Method for generating an electronic road map, electronic road map, method for generating a training data set, training data set and related devices

Abstract

The invention relates to a method for generating an electronic road map with classification information of road nodes, comprising: receiving position information of a spatial region of an electronic road map comprising road nodes; receiving position data of a plurality of vehicles driving through the road nodes, wherein the position data map the position courses of the vehicles when driving through the road nodes; classifying the road nodes by a classifier module based on the position information and based on the position data, wherein the classification comprises: recognizing the road nodes as roundabouts if, among a plurality of position courses, at least a predefined number of position courses each have a predefined curve course; adding the classification information of the road nodes to the electronic road map, and generating the electronic road map with classification information. Furthermore, the invention also relates to an electronic road map and a method for generating a training data set and a corresponding training data set. Furthermore, the invention also relates to a corresponding computing unit and computer program product.

Description

Technical Field

[0001] This invention relates to a method for generating an electronic road map with classification information of road nodes, and an electronic road map. The invention also relates to a method for generating a training dataset for training a classifier module for classifying road nodes, and a corresponding training dataset. Furthermore, the invention relates to corresponding computing units and computer program products. Background Technology

[0002] Electronic maps are known from existing technology. For the purpose of using road maps to assist vehicle control, it is of interest to find maps that can distinguish different road nodes from each other. Summary of the Invention

[0003] The objective of this invention is to provide an improved method for generating electronic road maps with classification information of road nodes, an electronic road map, a method for generating a training dataset for training a classifier module for classifying road nodes, and a training dataset.

[0004] This task is solved using the method, road map, and training dataset according to the invention. Advantageous implementations are the subject of extended technical solutions.

[0005] According to one aspect, a method is provided for generating an electronic road map with classification information of road nodes, the method comprising: Receive location information of spatial areas, including road nodes, from electronic road maps; Receive location data of multiple vehicles that have passed through the road node, wherein the location data maps the position and direction of these vehicles as they pass through the road node; The classifier module classifies the road nodes based on location information of the spatial region including the road nodes and location data of vehicles passing through the road nodes, wherein the classification includes: If, among multiple location directions, at least a predetermined number of location directions have a predetermined curved direction, then the road node is identified as a ring road. The classification information of the road nodes is added to the electronic road map, and an electronic road map with classification information is generated.

[0006] This provides the following technical advantages: an improved method for generating electronic road maps with classification information of road nodes can be provided. Specifically, in existing electronic road maps that do not include the classification of road nodes listed therein, location information of the spatial areas in which road nodes are displayed in the electronic road map is provided.

[0007] In addition, location data for multiple vehicles that passed through road nodes displayed on the electronic map at earlier points in time are provided. Here, the location data describes the trajectory of these vehicles as they passed through the road nodes.

[0008] Subsequently, based on location information from the electronic road map and location data of multiple vehicles, the corresponding road nodes are classified by a trained classifier module.

[0009] Here, if a predetermined number of vehicle trajectories have a predetermined curved trajectory among the multiple trajectories, the corresponding road node is identified as a circular traffic flow.

[0010] Subsequently, the corresponding classification information for road nodes categorized as ring road is entered into the electronic road map, thereby generating an electronic road map with classification information. The resulting electronic road map with classification information is an improved road map because the added road node classification information is important supplementary information in assisting vehicle control.

[0011] The driving behavior of other traffic participants can be inferred from the construction of the road nodes to be traversed. This enables improved vehicle control when considering the resulting electronic road map with categorized information about the road nodes.

[0012] According to one implementation, the classification includes: If, among multiple location orientations, the number of location orientations having the predefined curve orientation is less than the predefined number, then the road node is identified as one of the following: road intersection, fork entrance, entrance to a rural road, federal road, or highway, exit to a rural road, federal road, or highway, or highway intersection.

[0013] This provides the following technical advantages: in addition to ring road traffic, the classifier module can also classify other types of road nodes, such as road intersections or entrances and exits, and the corresponding classification information can be incorporated into the electronic road map.

[0014] By considering the predefined curve direction, it is possible to make an accurate distinction between roundabouts and nodes such as those constructed as road intersections, based on the position and direction of vehicles as they travel through each road node.

[0015] According to one embodiment, the predetermined curve orientation has one of the following orientations: quarter-circular orientation, semi-circular orientation, three-quarter-circular orientation, and circular orientation.

[0016] This provides the following technical advantages: by constructing the predefined curve direction as a quarter-circular, semi-circular, three-quarter-circular, or circular direction, a clear distinction can be made between road nodes constructed as circular traffic and road nodes constructed as road intersections, entrances, or exits.

[0017] According to one embodiment, the predetermined curve direction encloses a predetermined angle range, wherein the predetermined angle range is an angle selected from the following list: at least 90 degrees, at least 180 degrees, at least 270 degrees, and at least 360 degrees.

[0018] This provides the following technical advantages: by considering the predefined angle range of the curve direction, road nodes can be clearly and accurately classified again. In particular, it can accurately distinguish between roundabouts and road nodes, such as those constructed as road intersections.

[0019] According to one embodiment, the predetermined curve direction has a predetermined radius of curvature.

[0020] This provides the following technical advantages: by considering the curvature radius of the predefined curve direction, it is possible to clearly classify road nodes again and make a clear distinction between road nodes in a roundabout and those constructed as road intersections, for example.

[0021] According to one implementation, the classification further includes: Based on the location information of the spatial region of the road node and the location data of the vehicle, an image display of the road node is generated, the image display showing the direction of the vehicle extending from the road node, wherein the identification includes: Identify the predefined curve direction in the vehicle's positional trajectory.

[0022] This yields the following technical advantages: the classifier module can accurately classify road nodes. This results in the generation of image displays of road nodes on the road map, as well as image displays of the positions and directions of vehicles that passed through the corresponding road nodes at earlier points in time.

[0023] Based on the corresponding image display, a classifier module can classify road nodes by recognizing the curve direction of the vehicle's position and direction as it passes through the road node. The recognition of predefined curve directions within the vehicle's position and direction in the image display can be achieved by the classifier module, for example, through object recognition.

[0024] According to one embodiment, the generation of the image display includes: By selecting the location information in the spatial region that is consistent with the location data of the vehicle's direction, the mapping of the vehicle's corresponding location data to the location information of the spatial region of the road node is implemented. A vector display of the selected location information of the spatial region of the road node is generated; By mapping the selected spatial region of the road node, representing its orientation, onto a predefined image plane as continuous line elements, the vector display is rasterized. The identification of the curve orientation includes: The image display is converted into a tensor display.

[0025] This provides the following technical advantages: it enables further improvement in the classification of road nodes by the classifier module.

[0026] Here, the image display is obtained as follows: Location information is extracted from the spatial region of the corresponding road node in the electronic road map. This location information corresponds to the location data showing the movement of vehicles passing through that road node. Subsequently, the selected location information is displayed as continuous line elements on a predefined image plane.

[0027] Therefore, the generated image display primarily includes continuous line elements showing the positions of vehicles passing through the road nodes, facilitating the classifier module's identification of curves within these positions. To enable the classifier module to identify curves, a conversion from image display to tensor display is also performed, ultimately using these tensor displays for object recognition to identify the curves.

[0028] This enables the identification of curve directions as accurately as possible, and based on this, the corresponding road nodes can be classified accordingly.

[0029] According to one embodiment, the tensor display is constructed as a grayscale tensor display.

[0030] This yields the following technical advantages: by constructing the tensor display as a grayscale tensor display, the object recognition can be presented in the simplest possible way. This further improves the accuracy of object recognition.

[0031] According to one embodiment, the spatial region of the road node in the electronic road map is constructed as a polygon, and the spatial border of the road boundary of the roads intersecting at the road node is displayed.

[0032] This provides the following technical advantages: the spatial area of ​​the electronic road map, including the road nodes, can provide an accurate mapping of the road nodes, especially an accurate mapping of the relevant features of the road nodes.

[0033] According to one implementation, the classifier module includes at least one convolutional network that has been trained accordingly.

[0034] This provides the following technical advantages: through a properly trained convolutional network, it is possible to perform accurate object recognition based on the image display of the vehicle's position and direction as it passes through the road nodes, and to accurately identify the curve direction of the position and direction based on this image.

[0035] According to one embodiment, the location data of the vehicle is GPS tracking data.

[0036] This provides the following technical advantages: by using the vehicle's GPS tracking data, accurate location data can be provided, thereby enabling the precise location and direction of the vehicle as it passes through the road nodes.

[0037] According to one aspect, an electronic road map with classification information of road nodes is provided, wherein the road map is generated according to the method for generating an electronic road map with classification information of road nodes according to any of the above embodiments.

[0038] According to one aspect, a method is provided for generating a training dataset for training a classifier module for classifying road nodes, in order to generate an electronically generated road map with classification information according to any of the above embodiments, the method comprising: Receive location information of spatial areas, including road nodes, from electronic road maps; Receive location data of multiple vehicles that have passed through the road node, wherein the location data maps the position and direction of these vehicles as they pass through the road node; By selecting the location information in the spatial region that matches the location data of the vehicle's corresponding location direction, the mapping of the vehicle's location direction location data to the location information of the spatial region of the road node is implemented. A vector display of the selected location information of the spatial region of the road node is generated; The vector display is rasterized by mapping the selected location information of the spatial region of the road node, which shows the direction of the location, onto a predefined image plane as continuous line elements. Generating tensor data based on image display; and The tensor data are aggregated into the training dataset.

[0039] This provides the following technical advantages: it offers an improved method for generating training datasets.

[0040] According to one aspect, a training dataset is provided for training a classifier module for classifying road nodes, wherein the training dataset is generated according to the method for generating the training dataset.

[0041] This provides the following technical advantages: an improved training dataset can be provided, which enables improved training of the classifier module used to classify road nodes.

[0042] According to one aspect, a computing unit is provided, the computing unit being configured to implement the method for generating an electronic road map according to any of the above embodiments and / or to implement the method for generating a training dataset for training a classifier module for classifying road nodes.

[0043] According to one aspect, a computer program product is provided, comprising instructions that, when implemented by a data processing unit, cause the data processing unit to implement the method for generating an electronic road map according to any of the above embodiments and / or implement the method for generating a training dataset for training a classifier module for classifying road nodes. Attached Figure Description

[0044] Embodiments of the invention are described below with reference to the accompanying drawings. The drawings show: Figure 1 A schematic diagram of the method steps for generating an electronic road map with classification information of road nodes according to one embodiment; Figure 2 Another schematic diagram of the method steps for generating an electronic road map with classification information of road nodes according to another embodiment; Figure 3 A schematic diagram of method steps for generating a training dataset according to one embodiment, wherein the training dataset is used to train a classifier module for classifying road nodes; Figure 4 A flowchart of a method for generating an electronic road map with classification information of road nodes according to one embodiment; Figure 5 Another flowchart of a method for generating an electronic road map with classification information of road nodes according to another embodiment; Figure 6 Another flowchart of a method for generating an electronic road map with classification information of road nodes according to another embodiment; Figure 7 According to another flowchart of a method for generating a training dataset according to one embodiment, the training dataset is used to train a classifier module for classifying road nodes; and Figure 8 A schematic diagram of a computer program product. Detailed Implementation

[0045] Figure 1 A schematic diagram illustrating the method steps of a method 100 for generating an electronic road map 300 with classification information 301 including road nodes 303, according to one embodiment.

[0046] exist Figure 1 Figures a) and b) illustrate two different examples of road nodes 303, showing different steps of a method 100 for generating an electronic road map 300 with classification information 301 containing road nodes 303 according to the present invention.

[0047] Here, Figure a) shows road node 303 configured as a road intersection 319. Conversely, Figure b) shows road node 303 configured as a roundabout 315.

[0048] In order to classify the road nodes 303 of the electronic road map 302, the location information 305 of the spatial region 307 of the road nodes 303 of the electronic road map 302 is first received.

[0049] In the illustrated embodiment, the spatial region 307 is shown as a polygon that maps the extent of the road boundary 331 of the road 333 of the road node 303. The shape of the spatial region 307 may vary depending on the configuration of the corresponding road node 303, as shown in the two illustrated map displays 302 in Figures a) and b).

[0050] In addition, location data 309 of the location direction 311 of a vehicle that traveled through the corresponding road node 303 at an earlier time is received.

[0051] Here, the location data 309 may be, for example, GPS tracking data of the vehicle. More specifically, the location data 309 may be convoy data of multiple vehicles that have passed through the corresponding road node 303 at different times.

[0052] Based on the location information 305 of the map display 302 and the location data 309 of multiple vehicles, the classifier module 313 implemented on the shown computing unit 335 performs the classification of road nodes 303.

[0053] Therefore, if a predetermined number of location directions 311 of multiple vehicles have a predetermined curve direction 317, the classifier module 313 will identify the corresponding road node 303 as a ring road 315.

[0054] Here, the predetermined curve direction can be defined, for example, by the shape of the curve direction. Therefore, the predetermined curve direction 317 can be defined, for example, at least as a quarter-circular direction, a semi-circular direction, a three-quarter-circular direction, or a full-circular direction.

[0055] Alternatively or additionally, the predefined curve direction 317 can be defined by a predefined angle range enclosed by the corresponding curve area: the predefined curve direction 317 must be displayed in the area of ​​road node 303 in the vehicle's position direction 311 so as to classify the corresponding road node 303 as roundabout traffic 315.

[0056] The predefined curve direction 317 may include, for example, at least a 90-degree angle, at least a 180-degree angle, at least a 270-degree angle, or at least a 360-degree angle.

[0057] Alternatively or additionally, the predetermined curve direction 317 may be defined by a predetermined radius of curvature.

[0058] The predefined curve direction 317 is defined by a corresponding definition standard. Based on the position direction 311 of the vehicles that have passed the corresponding road node 303, the corresponding road node 303 can be identified as a circular traffic 315.

[0059] As shown in position 311 in Figure b), in the area of ​​road node 303, due to the ring structure of the ring traffic 315, there are multiple position 311 locations with similarly nearly ring-shaped curved directions 317.

[0060] Other road nodes 303, such as road intersections 319 shown in Figure a), do not have such circular positional orientations 311 for vehicles traveling through the road node 303 due to the corresponding road orientation within the road node 303.

[0061] Therefore, based on the circular orientation, such as a quarter-circle, semi-circle, three-quarter-circle, or full-circle, and / or based on the enclosed angular range, such as at least 90 degrees, at least 180 degrees, at least 270 degrees, or at least 360 degrees, and / or based on the respective radii of curvature of the predefined curve orientation 317, the corresponding road nodes 303 can be classified accordingly based on the more detailed location orientation 311 of the vehicle.

[0062] After the classifier module classifies the road node 303 to be classified, it provides corresponding classification information 301. In this classification information, the corresponding road node 303 is classified as a roundabout 315 or, for example, as a road intersection 319. This classification information is entered into the road map 302 as additional information. Based on this, a corresponding road map 302 is generated, containing the classification information 301 for the road node 303.

[0063] Figure 2 Another schematic diagram shows the method steps of a method 100 for generating an electronic road map 300 with classification information 301 containing road nodes 303, according to another embodiment.

[0064] Figure 2 The implementation method in is based on Figure 1 The implementation methods and included Figure 1 The method steps described herein. For simplicity, this description only applies to... Figure 1 The example in Figure b) illustrates this embodiment, in which road node 303 is constructed as a ring road 315.

[0065] First, feature extraction is performed based on the location information 305 of the road map 302 and the location data 309 of the vehicles.

[0066] Based on this, a vector display 323 is generated showing the location information 305 of map display 302 and the location data 309 of vehicle.

[0067] Based on this, an image display of road nodes is generated by rasterizing the vector display 321.

[0068] In order to generate the image display 321, firstly, the mapping of the vehicle's location data 309 to the location information 305 of the spatial region 307 of the road node 303 on the road map 302 is performed.

[0069] Here, the location information corresponding to the location data 309 of the road map 302 in the area of ​​road node 303 and the location direction 311 of the vehicle when it passes through the corresponding road node 303 is obtained.

[0070] Therefore, the selected location information corresponds to the mapping of the vehicle's location data 309 to the spatial region 307 of the road node 303 on the road map 302.

[0071] In order to generate image display 321, the position information 305 corresponding to the position data 309 of the map display 302 and the vehicle's position direction 311 is selected as a continuous line element 325 and mapped onto a predefined image plane 327.

[0072] Therefore, the resulting image display 321 includes a predefined image plane 327, which mainly or only shows the positional direction 311 of the vehicle as it travels through the road node 303, displayed as a continuous line element 325.

[0073] like Figure 2 As shown, the corresponding continuous line element 325 of the position direction 311 has a predefined curved direction 317 arranged in a near-circular shape around the center 337 of the displayed circular traffic 315.

[0074] In order to implement object recognition for identifying a predefined curve path 317 from the image display 321, in the illustrated embodiment, a tensor display 329 is generated from the correspondingly generated image display 321.

[0075] According to one implementation, a tensor display can be constructed as a grayscale tensor display.

[0076] By implementing classifier module 313 on tensor display 329, predefined curve directions 317 can be identified from continuous line elements 325 generated based on vehicle location direction 311, and road nodes 303 can be classified based on this.

[0077] according to Figure 1 The implementation method described in the text then generates corresponding classification information 301.

[0078] According to one embodiment, the classifier module 313 includes a convolutional network trained accordingly, the convolutional network being configured to cause the identification of a predefined curve direction 317 based on the described image display 321 and to classify the corresponding road node 303 as a roundabout 315 or selected from the following list: road intersection 319, fork entrance, entrance to a rural road, federal road or highway, exit of a rural road, federal road or highway, highway intersection.

[0079] According to one implementation, the convolutional network includes a VGG architecture with four convolutional blocks and one decoder block. These convolutional blocks are sequential layers that respectively include convolutional layers with ReLU activation, max-pooling layers, and dropout layers. In the dropout layers, for example, the probability can be set to 0.2.

[0080] In the final convolutional block, the probability can be set to 0.3. In the decoder block, the probability can be set to 0.5.

[0081] Here, the kernel size of the convolutional layer can be set to 3, and the kernel size of the max pooling layer can be set to 2, so that the kernel has the same dimension as the feature map.

[0082] The decoder portion of the network flattens the output of the feature extractor and includes two fully connected layers with ReLU activation and a dropout layer for regularization arranged between these fully connected layers. Furthermore, a Sigmoid activation function and thresholding are implemented.

[0083] The convolutional network can be constructed accordingly by using validation and by using early interruption when the validation loss is observed. Here, the loss function used can be constructed as binary cross-entropy. The optimizer used can be, for example, the Adam optimizer with a learning rate of 0.0005.

[0084] For example, a convolutional network can be constructed accordingly when implementing 100 cycles. The training dataset 400 used accordingly can, for example, include 1125 image displays 321, of which these image displays 321 show 334 road intersections 319 and 582 roundabouts 315.

[0085] Figure 3 A schematic diagram of method steps for generating a training dataset 400 according to one embodiment is shown, the training dataset being used to train a classifier module 313 for classifying road nodes 303.

[0086] In the illustrated embodiment, the method 200 shown for generating a training dataset 400 for training the corresponding classifier module 313 for classifying road nodes 303 is based on... Figure 2 The steps of the method 100 for generating a road map 300 with classification information 301 in the implementation.

[0087] To generate the training dataset 400, the following method steps were implemented until tensor representation 329 was generated. These method steps have been referenced. Figure 2 A detailed description was provided.

[0088] To generate the training dataset 400, tensor displays 329 generated for different road nodes 303 are aggregated into a corresponding training dataset 400. For this purpose, as is common practice in the prior art, the tensor displays 329 can be grouped for training and testing the classifier module 313.

[0089] Figure 4A flowchart is shown of a method 100 for generating an electronic road map 300 with classification information 301 including road nodes 303, according to one embodiment.

[0090] In order to generate an electronic road map 300 with classification information 301 containing road nodes 303, in the first method step 101, the location information of the spatial region 307 including road nodes 303 of the electronic road map 302 is first received.

[0091] In another method step 103, location data 309 of multiple vehicles passing through the road node 303 is received. Here, the location data 309 describes the positional trajectory 311 of these vehicles as they pass through the road node 303.

[0092] In another method step 105, the classifier module 313 classifies the road node 303 based on the location information 305 and the location data 309.

[0093] Therefore, in method step 107, if at least a predetermined number of location paths 311 of multiple vehicles have predetermined curve paths 317, then the road node 303 is identified as a ring road 315.

[0094] In method step 111, if the number of location paths 311 with a predefined curve path 317 is less than the predefined number among multiple location paths 311, then the road node 303 is identified as one of the following: road intersection, fork entrance, entrance to rural road, federal road or highway, exit of rural road, federal road or highway, highway intersection.

[0095] In another method step 109, the classification information 301 of the road node 303 is added to the electronic road map 302, thereby generating an electronic road map 302 with classification information 301.

[0096] Figure 5 Another flowchart of a method 100 for generating an electronic road map 300 with classification information 301 containing road nodes 303, according to another embodiment, is shown.

[0097] Figure 5 The implementation method in is based on Figure 4 The implementation methods described herein include all those in [the document / concept]. Figure 4 The methods and steps described in the document.

[0098] In the illustrated embodiment, the classification of road nodes 303 includes method step 113. In method step 113, an image display 321 of the road node 303 is generated based on the location information 305 of the spatial region 307 of the road node 303 and the vehicle location data 309. The image display includes the positional orientation 311 of these vehicles extending from the road node.

[0099] Furthermore, method steps 107 and 111 include method step 115. In method step 115, the predefined curve direction 317 is identified in the vehicle's positional direction 311 based on the image display 321.

[0100] Figure 6 Another flowchart of a method 100 for generating an electronic road map 300 with classification information 301 containing road nodes 303, according to another embodiment, is shown.

[0101] Figure 6 The implementation method in is based on Figure 5 The implementation methods described herein include all those in [the document / concept]. Figure 5 The methods and steps described in the document.

[0102] In the illustrated embodiment, the generation 113 of image display 321 includes method step 117. In method step 117, the location data 309 of the vehicle's location direction 311 is mapped onto the location information 305 of the spatial region 307 of the road node 303.

[0103] Therefore, position information 305 that is consistent with the position data 309 of the various positions and directions of the vehicle in the spatial region 307 is selected.

[0104] In method step 119, a vector display 323 of the selected location information 305 of the spatial region 307 of the road node 303 is generated.

[0105] In another method step 121, the vector display 323 is rasterized. To do this, the position information 305 of the selected display vehicle's position direction 311 in the spatial region 307 of the road node 303 is mapped as continuous line elements 325 onto a predefined image plane 327.

[0106] Furthermore, method step 115 in the illustrated embodiment includes method step 123. In method step 123, the image display 321 is converted into a tensor display 329.

[0107] Figure 7Another flowchart of a method 200 for generating a training dataset 400 according to one embodiment is shown, the training dataset being used to train a classifier module 313 for classifying road nodes 303.

[0108] In order to generate a training dataset 400 for training a classifier module 313 that classifies road nodes 303, in method step 201, the location information of the spatial region 307 of the electronic road map 302, which includes road nodes 303, is first received.

[0109] In method step 203, the location data 309 of multiple vehicles that have passed through the road node 303 are received.

[0110] In method step 205, the location data 309 of the vehicle's location direction 311 is mapped to the location information 305 of the spatial region 307 of the road node 303.

[0111] Therefore, position information 305 that is consistent with the position data 309 of the various positions and directions of the vehicle in the spatial region 307 is selected.

[0112] In method step 207, a vector display 323 of the selected location information 305 is generated.

[0113] In method step 209, the vector display 323 is rasterized by mapping the position information 305 of the selected display position direction 311 of the spatial region 307 as a continuous line element onto a predefined image plane.

[0114] In method step 211, a tensor display 329 is generated accordingly from the image display 321 generated by rasterizing the vector display 323.

[0115] In method step 213, the tensor representation is summarized into training dataset 400.

[0116] Figure 8 A schematic diagram of a computer program product 500 is shown. The computer program product includes instructions that, when implemented by a data processing unit, cause the data processing unit to implement a method 100 for generating an electronic road map 300 with classification information 301 of road nodes 303 and / or a method 200 for generating a training dataset 400 for training a classifier module 313 for classifying road nodes 303.

[0117] In the illustrated embodiment, the computer program product 500 is stored on a storage medium 501. Here, the storage medium 501 can be any storage medium known from the prior art.

Claims

1. A method (100) for generating an electronic road map (300) with classification information (301) of road nodes (303), the method comprising: Receive (101) the location information (305) of the spatial area (307) including road nodes (303) of the electronic road map (302); Receive (103) the location data (309) of multiple vehicles that have passed through the road node (303), wherein the location data (309) maps the position direction (311) of these vehicles when they pass through the road node (303). The classifier module (313) classifies the road node (303) based on the location information (305) of the spatial region (307) including the road node (303) and the location data (309) of vehicles passing through the road node (303), wherein the classification (105) includes: If at least a predetermined number of location directions (311) in a plurality of location directions (311) have a predetermined curve direction (317), then the road node (303) is identified (107) as a ring road (315). The classification information (301) of the road node (303) is added (109) to the electronic road map (302), and an electronic road map (300) with classification information (301) is generated.

2. The method (100) according to claim 1, wherein, The classification (105) includes: If, among multiple location paths (311), the number of location paths (311) having the predefined curve path (317) is less than the predefined number, then the road node (303) is identified (111) as one of the following: road intersection (319), fork entrance, entrance to rural road, federal road or highway, exit of rural road, federal road or highway, highway intersection.

3. The method (100) according to claim 1 or 2, wherein, The predefined curve orientation (317) has one of the following orientations: quarter-circular orientation, semi-circular orientation, three-quarter-circular orientation, and circular orientation.

4. The method (100) according to any one of the preceding claims, wherein, The predetermined curve direction (317) encloses a predetermined angle range, wherein the predetermined angle range is an angle selected from the following list: at least 90 degrees, at least 180 degrees, at least 270 degrees, at least 360 degrees.

5. The method (100) according to any one of the preceding claims, wherein, The predetermined curve direction (317) has a predetermined radius of curvature.

6. The method (100) according to any one of the preceding claims, wherein, The classification (105) also includes: Based on the location information (305) of the spatial region (307) of the road node (303) and the location data (309) of the vehicle, an image display (321) of the road node (303) is generated (113), the image display showing the vehicle's position and direction extending from the road node (311), wherein the identification (107, 111) includes: Identify (115) the predefined curve direction (317) in the vehicle's positional direction (311).

7. The method (100) according to claim 6, wherein, The generation (113) of the image display (321) includes: By selecting the location information (305) of the spatial region (307) that is consistent with the location data (309) of the vehicle's position direction (311), the mapping (117) of the location data (309) of the vehicle's corresponding position direction (311) to the location information (305) of the spatial region (307) of the road node (303) is implemented. A vector display (323) of the selected location information (305) of the spatial region (307) of the road node (303) is generated (119). By mapping the selected location information (305) of the spatial region (307) of the road node (303) showing the location direction (311) as a continuous line element (325) onto a predefined image plane (327), the vector display (323) is rasterized (121), wherein the identification (115) of the curve direction (317) includes: The image display (321) is converted (123) into a tensor display (329).

8. The method (100) according to claim 7, wherein, The tensor display (329) is constructed as a grayscale tensor display (329).

9. The method (100) according to any one of the preceding claims, wherein, The spatial region (307) of the road node (303) of the electronic road map (300) is constructed as a polygon, and the spatial border of the road boundary (331) of the roads (333) that intersect at the road node (303) is displayed.

10. The method (100) according to any one of the preceding claims, wherein, The classifier module (313) includes at least one convolutional network that has been trained accordingly.

11. The method (100) according to any one of the preceding claims, wherein, The location data (309) of the vehicle is GPS tracking data.

12. An electronic road map (300) with classification information (301) including road nodes (303), wherein, The road map (300) is generated according to the method (100) for generating an electronic road map (300) according to any one of claims 1 to 11.

13. A method (200) for generating a training dataset for training a classifier module (313) for classifying road nodes (303) to generate an electronic road map (300) with classification information (301) according to any one of claims 1 to 11, the method comprising: Receive (201) the location information (305) of the spatial area (307) including road nodes (303) of the electronic road map (302); Receive (203) the location data (309) of multiple vehicles that have passed through the road node (303), wherein the location data (309) maps the position direction (311) of these vehicles when they pass through the road node (303). By selecting the location information (305) of the spatial region (307) that is consistent with the location data (309) of the vehicle's position direction (311), the mapping (205) of the location data (309) of the vehicle's corresponding position direction (311) to the location information (305) of the spatial region (307) of the road node (303) is implemented. A vector display (323) of the selected location information (305) of the spatial region (307) of the road node (303) is generated (207); The vector display (323) is rasterized (209) by mapping the selected location information (305) of the spatial region (307) of the road node (303) showing the location direction (311) as a continuous line element onto a predefined image plane. Based on the image display (321), generate (211) a tensor display (329); and The tensor display (329) is summarized (213) into the training dataset (400).

14. A training dataset (400) for training a classifier module (313) for classifying road nodes (303), wherein, The training dataset (400) is generated according to the method (200) for generating the training dataset (400) according to claim 13.

15. A computing unit (335) configured to implement a method (100) for generating an electronic road map (300) according to any one of claims 1 to 11 and / or implement a method (200) for generating a training dataset (400) according to claim 13, the training dataset being used to train a classifier module (313) for classifying road nodes (303).

16. A computer program product (500) comprising instructions that, when implemented by a data processing unit, cause the data processing unit to implement a method (100) for generating an electronic road map (300) according to any one of claims 1 to 11 and / or implement a method (200) for generating a training dataset (400) according to claim 13, the training dataset being used to train a classifier module (313) for classifying road nodes (303).

Images