method according to one of claims 1 to 3, - a central data storage device for storing segments (3) of recording journeys (2), wherein the central data storage device is designed such that new segments (3) can be inserted into the central data storage device and existing segments (3) can be removed from the central data storage device, and - a data alignment device (17) for generating an aligned pose graph based on the segments (3) stored in the central data storage.

Images