Methods for merging map datasets

Abstract

Method for fusing map datasets (1, 2, 3), comprising the steps of: - providing multiple map datasets (1, 2, 3) containing positional information relating to the positions of respective features (6, 7, 9, 15-25), - selecting multiple pairs of features (6, 7, 9, 15-25) in each map dataset (1, 2, 3) and assigning distance information to each pair, which is specified by or dependent on the positional information and relates to the distance between the assigned features (6, 7, 9, 15-25), - identifying mutually associated features (6, 7, 9, 15-18) that describe the same feature (6, 7, 9, 15-18) in the different map datasets (1, 2, 3), - determining coordinates for at least parts of the features (6, 7, 9, 15-25) by Solving an optimization problem that depends on distance information and coordinates, where associated features (6, 7, 9, 15 - 18) are assigned the same coordinates,wherein the position information describes an absolute position of the respective features (6, 7, 9, 15-25) with respect to a map dataset-side coordinate system or a relative position of the respective features (6, 7, 9, 15-25) to each other and / or an expected error of this absolute or relative position, wherein several groups (28, 29) of features (6, 7, 9, 15-25) are selected from the map datasets (1, 2, 3), wherein a measure of similarity is determined for pairs of groups (28, 29) from different map datasets (1, 2, 3) with respect to the relative positions of the respective features (6, 7, 9, 15-25) of the group (28, 29) in order to identify maximally similar groups (28, 29) and thus the associated features (6, 7, 9, 15-18).

Description

[0001] The invention relates to a method for fusing map datasets. The invention also relates to a motor vehicle.

[0002] Map data is used in motor vehicles for route planning, among other things. A corresponding method is known, for example, from German patent application DE 4104351 A1. This method allows information from multiple map sections or information from multiple sources to be combined, as is known, for example, from German patent applications DE 202006021132 U1 and DE 69924772 T2. A problem with combining different map sections or information from different sources is that these maps or sources can exhibit different positional or orientation errors and / or distortions. This is particularly problematic when high-resolution location information, such as lane-specific information about certain features on a planned route, is required, as even minor orientation errors or distortions can lead to significant deviations in the absolute position.For example, when using map data in the context of highly automated driving or for novel assistance systems, a high degree of accuracy with a position error of less than a few tens of centimeters or less than 10 cm should be achieved.

[0003] The invention is therefore based on the objective of providing a method that enables a fusion of map data sets with increased accuracy, in particular to also enable a robust fusion of high-precision map data with, for example, lane-accurate information.

[0004] The problem is solved according to the invention by a method for fusing map data sets, which comprises the following steps: - Providing multiple map datasets that include positional information relating to the positions of respective features, - Selection of multiple pairs of features in each map dataset and assignment of distance information to each pair, which is specified by or depending on the position information and relates to the distance of the assigned features, - Identifying related features that describe the same feature across different map datasets, - Determining coordinates for at least parts of the features by solving an optimization problem that depends on the distance information and the coordinates, assigning the same coordinates to features associated with each other.

[0005] According to the invention, the coordinates of individual features, such as objects described by map data, are determined by solving an optimization problem. For this purpose, as will be explained in more detail later, mutually associated features are first identified that describe the same characteristics in the various map datasets, i.e., that are assigned to the same object or the same segment of an object. Since a feature contained in several of the map datasets should only be represented once in the merged map dataset with a specific position or with specific coordinates, the same coordinates are assigned to mutually associated features as a boundary condition of the optimization problem. In the simplest case, this can be achieved by specifying coordinates for the associated features using one of the map datasets.It is not necessary to identify all features that are associated with each other, i.e., that relate to the same object. For the method to be robust, it is sufficient to identify some associated features that describe the same characteristic in the different map datasets. This can, for example, be enough to detect and correct relative misorientation, shifts, and / or distortions between the map datasets.

[0006] The remaining features can directly form a pair with one of the associated features or be linked to them via a series of pairwise relationships. The coordinates of the remaining features can then be chosen to minimize the difference between a given distance calculated from the coordinates for a pair and the actual distance information for that pair. This approach allows for a fusion of the map datasets with relatively little computational effort. Alternatively, the coordinates for the associated features could be determined using an appropriate optimization procedure.

[0007] The selection of multiple pairs of features can also be viewed as the construction of a graph relating to the respective map dataset. Here, the respective features correspond to nodes in this graph, and the pairs of features, particularly with the associated distance information, form an edge of the graph. The method according to the invention can thus be considered a graph optimization problem.

[0008] In addition to the distance between the two features in the respective map dataset, the distance information can describe an expected error for this distance. Alternatively or additionally, an expected absolute and / or relative distance error can be specified or assumed for a given map dataset. For example, a map dataset may originate from a specific provider or source, and for this provider or source, empirical data regarding the expected accuracy of the map dataset may be known and used to determine an expected distance error.

[0009] Features can describe individual objects and / or landmarks, such as road signs, guideposts, trees, intersections, potholes, construction sites, or similar features. These can be present in multiple map datasets, thus providing associated features. Conversely, certain features or landmarks may only be present in one or some of the map datasets. A map dataset that describes certain features or landmarks, such as guideposts, can be combined with another map dataset that does not describe them. The resulting merged map dataset can describe the features or landmarks of both map datasets, including, for example, guideposts described in only one of the map datasets. The resolution or accuracy of the merged map dataset can be in the centimeter range.

[0010] Another highly relevant application of the method according to the invention is the supplementation of a high-precision base map with location-based, dynamically changing information, for example, lane-specific traffic information, highly accurate (e.g., centimeter-accurate) recorded road friction coefficients, or similar data. A first map dataset can provide static landmarks, objects, or similar features, for example, in the form of a high-precision road map. A second map dataset provides the dynamic information. Optionally, the second map dataset can also describe at least some features, for example, individual landmarks, that can be associated with features of the first map dataset.As part of the fusion of the map data, the coordinates of the features of the first map dataset can be assumed to be fixed, and coordinates can be determined exclusively for the features of the second map dataset, in particular exclusively for the dynamically changing features, in order to supplement the base map with the dynamic information.

[0011] If the map dataset describes elongated objects, such as individual lanes, guardrails or similar features, these can also be described by several separate features, which are arranged, for example, at a predetermined distance along the object, as will be explained in more detail later.

[0012] The position information can describe the absolute position of the respective features in a map dataset-based coordinate system and / or the relative position of the respective features to each other and / or an expected error of this absolute or relative position. A relative position can, for example, be specified as a direction vector between two features, i.e., as a distance and a direction. If absolute positions are specified for the individual features by a map dataset, this is initially done in a map dataset-based coordinate system. While this may approximate a global coordinate system in which the fusion of the various map datasets is to take place, it may be distorted, offset, and / or misoriented relative to it. The error in this registration can be corrected during the fusion process as explained above.Errors can be specified globally for the map dataset, but it is particularly preferred to specify a respective expected error for individual objects or object types.

[0013] As explained earlier, it is essential to assign the same coordinates to associated features that are assigned to the same feature in the different map datasets. To identify such features, several groups of features can be selected from the map datasets. A measure of similarity is then calculated for pairs of groups from different map datasets with respect to the relative position of the respective features within each group. This aims to identify maximally similar groups and thus the associated features. The groups of features can be chosen randomly, with a preference given to determining similarities between pairs of groups with the same number of features.

[0014] Preferably, the groups of features comprise only those features that are present in multiple map datasets, since only these can be associated features and this helps to avoid misidentifications of similar groups of features. Each group of features forms a line segment or polygon in the respective map dataset. In particular, the features can be sorted such that a simple polygon results whose sides do not intersect. A measure of similarity can then be determined for polygons or groups from different map datasets. For this purpose, a cost function can be established, for example, which defines the respective costs for a relative rotation, scaling, and / or compression of the polygons and for the remaining distances between the features after this preprocessing.Polygons for which, for example, a limit value for this cost function is not exceeded, or for which sufficient similarity is otherwise established, are identified with each other, such that the individual features of these groups, or the individual vertices of the respective polygon, form associated features. In other words, groups of features are found whose local arrangement is very similar to each other, leading to the assumption that they are the same features. Additional known properties of the features, such as reflectivity (if the features are sensorially detected) or a specific feature type, can be taken into account.

[0015] For map datasets, it can typically be assumed that the relative position of features that are close to each other has a relatively small error. It is therefore advantageous if at least one of the map datasets describes the distances between the features of that map dataset, selecting pairs of features whose distance falls below a predefined threshold. The map dataset can specify the distances directly, for example, as a direction vector between features. Alternatively, a map-specific coordinate system can be defined for the map dataset, and the features can have coordinates within this coordinate system, thus predefining the distances between the features.If an expected accuracy of the distance information is taken into account for the individual pairs, the expected accuracy can also depend on the distance, since a higher distance accuracy is expected for features that are closer together.

[0016] At least one of the map datasets can relate to at least one extended object, such as a guardrail, a lane, or similar feature, whereby this object is subdivided into segments, particularly overlapping ones, with each of these segments forming one of the features. The features can be arranged at defined intervals along the object. By considering extended objects as multiple features, the number of features available for map fusion can be increased, thereby achieving higher fusion accuracy.

[0017] Features assigned to adjacent segments can be selected as a pair. Due to the procedure described above, the distances between these features are known with high accuracy, which is why these features can be advantageously selected as a pair.

[0018] Fixed coordinates can be assigned to the features of one of the map datasets, and the coordinates are determined for the features of at least one further map dataset. In other words, the fixed coordinate can be used for the associated features that are present in both map datasets. Only for those features for which no fixed coordinate is specified can the coordinate be specified according to the method according to the invention. This variant of the method according to the invention can therefore be carried out with relatively low computational effort and is thus particularly suitable if the fusion of the map datasets is to be carried out, for example, locally by a processing unit of a motor vehicle. For example, a highly accurate navigation map may be available on the vehicle side, which is to be supplemented by externally provided information, for example, the positions of construction sites, traffic accidents, and / or potholes.

[0019] Alternatively, it is also possible to determine the coordinates of all features within the framework of the inventive method. This is particularly advantageous if it can be assumed that all map datasets to be merged contain errors. In this case, the position of the associated features is also determined by optimizing the coordinates, so that a merged map that is at least highly consistent internally can be obtained.

[0020] At least one of the map datasets can be updated at least once, adding a previously missing feature to the map dataset and determining a coordinate for at least this feature. Specifically, a coordinate can be determined exclusively for the added feature(s). This allows time-varying data, such as traffic information or speed profiles of traffic flow, weather conditions (e.g., ice formation on roads), or similar data, to be merged with existing, especially previously merged, map datasets with minimal computational effort.

[0021] At least one of the map datasets can specify the position of a feature with track-level accuracy or within a track. The method according to the invention is particularly advantageous for achieving a highly accurate fusion of map data. For example, a resolution of a few centimeters or several tens of centimeters can be achieved.

[0022] In addition to the method according to the invention, the invention relates to a motor vehicle comprising a processing unit configured to carry out the method according to the invention. Computing power in motor vehicles is typically limited. Therefore, when carrying out the method according to the invention in motor vehicles, it can be advantageous to use one of the map datasets as a reference dataset, the features of which, as explained above, are assigned fixed coordinates. The method according to the invention can thus be used within the motor vehicle, in particular, to add further features to an existing map dataset, which are taken from an externally provided map dataset.

[0023] The motor vehicle may have a communication device to retrieve or receive at least one of the map data sets to be merged from an external device, such as a backend server.

[0024] Further advantages and details of the invention will become apparent from the following exemplary embodiments and the accompanying drawings. These schematically illustrate: Fig. 1 different map data sets which are fused by an embodiment of the method according to the invention, and Fig. 2 an embodiment of a motor vehicle according to the invention.

[0025] Fig. Figure 1 shows three different map datasets 1 , 2 , 3 , which are to be merged for use in a motor vehicle. The map datasets 1 , 2 , 3 Each describes a road network 4 as well as additional features of the road network 4 or objects in the vicinity of the road network 4 For the sake of clarity, in Fig. 1 each only a very short section of the road network 4in the immediate vicinity of an intersection 5 depicted. In real-world applications, the map datasets, 1 , 2 , 3 typically considerably larger sections of the road network 4 This can describe, for example, a route length or a radius of several kilometers. However, it can also involve the fusion or optimization of map data for smaller areas, such as a radius of a few hundred meters, for example, when highly accurate map data needs to be merged to achieve a resolution in the centimeter range or in the range of a few tens of centimeters. To achieve a robust fusion of the map datasets 1 , 2 , 3 To enable this, the map data sets indicate 1 , 2 , 3 They share common features. The junction 5 is characterized by 6 and 7, which are positioned at the edges of the junction, are described. Furthermore, it is described as a characteristic 9 the position of a sign 36 described. The extended object 10 , namely a guardrail, is, as for the map data set 3 It is shown as being divided into several overlapping segments. 11 , 12 , 13 , 14 subdivided, and the respective center of these segments 11 - 14 as a characteristic 15 - 18 saved.

[0026] In addition to these common features, the various map datasets 1 , 2 , 3 The map dataset exhibits the following characteristics: 19 - 21 Positions of potholes 8 on. The map data set 2 Generally describes traffic information. Features include: 22 , 23Schematic speed profiles for individual lanes are shown. Local traffic information is also included, for example as a feature. 24 The presence of a traffic accident was described. The map data set 3 additionally describes weather information, especially as a feature 25 the position of an iced-up area 26 .

[0027] The various map datasets 1 - 3 For example, they may be provided by different vendors or have been collected in different ways. This can lead to discrepancies in the coordinate systems of the individual map datasets. 1 , 2 , 3 differ slightly from each other, in particular exhibit a relative orientation or scaling error and / or are distorted relative to each other. For example, if the features 22 - 25 the map data sets 2 ,3 directly into the map data set 1 If this were adopted, relatively large positioning errors could result, meaning that, for example, a track-accurate assignment of the individual features would no longer be possible.

[0028] To avoid such coordinate errors, the coordinates of the individual features are determined by solving an optimization problem. This takes advantage of the fact that the previously described deviations between the different map datasets, 1 , 2 , 3 While this typically leads to relatively large errors in the relative position of widely spaced features, the effect on the distances between closely spaced features is typically relatively small. Therefore, in the map datasets 1 - 3First, pairs of features were selected that are relatively close to each other or where, for other reasons, it can be assumed that these distances are subject to only small errors. Such a distance 27 is in the example for the map data set 3 for the characteristics 6 and 18 The selection of pairs is illustrated. Two approaches are used for this purpose. Firstly, pairs are formed from features whose distance falls below a predefined threshold. Secondly, as explained previously, extended objects are used. 10 into several segments 11 - 14 subdivided and into segments 11 - 14 a respective characteristic 15 - 18 assigned. Since these features have a defined spatial relationship to each other, adjacent segments are assigned. 11 - 14 Assigned characteristics 15 - 18also selected as a pair. Thus, pairs of the characteristics are formed. 15 and 16 , 16 and 17 as well as 17 and 18 in addition to the pairs chosen based on the distance between their characteristics.

[0029] For each of these pairs, distance information is determined, which indicates the distance. 27 the characteristics of the respective pair in the respective map dataset 1 - 3 The optimization problem then consists of choosing the coordinates of the individual features such that the distances between the coordinates of the features of the pairs deviate as little as possible from the respective distance information.

[0030] As an additional condition within this optimization, it is used that associated features, which are in different map datasets, are 1 - 3The same feature is described, and the same coordinates are assigned to it. In other words, an object or segment mapped by multiple map sets is assigned only one position in the merged map dataset. To achieve this, the associated features are first identified. As explained earlier, it is sufficient to identify parts of the associated features that describe the same feature in the different map datasets. The association therefore does not need to be perfect, and the fusion of the map datasets is still robust even if it is not immediately apparent that individual features refer to the same object or landmark in the different map datasets.

[0031] This can be achieved by including in the individual map data sets 1 - 3 each group 28 , 29Features, especially polygons, are selected. In the example, groups are used. 28 , 29 each 3 Features 6 , 7 , 18 or 9 , 16 , 17 as shown. However, it can be advantageous to group with more than three characteristics, for example with at least 5 or at least 7 features to be used to achieve more robust detection of identical features across different map datasets 1 - 3To achieve this, preferably several groups of features are selected for each map dataset, all containing the same number of features. For pairs of groups from different map datasets, the similarity of these groups can then be determined. From the respective map dataset, the relative position of neighboring points within the group, and thus the shape of the respective polygon, can be determined. By comparing the groups from the different map datasets... 1- 3. Similar groups can be identified. For example, a measure of the similarity between two groups can be determined by establishing a cost function that assigns a cost factor to each rotation, scaling, and / or distortion of the group, as well as to the remaining point distances between the features of the two groups after these operations are applied. This cost function can be minimized, and its minimum is a measure of the dissimilarity between the groups. If a pair of groups is found to be sufficiently similar—for example, if the minimum of an assigned cost function for this pair is less than a predefined limit—it can be assumed that the individual features of the group, or the individual vertices of the respective polygon in the different map datasets, are assigned to the same object or feature. In this way, the features can be 6 , 7 , 9 and15 - 18 The features can be found in all three map datasets, and as a boundary condition of the optimization problem, it can be specified that these features have the same coordinates.

[0032] In the simplest case, one of the map data sets, for example the map data set, can be used. 1 , can be used as a reference map dataset, and the coordinates of the associated features can be predefined by this map dataset. The coordinates for the features from the other map datasets 2 and 3 adopted features 22 - 25 They can then be determined by solving the optimization problem. For each of the features, the values ​​can be calculated. 22 - 25 Distances to all features within a specified radius around the respective feature are taken into account. 22 - 25 lie. For example, for the characteristic 24The coordinate is determined by taking into account the error in the distance to the features. 6 , 7 and 18 The coordinates of the features are minimized. 6 , 7 and 18 as explained by the first map data set 1 specified and the distances between the feature 24 and the characteristics 6 , 7 and 18 through the map data set 2 .

[0033] Are in the immediate vicinity of those features 22 - 25 , for which coordinates are determined, only mutually associated features are present that are present in each of the map datasets, the coordinates of the individual features can be determined. 22 - 25They can be determined independently of each other. The optimization problem thus decomposes into several separate optimization problems. On the other hand, if features with larger radii are considered during pairing, it may be necessary, for example, to determine the distance between the features. 23 and 24 and the distance between the features 22 and 23 to be taken into account within the framework of the optimization problem, whereby the coordinates of the features 22 - 24 can no longer be determined independently of each other.

[0034] The fixed specification of the coordinates for the associated features by the map dataset 1This enables the determination of coordinates for all features with relatively low computational effort, which is particularly advantageous if the process is to be carried out by a mobile device, for example, in a motor vehicle. However, the quality of the merged map dataset can often be further improved if the coordinates of the associated features are also determined or corrected as part of solving the optimization problem. Thus, not only the map datasets are improved. 2 and 3 locally deformed and / or shifted and / or rotated and / or scaled to achieve optimal fusion, but all data sets 1 , 2 and 3 .

[0035] Parts of the map datasets, for example the map dataset relating to traffic flow 2 and the map data set relating to weather conditions 3Features can change on relatively short timescales, for example, less than a day or less than an hour. It can be advantageous in this case not to re-merge the map datasets every time one of them changes, but rather to simply delete features that are no longer present from the merged map data and include newly added features. This can be achieved with minimal computational effort, for example, by assuming the coordinates of the static or unchanged features are fixed and determining only the coordinates of the newly added features by solving the optimization problem.If the newly added features are sufficiently far apart in this case so that they exclusively form pairs with features with fixed coordinates, the optimization problems for determining the coordinates of the individual newly added features can be solved separately, making it possible to determine the coordinates with particularly low computational effort.

[0036] In an embodiment not shown, one of the map data sets could also be used. 1 , 2 , 3 or another map dataset to be merged can describe features such as objects or landmarks, for example, guideposts, that are not present in another map dataset or in the other map datasets. Through the described merging of the map datasets, a map dataset that does not include these objects or landmarks can then be supplemented with them. For example, in the map dataset1 Positions of guideposts are added, which originate from another map dataset, whereby an accuracy of, for example, a few centimeters can be achieved due to the described map fusion.

[0037] Fig. Figure 2 schematically shows a motor vehicle 30 , which is trained to carry out the procedure described above. The map dataset 1 For example, it can be stored internally in the vehicle and used by a navigation system. 31 of the motor vehicle 30 will be provided. The map datasets 2 and 3 can be accessed via the communication device 32 of the motor vehicle 30 from two different server facilities 33 , 34 be received, that is, provided in particular by separate providers. The previously described steps for merging the map data are carried out by the processing facility.35 of the motor vehicle 30 carried out. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] DE 4104351 A1

[0002] DE 202006021132 U1

[0002] DE 69924772 T2

[0002]

Claims

[1] Method for merging map datasets (1, 2, 3), comprising the steps: - Providing multiple map datasets (1, 2, 3) that include position information relating to the positions of respective features (6, 7, 9, 15 - 25), - Selection of several pairs of features (6, 7, 9, 15 - 25) in each map dataset (1, 2, 3) and assignment of distance information to each pair, which is specified by or depending on the position information and concerns the distance of the assigned features (6, 7, 9, 15 - 25), - Identifying related features (6, 7, 9, 15 - 18) that describe the same feature (6, 7, 9, 15 - 18) in the different map datasets (1, 2, 3), - Determining coordinates for at least parts of the features (6, 7, 9, 15 - 25) by solving an optimization problem that depends on the distance information and the coordinates, assigning the same coordinates to mutually associated features (6, 7, 9, 15 - 18). [2] Method according to claim 1, characterized by , that the position information describes an absolute position of the respective features (6, 7, 9, 15 - 25) with respect to a map dataset-side coordinate system or a relative position of the respective features (6, 7, 9, 15 - 25) to each other and / or an expected error of this absolute or relative position. [3] Method according to claim 2, characterized by, that from the map datasets (1, 2, 3) several groups (28, 29) of features (6, 7, 9, 15 - 25) are selected, whereby a measure of similarity for pairs of groups (28, 29) from different map datasets (1, 2, 3) is determined with respect to the relative positions of the respective features (6, 7, 9, 15 - 25) of the group (28, 29) in order to determine maximally similar groups (28, 29) and thus the associated features (6, 7, 9, 15 - 18). [4] Method according to any of the preceding claims, characterized by , that at least one of the map datasets (1, 2, 3) describes distances of the features (6, 7, 9, 15 - 25) of the respective map dataset (1, 2, 3), wherein pairs of features (6, 7, 9, 15 - 25) are selected whose distance is below a specified limit. [5] Method according to any of the preceding claims, characterized by, that at least one of the map data sets (1, 2, 3) relates to at least one extended object (10), wherein this object (10) is subdivided into, in particular overlapping, segments (11 - 14), each of these segments (11 - 14) forming one of the features (15 - 18). [6] Method according to claim 5, characterized by , that features (15 - 18) assigned to adjacent segments (11 - 14) are selected as a pair. [7] Method according to any of the preceding claims, characterized by , that the features (6, 7, 9, 15 - 25) of one of the map datasets (1, 2, 3) are assigned fixed coordinates, wherein the coordinates are determined for the features (6, 7, 9, 15 - 25) of at least one further of the map datasets (1, 2, 3). [8] Method according to any of the preceding claims, characterized by, that at least one of the map data sets (1, 2, 3) is updated at least once, adding a previously non-existent feature (6, 7, 9, 15 - 25) to the map data set (1, 2, 3), and determining a coordinate for at least this feature (6, 7, 9, 15 - 25). [9] Method according to any of the preceding claims, characterized by , that at least one of the map data sets (1, 2, 3) specifies a position of a feature (6, 7, 9, 15 - 25) with lane accuracy or within a lane. [10] motor vehicle, characterized by that it comprises a processing facility (35) equipped to carry out the process according to any of the preceding claims.

Images