Method for segmenting trajectory of driving record

Abstract

A method for segmenting a trajectory (1) of a driving record (2) for formulating an HD map based on population data, sensor data being acquired along the trajectory (1) during the driving record (2), which sensor data may comprise GNSS measurement data, the method comprises the following steps: a) detecting points (4) on the trajectory (1) where the position cannot be determined by using the GNSS measurement data alone or the position accuracy determined by using the GNSS measurement data alone does not reach the predetermined positioning accuracy; and b) segmenting the trajectory (1) into a plurality of segments (3) in such a way that each segment (3) reaches a predetermined minimum length (5) and that the at least one point (4) detected in step a) is not located within the starting region (6) and the ending region (7) of each segment (3).

Description

Technical Field

[0001] The method described herein involves generating digital high-precision map data (i.e., so-called high-definition maps / HD maps) that can be used in highly automated and potentially autonomous driving applications. Background Technology

[0002] Digital maps play a crucial role in enabling driver assistance and autonomous driving functions in vehicles. As a dataset describing the environmental conditions along a vehicle's route and supplementing real-time environmental data, digital maps significantly improve the performance of driver assistance and autonomous driving functions, exceeding what can be achieved based solely on real-time environmental data.

[0003] These digital maps (distinct from the simpler digital navigation maps used in road navigation systems) include environmental condition data in the immediate vicinity of the route. In a sense, this allows assisted driving and autonomous driving functions to extend beyond the field of vision of the vehicle's environmental sensor systems and to better analyze and evaluate the environmental data acquired by these systems. For example, if the environmental sensor system detects a specific traffic condition in the immediate vicinity of the vehicle, data from a digital map describing the extended conditions of the road can be used to better assess that condition. For instance, this can lead to inferences about the likely evolution of that traffic condition along the subsequent journey along the road.

[0004] Such digital maps typically include a wealth of data describing routes and their adjacent areas, such as historical data on routes and road markings, data related to objects around the route, lane-level road topology and geometry, and semantic contextual information (e.g., the locations of traffic signs and traffic lights). The accuracy and richness of information in such maps often exceed what can be obtained using sensors alone. Extracting comprehensive semantic information directly from raw sensor data in real time (especially in complex traffic scenarios) presents significant challenges. By using digital maps that include data illustrated graphically, environmental data acquired using environmental sensors can be evaluated more richly and efficiently. In particular, the use of digital maps means that in many scenarios or with large amounts of information, it is sufficient to correctly evaluate the information contained in the digital map based solely on sensor data.

[0005] These types of digital maps (unlike the simple digital navigation maps used in road navigation systems) are typically not available in a standard format for the entire road network. These digital maps are provided by various service providers. Unlike simple digital navigation maps, these types of digital maps are usually not available in a fixed standard global coordinate system. The data upon which these digital maps are based is typically acquired through test drives of vehicles along routes / roads. Very independent coordinate systems are often used to describe the location of the data in the map, each of which typically refers to, for example, the position of the vehicle performing the test drive, and, if necessary, the alignment of the vehicle performing the test drive. In this case, there is a coordinate system referencing the route. It is often impossible to accurately convert the location information of a digital map or individual objects included in a digital map into a standard global coordinate system. This is often because the precise location required to convert the location information into a global coordinate system cannot be obtained during the acquisition of the data upon which the digital map is based (e.g., due to the lack of GNSS signals or insufficient GNSS quality at the time of acquisition to determine precise positioning). A very typical example could be a digital map describing the surrounding environment of a vehicle inside a tunnel. Here, no GNSS is available at all. However, in such cases, driver assistance and autonomous driving functions require digital maps. Summary of the Invention

[0006] The methods described herein involve processing and conditioning data collected in a population-based manner to create and update high-precision digital maps (specifically, so-called HD maps). The initial creation (production) and updating of HD maps are therefore distinct tasks, but they share many similarities, particularly in the collection of data required for initial production and / or updating. When HD map production is mentioned below, it generally refers to updating, and vice versa.

[0007] Traditionally, high-precision maps are created for highly automated and potentially autonomous driving applications, relying on manufacturer-specific fleet log data used for map building or other information sources used to acquire data.

[0008] Currently, this traditional approach presents two fundamental challenges for typical map providers: providing options for map updates and the difficulty of collecting enough data to achieve highly accurate (granular) maps. Both challenges are related to the fleet size of the vehicles collecting map processing data. To date, the data collected by typical map providers is usually only sufficient for mapping at the conventional granularity of typical navigation system operation in vehicles. Such maps are often referred to as SD maps. Highly automated, and potentially even fully autonomous driving functions, require maps that include data for each individual lane of the road and additionally, a detailed description of the road environment. Such maps typically fall under the category of the previously introduced "HD maps" terminology.

[0009] To address the challenges of sufficiently frequent map updates and adequate map granularity, methods for creating maps based on swarm data have recently been discussed. Swarm data is preferably collected indirectly by the vehicle manufacturer's customers (i.e., each driver operating the vehicle). During everyday driving, sensor data from the vehicle is sent to the vehicle manufacturer and then transmitted to map service providers. The map service providers process this data to compile highly accurate HD maps.

[0010] Producing HD maps from such group data is a complex process typically involving multiple steps. The data processing workload for compiling HD maps from group data is usually very high. In principle, measures that can be used to reduce the data processing workload are desirable.

[0011] A key step is the so-called "segmentation" of the data collected in a population-based manner. This segmentation produces processable data portions from the population-based data, which can be processed to compile and / or update HD maps.

[0012] The purpose of the method described herein is to disclose a particularly advantageous method for segmenting recorded driving. This objective is achieved by the method featuring the characteristics described in the independent claim. Other advantageous configurations are found in the dependent claims, the specification, and the description of the drawings. It should be noted that those skilled in the art can combine the various technical features with each other in a technically meaningful manner to obtain other configurations of the invention.

[0013] This specification relates to a novel method for segmenting / segmenting data or driving records collected in a population-based manner, which is particularly suitable for subsequent HD map compilation.

[0014] The following presents a novel algorithm for segmenting / dividing a trip into driving records (“drives”). It also describes the particular advantages of this type of segmentation / division for specific applications of data processing (initial map creation and / or updating of HD maps).

[0015] This specification first describes a method for segmenting a trajectory of a driving record for compiling HD maps based on group data, wherein sensor data is acquired along the trajectory during the driving record, and wherein the sensor data may include GNSS measurement data, comprising the following steps:

[0016] a) Points on the detection trajectory whose location cannot be determined solely using GNSS measurement data, or whose location determined solely using GNSS measurement data does not achieve the predetermined positioning accuracy; and

[0017] b) Divide the trajectory into multiple segments in such a way that each segment reaches a predetermined minimum length and at least one point detected in step a) is not located in the start and end regions of each segment.

[0018] One step in creating HD maps using data collected in a group-based manner from driving logs is to align, reference, or map the individual driving logs to each other. Reference or alignment refers to aligning identical sensor data identified in different driving logs to each other. This alignment is necessary because all sensor data contains measurement errors, or data is collected in different coordinate systems during the driving log (as described above). Typically, all objects detected during the driving log (e.g., a group of traffic lights) are assigned specific vehicle-specific coordinates based on global coordinates determined during the driving log. Such global coordinates are often determined via GNSS, for example. GNSS has limited accuracy. It can be in the centimeter range, but its accuracy tends to be low, especially for vehicles without dedicated settings for generating map data. Furthermore, there are situations where GNSS signals are blocked, such as in tunnels as described above. In such cases, odometer data is often additionally used to determine the global coordinates of the vehicle's current position. For example, odometer data comes from the vehicle's wheel speed and steering sensors. Such odometer data can be used to update the vehicle's position even when GNSS signals are (temporarily) unavailable, meaning that the global position can also be determined at this time. For example, in tunnels, in addition to GNSS, odometry data is needed to determine the vehicle's global position. However, this can lead to position drift because odometry data often needs to be integrated in order to be used to determine the global position.

[0019] When using population data to create HD maps, measurement errors caused by inaccurate GNSS and / or the use of odometry data can likely be at least partially corrected by the large volume of processed population data. In principle, appropriate methods allow for the creation of high-quality HD maps using population data. The redundancy of available information in the population data can compensate for individual measurement errors.

[0020] Methods for processing population data to generate highly accurate maps are often based on so-called SLAM techniques. SLAM stands for Simultaneous Localization and Mapping.

[0021] SLAM (Simultaneous Localization and Mapping) involves comparing all available supplementary information about the vehicle's surroundings, determined by sensors, with existing information in map form, and using this supplementary information to improve or correct the map. This means continuously adding to or correcting map data using additional information determined by sensors. SLAM is analogous to how the human brain perceives its surroundings. When people observe a room, they know what it looks like and what's in it. A virtual map of the room exists in the human mind. Through this perception, humans automatically add to an image of their surroundings, or to a virtual map of their environment.

[0022] Regarding the processing of swarm data to generate highly accurate maps, a SLAM method is preferably used to acquire mutually aligned driving trajectories or routes of individual trips for collecting swarm data. When determining the corresponding data, sensor measurements from all vehicle sensors relative to the vehicle's position are recorded. Errors in determining the vehicle's position are reflected in the driving trajectories or routes of the vehicle's trips. Typically, SLAM methods can be used to align the driving trajectories or routes of individual driving records in the data collection. Preferably, landmarks detected by vehicle sensors in different driving records are assigned to each other and used for mutual reference / alignment between driving records. In particular, this step allows for the creation of HD maps using swarm data.

[0023] However, using group data to create HD maps also introduces new challenges. For trips to be aligned with each other, trips covering a certain distance are required. Only then will sufficient features (especially enough landmarks in the surrounding environment of the driving track or route) exist to allow individual trip records to be referenced or aligned with each other. There is no mandatory upper limit on trip length. However, in principle, there is an upper limit to the number of data packets related to driving records transmitted from vehicles to the central authority.

[0024] In a preferred variant embodiment, each travel record (or data packet processed as a travel record) reflects a portion of a route with a specific route length. In the variant embodiment, this length can be defined by length (e.g., in meters or kilometers). This length can also be defined by time, such that specific time intervals recording the vehicle's journey are interpreted as travel records. This segmentation aims to keep the data packets to be processed for use as travel records within a specific size range. The process of segmenting the actual journey of a vehicle along a specific route (e.g., from Frankfurt to Munich) into individual travel records or data packets for the creation of map data can also be referred to as slicing or segmenting of travel records or data packets. By forming segments of travel records within a specific length range, the use of SLAM methods can be significantly simplified. The formation of segments with suitable characteristics can normalize the data fed into the SLAM algorithm. This makes the SLAM algorithm more efficient.

[0025] Preferably, the driving record segments are formed in such a way that the start and end regions of adjacent segments partially overlap, or where overlapping segment regions are formed where data exists in two adjacent segments. This allows the segments to be reconnected or processed collaboratively in subsequent methods for processing the data within the segments.

[0026] Appropriate and particularly advantageous slicing, or appropriate and particularly advantageous segmentation and / or partitioning of driving records and data packets allows for particularly efficient processing of driving records and data packets to generate maps. In particular, appropriate slicing or appropriate partitioning / segmentation can facilitate the parallel processing of large volumes of driving records. For example, such slicing or partitioning can also be selected differently depending on the circumstances. For example, it can be differentiated based on the type of highway, rural road, or urban road involved.

[0027] In principle, the amount of driving records used for map data is of great interest within a specific range. At the same time, the frequency of data collection in a group-based manner, or the frequency of driving records being performed for specific sections of a route, varies considerably. For example, while many vehicles travel on highways, thus generating many (potential) driving records to generate data packets, the frequency of travel on secondary roads is typically much lower. This means that the amount of data on secondary roads is also much smaller.

[0028] Appropriate slicing or segmentation of driving records allows the driving records or corresponding data packets to be well-suited for creating HD maps. While it may make more sense to create HD maps for primary roads using more data than for secondary roads, appropriate segmentation allows for selection from the total amount of data collected, which is then used to compile HD maps, enabling customized data coverage for HD map compilation.

[0029] As mentioned above, when compiling HD maps from group data, it is crucial to always distinguish between the initial generation of the HD map and its updates (due to the need for periodic route changes). During the initial generation of an HD map, the map itself does not exist; data packets from a large number of driving records are combined to create the initial (initial) HD map. Regular updates are necessary because HD maps typically include numerous features that can change frequently. For example, such features might be related to construction sites. Because of the high level of detail in HD maps, changes are even more frequent. For example, updates may require replacing small portions of the map. In principle, it is desirable for HD map updates to alter the map as little as possible. This can be advantageous, for example, if the map is safety-related and requires backup through review or other measures.

[0030] The two different stages or types of map production (i.e., initial map generation and HD map updating) present different challenges for data processing used with data acquired from population data. These different challenges can be addressed, in particular, through appropriate segmentation / segmentation of data or driving records.

[0031] The method described in this article for segmented driving records or their trajectories comprises two basic steps: First, detecting points with poor GNSS accuracy. In the second step, segments are formed such that neither the starting nor ending region of each segment includes such points.

[0032] The first step, a) is important to detect segments of the journey with poor GNSS measurement data (e.g., in tunnels) because, preferably, it is unnecessary to generate segments whose generation involves no GNSS measurement data or only poor GNSS measurement data. This specifically refers to the start and end sections of segments. Wherever possible, the start and end sections of segments should have good GNSS measurement data. For example, if a segment contains GNSS measurement data only at its start and not at its end, odometry drift will produce potentially invalid results for SLAM optimization. For route description, GNSS measurement data is important in the driving logs associated with a particular segment and in the driving data packets belonging to those logs, used by sensors on the recording vehicle to record data in the data packets, referencing all recorded data in space. In areas where there is no GNSS measurement data, this route determination must be based entirely on odometry data (e.g., determined by vehicle wheel speed sensors and / or steering angle sensors). Such odometry data is often subject to drift. To limit the effects of drift in odometer data, it is highly advantageous to generate segments in such a way that GNSS measurement data available for describing the route is always present at the beginning and end of each segment or driving record, and the route description is provided solely based on odometer data only between the beginning and end of each segment.

[0033] When the method described herein refers to segmenting the trajectory, it means dividing all data collected during the driving record along the trajectory into segments. Within each segment, the data collected during the driving record is used to form individual data packets associated with a portion (segment) of the trajectory, which have a range that provides good further processability for generating and / or updating HD maps.

[0034] This method is particularly advantageous if the minimum length of the segment is defined as greater than 200 meters.

[0035] The minimum length of the segments can also be larger, for example, 500 meters. This minimum length is preferably chosen so that the data storage of the segments is likely to be efficient, and the algorithms used to process the data contained in the segments can work efficiently.

[0036] This minimum length corresponds to the route taken by the recording vehicle during data collection. The segment must be long enough to contain sufficient information. However, the length should not be so long as to require unnecessarily large amounts of data for updating and processing.

[0037] Particularly preferred is that, in step b), when the trajectory has a start point and an end point, the trajectory is continuously segmented into multiple segments from the start point to the end point according to the minimum length. If the point detected in step a) is located in the start region and / or end region, the segment is combined with its adjacent segments in front and / or behind to form a new segment.

[0038] Data packets collected in a group-based manner typically initially include data recorded by the vehicle during its entire journey. For example, a customer vehicle travels from the driver's residence to their workplace. The starting point is the residence, and the destination is the workplace. If the total route length is 10 kilometers, the method described above can be used to initially divide this route into 20 segments, each 500 meters long.

[0039] This document specifically aims to describe a method for providing HD maps using crowd data collected from customer vehicles, comprising the following steps:

[0040] A) Receive data related to multiple driving records of a customer vehicle in the following manner: For each driving record, acquire the trajectory and the corresponding sensor data, which may include GNSS measurement data;

[0041] B) According to the method including steps a) and b), the trajectories obtained in step A) are divided into multiple segments (3);

[0042] C) Assigning segments to predetermined map portions to form multiple segment groups, each segment group corresponding to a map portion, wherein multiple map portions can be generated from multiple segment groups and merged to form a complete positioning map; and

[0043] D) HD maps are provided by the resulting segment groups.

[0044] The purpose of the method according to steps A) to D) is to create a collection of population data with the most uniform data density possible for subsequent HD map compilation and / or updating. The segmentation described in steps a) and b) provides a prerequisite for this.

[0045] Step A) includes receiving data from driving records, preferably acquired by customer vehicles in a group-based manner. This data specifically includes trajectories describing the routes taken by the recorded vehicles during the driving record, as well as other data determined by sensors during the driving record. The trajectories in this data are preferably generated using GNSS measurement data / GNSS measurement signals, and particularly preferably are available in global coordinates. Preferably, the trajectories can be used to assign a global position to all data acquired in the driving record (e.g., data related to objects on the road). However, due to the foregoing reasons and the aforementioned uncertainties, this global position may be subject to error.

[0046] In step B), the received driving records are segmented according to steps a) and b).

[0047] Further, according to step C), the created segments of the driving record are assigned to map portions. These map portions are preferably specified by a fixed-defined grid associated with the area to be mapped. All generated segments are added to this grid. The grid of the map portion is preferably used to subsequently determine the corresponding driving record for a specific map portion.

[0048] Step D) involves the actual generation of the HD map. According to step D), the initial work performed in steps A) to C) means that the HD map can be created piecemeal, and map portions can be selectively updated.

[0049] Associating or grouping segments of driving records based on map portions significantly improves the processability of segments used to generate HD maps. SLAM algorithms for aligning segments can selectively feed segments from individual map portions or, if necessary, segments from adjacent map portions.

[0050] In particular, associating segments with map portions greatly enhances the possibility of parallelizing data processing to compile HD maps. SLAM optimization for aligning driving records can be performed in parallel for different locations defined by the grid of map portions.

[0051] Associating segments of driving records with portions of the map that form a grid also allows for the creation of search structures that can be used to efficiently locate segments of driving records. For example, this structure can be used to determine the data to be updated.

[0052] This method is particularly advantageous when performing analysis between steps A) and D) to verify whether the number of segments in each group exceeds the upper threshold or falls below the lower threshold.

[0053] In this context, it is advantageous to remove redundant segments from a group if the number of segments in the group exceeds the upper limit threshold.

[0054] If the number of segments in a group is below the lower threshold, it is advantageous to insert new segments into that group.

[0055] For example, the upper threshold could be ten segments per group (or per map portion). The lower threshold could be "five". In principle, it is desirable to cover the entire area or road system to be mapped as consistently as possible in the form of segments of driving records. A consistent quality HD map can then be constructed. This can be achieved by setting upper and lower thresholds. Particularly preferably, the system for performing the method can also be configured to request further (or additional) driving records, if necessary (optionally), if there is insufficient data collected in a group-based manner in a particular area to be mapped.

[0056] Additionally, this method is preferred when step D) of providing the location map includes creating an initial location map.

[0057] Additionally, the method is preferred when step D) of providing the location map includes updating the initial location map, wherein only selected portions of the map are updated.

[0058] The method is applicable to both application scenarios and is advantageous, especially due to the grouping of segments by map portion.

[0059] Furthermore, it is advantageous to update the selected map portion (especially according to step C) so that new segments that can be assigned to that map portion are added to the applicable group and existing segments are removed from that group.

[0060] Specifically, for each map segment, there will always be enough segments with specific up-to-date driving records (e.g., no more than one year; for map segments with construction sites, e.g., no more than one week). The effect achieved is that the data storage device for the segments of driving records can always provide driving records with a predetermined up-to-date state.

[0061] Furthermore, it is advantageous to align the segments in each group with each other in step D) to form an aligned pose map.

[0062] Pose maps include records of all vehicle poses across various driving records, allowing data collected from different driving records to be distributed among them.

[0063] Aligning segments of driving records with each other is a crucial step for processing data from different segments together to create or update HD maps. The recommendation here is to first create the segments of the driving records, then store them in an unaligned form (as the raw data of the segments) in a data storage device. This allows access to the raw data of these segments for compilation and / or updating. This prevents information loss due to data alignment from being merged into the data storage, instead allowing the basic raw data of the records to be used when compiling HD maps.

[0064] Additionally, it is advantageous to combine all the segments obtained in step B) into a data structure and store it in a central data memory (the data structure is designed so that new segments can be inserted and stored segments can be deleted).

[0065] A particularly preferred approach is to use a kD-tree to insert the center of the corresponding segment into the kD-tree to perform the allocation in step C).

[0066] A kD-tree, or "k-dimensional tree," is a balanced search tree used for storing spatial data. It provides the ability to efficiently search stored data.

[0067] This forms an efficient data management structure that allows for the advantageous provision of raw data acquired for map creation during data transmission, so that HD maps can be subsequently (at any time) compiled and / or updated from it.

[0068] This specification is also intended for a map building system for producing HD maps, including:

[0069] - A data collection device for collecting data, wherein segments of driving records are formed and collected in the data collection device according to the method based on steps a) and b);

[0070] - A central data storage unit for storing segments of driving records, wherein the central data storage unit is designed to allow new segments to be inserted into it and existing segments to be removed from it; and

[0071] - A data alignment device for generating aligned pose maps based on segments stored in a central data memory.

[0072] It should be noted that the special advantages and configuration features associated with the above methods are also applicable to and can be transferred to map building systems.

[0073] The map building system is preferably operated by a central authority and is used to allow for the continuous provision of the latest HD maps to users.

[0074] Specifically, the map building system is also configured to perform segmented management, storage, and processing of driving records according to steps A) to D) already described. Attached Figure Description

[0075] The present invention and its technical field are explained in detail below with reference to the accompanying drawings. The drawings illustrate preferred exemplary embodiments (the invention is not limited thereto). It should be particularly noted that the dimensions shown in the drawings, especially the proportions shown, are merely illustrative. In the figures:

[0076] Figure 1 The diagram shows a road system including main roads and side streets, with driving records schematically drawn.

[0077] Figure 2a and 2b An example of driving records in the area where the route passes through the tunnel is shown;

[0078] Figure 3a and 3b This illustrates an example of segmenting a vehicle's journey into travel records within a tunnel area;

[0079] Figure 4a and 4b Examples show which segments of driving records are used to update parts of the map; and

[0080] Figure 5 A flowchart and apparatus for creating HD maps using the method described herein are shown. Detailed Implementation

[0081] Figure 1This diagram illustrates a road system comprising a main road 11 and secondary roads 12 branching from it. This is a simplified example of a road system for which a map is to be constructed or for which an HD map is to be created. Regular vehicle traffic occurs on this road system. These vehicles are preferably normal customer vehicles making normal journeys, not primarily used for collecting data for map creation. In practice, during the journey, the journey is collected by sensors, and this journey is used as a driving record 2. Each driving record 2 has a trajectory 1, which describes the path traversed by the corresponding vehicle during the driving record 2. Figure 1 As can be seen, in principle, more trips occurring on main road 11 that can be used as driving records 2 occur than on secondary road 12. This is particularly due to the greater traffic volume on main road 11 compared to secondary roads. It can also be seen that driving records 2 performed by normal customer vehicles 2 are not subject to central planning control. From a data collection perspective, the choice of which route / road to use is more or less random. The natural distribution regularity of trips creates some unfavorable clusters of driving records for HD map compilation. Consequently, conventional algorithms used to process group data collected in this way are fed an unnecessarily large amount of data. This consumes computation time, increases the workload of map production, and does not create any added value. The method proposed in this paper reduces this problem by segmenting driving records according to steps a) and b); and further processing the segmented driving records according to steps A), B), C), and D). In summary, these methods reduce the computational workload required to create, provide, and update HD maps.

[0082] Figure 2a and 2b An example of a driving record 2 showing a portion of the passage through tunnel 13 is shown. Figure 2a and 2b This illustrates that when strictly applying the minimum length 5 to divide the driving record 2 into segments 3, it may result in the start region 6 and / or end region 7 of segment 3 terminating within tunnel 13. GNSS reception is poor within tunnel 13, therefore all points on the trajectory 1 of the driving record within tunnel 13 are points 4 for which precise GNSS positioning is impossible. An example of such a point 4 located within the end region 7 of segment 3 is shown here. This relates to segment 3 terminating within tunnel 13.

[0083] In areas with poor GNSS measurement data, for example, the trajectory 1 of driving record 3 can only be determined by odometer data. Figure 2bThe diagram illustrates (with enhanced illustration) how odometry data drift can have an impact. Segment 3 of travel record 2, or its trajectory 1, may have a heading with a high degree of error, which could be triggered, for example, by odometry data drift. To avoid such effects, it is advantageous that point 4, which has poor GNSS accuracy, is specifically not located within the start region 6 and end region 7 of segment 3. The start region 6 and end region 7 of a segment typically form an area 19 that overlaps with other segments 3. In this case, it is necessary to correlate segment 3 with other segments 3 to perform SLAM algorithms. At this point, high positional accuracy of GNSS determination, unaffected by odometry data drift, is crucial.

[0084] Figure 3a and 3b This illustrates how the longer route along travel record 2 is adjusted to divide travel record 2 into segments 3, to prevent the starting area 6 or ending area 7 of the segment from terminating within tunnel 13. Therefore, Figure 3a and 3b Specifically, the execution of method steps a) and b) is shown, which can prevent the generation of segment 3 with point 4 in the starting region 6 or the ending region 7.

[0085] Figure 3a and Figure 3b All show the route through tunnel 13. Figure 3a and 3b The route shown has a starting point 8 and an ending point 9, and has been subdivided into segments 3, or into segments of driving records 2, to achieve better processability for generating HD maps / map building.

[0086] according to Figure 3a The route is subdivided into segments 3 using only the minimum length 5. It can be seen that in each segment, the starting area 6 and / or the ending area 7 terminate in the tunnel 13, and therefore there are points 4 in the starting area 6 and / or the ending area 7 with poor positional accuracy based on GNSS measurement data.

[0087] according to Figure 3b Segment 3 with a starting region 6 or an ending region 7 has now been merged with other segments 3, so that there are no longer segments 3 with a starting region 6 or an ending region 7 located in tunnel 13 (and therefore in areas with poor positional accuracy based on GNSS measurement data).

[0088] Figure 4a and 4b In particular, this relates to the methods described according to steps A), B), C), and D), wherein the methods described according to steps a) and b) are embedded. Figure 4a and 4bMultiple segments 3 are shown, including data associated with a specific route section 20, upon which HD maps can be compiled and / or updated. As described, segment 3 is a portion of the driving record 2. Each segment 3 has associated data acquired during the driving record 2 traversing that segment 3.

[0089] Figure 4a and 4b A problem has emerged: a specific map portion 10 of an existing, compiled HD map needs to be updated. Figure 4a and 4b The segment 3 shown is stored in a database that provides segments 3 of multiple driving records 2 with related data. These segments 3 are typically collected in a group-based manner. The segments 3 stored in the database are usually provided with timestamps of when they were recorded. It is now preferable to use the most up-to-date segment 3 to update map region 10. Furthermore, it is important to use segments 3 that include information related to the map region 10 to be updated.

[0090] Figure 4a This shows an example of the entire set of segments 3 stored in the database around the map portion 10 to be updated.

[0091] Figure 4b Examples of new segments 21 constituting a subset of the entire set of segment 3 are shown. For these new segments 21, a distinction is made between new segments 21 located entirely outside the map portion 10 to be updated (indicated by dashed lines) and new segments 21 involving the map region 10. New segments 21 shown only by solid lines include additional information applicable to updating the map portion 10. Preferably, only these new segments 21 are used to update the map portion 10.

[0092] Figure 5 The diagram schematically illustrates a method for segmented driving record recording with steps a) and b), and a (higher-level) method for providing an HD map according to steps A) to D). Method steps A) to D) describe the higher-level process for generating the HD map, while method steps a) and b) relate to the execution of method step B), or are sub-steps of method step B).

[0093] Both methods are executed using data processing system 22, which is schematically broken down into several components. This breakdown is not mandatory. The structure of data processing system 22 can also differ. The current illustration is merely an example. The components of the data processing system 22 shown are: data collection device 14, data storage device 18 (which is further divided into data update unit 15 and data reduction unit 16), and data processing device 17.

[0094] Data collection device 14 is initially used to collect raw data that can be used to compile HD maps. The raw data is collected in the field, such as driving records 2 from vehicles in a fleet. During their regular operation, these vehicles preferably record raw data (in a group-based manner), which can then be used with the data processing system described herein to compile HD maps. Therefore, step A) shown is preferably performed outside of the data processing system 22. The collected data is preferably transmitted from the vehicles in the field to the data processing system 22 operated by a central agency via a transmission interface 23 (e.g., a mobile radio connection). The raw data includes initially unsegmented driving records. The raw data typically only reflects data collected along routes frequently used by vehicles in the field. For example, these routes might be routes taken by drivers of vehicles in the field to work or to shops. Method step B), with sub-steps a) and b), is now used to create segments from this raw data suitable for further processing of the raw data to compile and / or update HD maps. Segmentation has been described in detail above, and according to… Figure 5 The segment is associated with the data collection device 14 shown in the figure.

[0095] The data is initially collected independently of the subsequent creation and / or updating of the HD map. Segments of driving records already created using data collection device 14 are stored in data storage device 18 for later use in creating and / or updating the HD map. Data storage device 18 is shown here in a simplified form, having a data update unit 15 and a data reduction unit 16. Data storage device 18 is used only to store and provide data or segments 3 that provide relevant added value for creating and / or updating the HD map. For example, if thousands of segments of driving records on major roads are collected using data collection device 14, it would typically be useless to store all segments in the data storage device for subsequent HD map creation. The information in the individual segments would then be largely redundant. Instead, the task of data storage device 18 is to provide the most up-to-date segments 3, ensuring uniform coverage of roads and routes to be constructed. Data update unit 15 is preferably used to add correspondingly relevant newly added segments. Data reduction unit 16 is preferably used to remove older segments or segments that have been replaced by better, newer segments. Therefore, data storage device 18 is preferably configured to always provide the most up-to-date original dataset, which has the most consistent quality possible across all roads and routes for map construction. This is primarily achieved by performing method step C) (according to...) Figure 5 As shown, the method steps are associated with data storage device 18.

[0096] The appropriate segmentation in steps a) and b) achieves a granularity in the data stored in data storage device 18. This granularity significantly improves the efficiency of data storage device 18. If complete real-world vehicle journeys (e.g., routes taken by vehicle users to work or shops) are stored in data storage device 18, these complete real-world journeys must also be added to or removed. Segmentation means adding only specific, relevant segments from the real-world journeys that have added value for the creation and / or updating of HD maps to the dataset provided by data storage device 18. Since each segment always has overlapping areas with adjacent segments in its start and end regions, these segments can be used to generate complete images of each independent map portion of the system of roads and routes to be mapped at any time.

[0097] The dataset provided in data storage device 18 is accessed by data processing device 17, which is related to the actual creation and provision of HD maps based on the raw data. Data processing device 17 is preferably used to create (new) HD maps. Furthermore, data processing device 17 is also preferably used to update map portions 10 in existing HD maps.

Claims

1. A method for segmenting a trajectory of a driving record, used to compile a high-definition (HD) map based on group data, wherein sensor data is acquired along the trajectory during the driving record, the sensor data including GNSS measurement data, the method comprising the following steps: a) Detect points on the trajectory where it is impossible to determine the location using GNSS measurement data alone, or where the accuracy of the location determined using GNSS measurement data alone does not reach the predetermined positioning accuracy; as well as b) Divide the trajectory into multiple segments in such a way that each segment reaches a predetermined minimum length and at least one point detected in step a) is not located in the start and end regions of each segment.

2. The method according to claim 1, wherein the predetermined minimum length is defined as greater than 200 meters.

3. The method according to claim 1, wherein the trajectory in step b) has a start point and an end point, thereby continuously segmenting the trajectory from the start point to the end point into the plurality of segments according to the predetermined minimum length, and if the point detected in step a) is located in the start region and / or the end region, the segment is combined with the adjacent preceding and / or following segments to form a new segment.

4. A method for providing high-definition (HD) maps from group data collected from customer vehicles, comprising the following steps: A) Using a customer vehicle, data related to multiple driving records is received in the following manner: for each driving record, a trajectory and sensor data corresponding to the trajectory are acquired, the sensor data including GNSS measurement data; B) According to the method described in claim 1, the trajectories obtained in step a) are divided into multiple segments. C) Assign the segments to predetermined map portions to form multiple segment groups, wherein each segment group corresponds to a map portion, wherein multiple map portions can be generated from the multiple segment groups, and the multiple map portions are merged to form a complete positioning map; as well as D) Provide HD maps based on the formed segment groups.

5. The method of claim 4, wherein an analysis is performed between step A) and step D) to determine whether the number of segments in each segment group exceeds an upper threshold or falls below a lower threshold.

6. The method of claim 5, wherein if the number of segments in a segment group exceeds the upper limit threshold, then redundant segments in the segment group are removed.

7. The method of claim 5, wherein if the number of segments in a segment group is less than the lower threshold, a new segment is inserted into the segment group.

8. The method of claim 4, wherein providing the location map in step C) includes creating an initial location map.

9. The method of claim 8, wherein providing the location map in step D) includes updating the initial location map, wherein only selected map portions are updated.

10. The method of claim 9, wherein in step D), the selected map portion is updated such that new segments that can be assigned to the selected map portion are added to the applicable group, and existing segments are removed from the applicable group.

11. The method of claim 4, wherein in step D), the segments in each segment group are aligned with each other to form an aligned pose map.

12. The method of claim 4, wherein all segments obtained in step B) are combined into a data structure and stored in a central data memory, the data structure being designed to allow the insertion of new segments and the deletion of stored segments.

13. The method of claim 4, wherein the allocation in step C) is performed using a kD-tree by inserting the centers of the respective segments into the kD-tree.

14. A map building system for generating high-definition (HD) maps, having - A data collection device for collecting data, wherein segments of driving records are formed and collected in the data collection device according to the method of claim 1; - A central data storage unit for storing segments of the driving record, the central data storage unit being designed to allow new segments to be inserted into the central data storage unit and to allow existing segments to be removed from the central data storage unit; as well as - A data alignment device for generating an aligned pose map based on the segments stored in the central data memory.

Images