Assignment of regulatory signs to a lane

Abstract

The present invention relates to a method, a computer program with instructions, and a device for training a neural network for assigning a regulatory sign to a lane, as well as a neural network trained in this way. In a first step, input variables are provided for the neural network (10). The input variables comprise at least a representation of lanes and regulatory signs, as well as prior knowledge. The neural network is then trained with the input variables (11). Training is terminated (12) when a predetermined criterion is met. The trained neural network can then be made available for use (13).The invention also relates to a method, a computer program with instructions and a device for assigning a regulatory sign to a lane, as well as to a means of transport that uses a method or a device according to the invention.

Description

[0001] The present invention relates to a method, a computer program with instructions, and a device for training a neural network for assigning a regulatory sign to a lane, as well as to a neural network trained in this way. The invention also relates to a method, a computer program with instructions, and a device for assigning a regulatory sign to a lane, as well as to a means of transport that uses a method or device according to the invention.

[0002] Automated driving, also known as autonomous driving, automated driving, or guided driving, is the movement of vehicles, mobile robots, and driverless transport systems that are largely autonomous. There are different degrees of automated driving. In Europe, various transport ministries, such as the Federal Highway Research Institute in Germany, have defined the following levels of automation: • Level 0: “Driver only”, the driver drives, steers, accelerates, brakes, etc. • Level 1: Certain assistance systems help with vehicle operation, including a cruise control system such as ACC (Automatic Cruise Control). • Level 2: Partial automation. Automatic parking, lane guidance, general longitudinal guidance, acceleration, deceleration, etc., including collision avoidance, are handled by the assistance systems. • Level 3: High automation. The driver does not need to constantly monitor the system. The vehicle performs functions such as activating the turn signal, changing lanes, and maintaining lane guidance independently. The driver can attend to other tasks but must assume control upon request within a certain warning period. • Level 4: Full automation. The system permanently assumes control of the vehicle. If the system is no longer able to handle the tasks, the driver can be asked to take over. • Level 5: No driver required. Aside from setting the destination and starting the system, no human intervention is required.

[0003] A slightly different definition of steps is provided by the Society of Automotive Engineers (SAE). In this context, reference is made to the SAE J3016 standard. Such definitions can be used as an alternative to the definitions given above.

[0004] The ability to understand the stationary part of the environment is a key aspect of the driving function of automated vehicles. To overcome the deficiencies of current perception and environment modeling algorithms, modern driving functions rely on high-resolution maps that represent the stationary environment. However, a driving function that relies on such maps is prone to errors, as discrepancies can occur between the real world and the map data. For example, invalidated or reapplied road markings in a construction zone can cause the map to become outdated.

[0005] In addition, there is information that cannot be directly recognized. One example is whether a traffic sign or traffic light, hereinafter collectively referred to as "regulatory signs," belongs to a specific lane, i.e., is assigned to a specific lane. Typically, such information is manually entered into maps by map providers based on image recordings.

[0006] The driving functions of automated vehicles require knowledge of the relationships between regulatory signs and lanes, as this is key information for applications such as behavior prediction and path planning. These relationships are also important for map creation.

[0007] In this context, US 2021 / 0 404 841 A1 describes a method for deriving information about stationary elements based on semantic relationships. The method identifies a stationary element in a real-world environment for which information is to be derived, and detects a semantic relationship between the stationary element and other stationary elements in the real-world environment. Based on the detected semantic relationship, information about the stationary element is derived, which is then incorporated into a set of information describing the stationary element.

[0008] US 2022 / 0 266 824 A1 describes a device for generating road information. The device comprises an in-vehicle detection unit for detecting a situation in the surroundings of a vehicle, a microprocessor, and a memory connected to the microprocessor. The microprocessor is configured to detect a position of the vehicle, a lane of the vehicle, and a traffic light corresponding to the lane installed at an intersection before the vehicle enters the intersection, and to generate road information linking the detected traffic light to the vehicle's lane.

[0009] US 2023 / 0 343 109 A1 describes a method for detecting traffic lights on lanes using a camera and multiple models. Using a first model, depth and orientation information of traffic lights relative to a vehicle's lane is estimated from an image. Using a second model, relevance scores for the traffic lights are calculated based on geometric inferences between the depth and orientation information. Using the second model, a traffic light associated with the lane is assigned a primary relevance score based on the depth and orientation information.

[0010] It is an object of the invention to provide improved solutions for assigning a regulatory sign to a lane.

[0011] This object is achieved by a method having the features of claim 1 or 9, by a device having the features of claim 7 or 14, by a computer program with instructions according to claim 6 or 13, and by a means of transport according to claim 15. Preferred embodiments of the invention are the subject of the dependent claims.

[0012] According to a first aspect of the invention, a method for training a neural network for assigning a regulatory sign to a lane comprises the steps: - Providing input variables for the neural network, wherein the input variables comprise at least a representation of lanes and regulatory signs as well as prior knowledge; - Training the neural network with the input variables; and - End training when a predetermined criterion is met.

[0013] According to a further aspect of the invention, a computer program includes instructions which, when executed by a computer, cause the computer to perform the following steps for training a neural network to assign a regulatory sign to a lane: - Providing input variables for the neural network, wherein the input variables comprise at least a representation of lanes and regulatory signs as well as prior knowledge; - Training the neural network with the input variables; and - End training when a predetermined criterion is met.

[0014] The term "computer" should be understood broadly. In particular, it also includes workstations, distributed systems, and other processor-based data processing devices. The execution of these steps can be performed directly by the computer or involve the computer controlling a component intended to execute a step.

[0015] The computer program may, for example, be made available for electronic retrieval or stored on a computer-readable storage medium.

[0016] According to a further aspect of the invention, a device for training a neural network for assigning a regulatory sign to a lane comprises: - an input module for providing input variables for the neural network, wherein the input variables comprise at least a representation of lanes and regulatory signs as well as prior knowledge; and - a training module for training the neural network with the input variables, wherein the training module is configured to terminate the training when a predetermined criterion is met.

[0017] In the inventive solution, a neural network is trained to be able to determine a valid association between a regulatory sign and a lane, taking prior knowledge into account. This prior knowledge can relate in particular to geographical aspects, e.g., whether the regulatory sign is visible from the lane, whether the regulatory sign is located to the left, right, or above the lane, or whether the regulatory sign is the lane's nearest neighbor. The neural network thus considers relationships between regulatory signs and lanes in a similar way to how humans judge such situations.

[0018] One advantage of a neural network trained with prior knowledge over conventional neural networks or rule-based approaches is that the neural network trained in this way decides how the various aspects of prior knowledge should be weighted when modeling the assignments. Unlike conventional algorithms that process prior knowledge in the form of assumptions and use thresholds for this purpose, the inventive solution does not make binary decisions regarding the application of aspects of prior knowledge. The inventive solution therefore combines the advantages of neural networks, i.e., the learning of patterns, with the advantages of rule-based approaches, i.e., the application of knowledge.

[0019] The neural network can be implemented, for example, as a multi-layer perceptron with an input layer, several hidden layers, e.g., three hidden layers, and a neuron in the output layer. The neuron in the output layer can specify the probability that a combination of a rule-defining sign and a lane forms a valid assignment. For example, the probability can be specified as a value in the range from 0 to 1.

[0020] According to one aspect of the invention, the predetermined criterion is the maximization of the degree of fulfillment of the prior knowledge. Preferably, the learning process does not aim at pure loss minimization, but rather at maximizing the degree to which the defined prior knowledge is fulfilled, with the relationship loss = 1 - degree of fulfillment holding. When the degree of fulfillment converges to the global maximum, the prior knowledge has been learned as best as possible by the neural network.

[0021] According to one aspect of the invention, the representation of lanes and regulatory signs is a representation in the form of a tensor. The tensor can, for example, contain multiple coordinates of points belonging to a regulatory signal or a lane. Which features are used to create the tensor depends on which features are relevant for inferences. For example, if inferences are to be drawn based on geometric parallelism, the features of the tensor must be suitable as a basis for calculating parallelism.

[0022] According to one aspect of the invention, the prior knowledge is provided in a Real Logic syntax. Real Logic syntax is a fully differentiable logical language that is well suited to representing prior knowledge in a form suitable for machine learning.

[0023] According to one aspect of the invention, the neural network is a logic tensor network. A logic tensor network is a neurosymbolic framework that supports querying, learning, and reasoning with large-scale data and abstract knowledge about the world.

[0024] Advantageously, a neural network for assigning a regulatory sign to a lane is trained using a method according to the invention. Such a neural network is capable of assigning regulatory signs detected by a sensor system of a vehicle to a lane with high reliability.

[0025] According to a further aspect of the invention, a method for assigning a regulatory sign to a lane comprises the steps: - Detecting the surroundings of a means of transport; - Creating a representation of lanes and regulatory signs of the detected environment; and - Processing the representation with a neural network according to the invention.

[0026] According to a further aspect of the invention, a computer program contains instructions which, when executed by a computer, cause the computer to perform the following steps for assigning a regulatory sign to a lane: - Detecting the surroundings of a means of transport; - Creating a representation of lanes and regulatory signs of the detected environment; and - Processing the representation with a neural network according to the invention.

[0027] The term "computer" should be understood broadly. In particular, it also includes control units, embedded systems, and other processor-based data processing devices. The execution of these steps can be performed directly by the computer or involve the computer controlling a component intended to execute a step.

[0028] The computer program may, for example, be made available for electronic retrieval or stored on a computer-readable storage medium.

[0029] According to a further aspect of the invention, a device for assigning a regulatory sign to a lane comprises: - a detection module for detecting an environment of a means of transport; - a processing module for creating a representation of lanes and regulatory signs of the detected environment; and - an inventive neural network for processing the representation.

[0030] In the inventive solution, the trained neural network, based on the prior knowledge used during training, is able to interpret the stationary environment to determine which regulatory sign is valid for which lane, similar to a human driver. It navigates through unknown environments by taking prior knowledge into account, i.e., heuristics such as searching for the visibility and orientation of regulatory signs and adhering to general rules, such as permissible distances between regulatory signs and lanes.

[0031] According to one aspect of the invention, the representation of lanes and regulatory signs is a representation in the form of a tensor. The tensor can, for example, contain multiple coordinates of points belonging to a regulatory signal or a lane.

[0032] According to one aspect of the invention, an output of the neural network is provided as an input for a driving function. The driving function is thus enabled to plan the further journey with the correct information about which regulatory sign is valid for which lane. This increases the safety of an autonomous or semi-autonomous driving function.

[0033] According to one aspect of the invention, an output of the neural network is provided as an input for a map update. The update can be performed directly in the vehicle. Alternatively or additionally, updates can be performed on an external server. This server can then make the updated map available to other vehicles that may be affected by the critical areas at a later time.

[0034] Advantageously, a means of transport comprises a device according to the invention or is configured to carry out a method according to the invention for assigning a regulatory sign to a lane. The means of transport can be, in particular, a passenger car, a bus, or a commercial vehicle, e.g., a truck, an agricultural machine, or a construction machine. The solution according to the invention can be used particularly advantageously in autonomous or semi-autonomous means of transport. However, the solution according to the invention can also be used in manually controlled means of transport, e.g., to provide data for an assistance system.

[0035] Further features of the present invention will become apparent from the following description and the appended claims taken in conjunction with the figures. Fig. 1 schematically shows a method for training a neural network for assigning a regulatory sign to a lane; Fig. 2 shows a first embodiment of a device for training a neural network for assigning a regulatory sign to a lane; Fig. 3 shows a second embodiment of an apparatus for training a neural network for assigning a regulatory sign to a lane; Fig. 4 schematically shows a method for assigning a regulatory sign to a lane; Fig. 5 shows a first embodiment of a device for assigning a regulatory sign to a lane; Fig. 6 shows a second embodiment of a device for assigning a regulatory sign to a lane; Fig. Figure 7 schematically represents a means of transport in which a solution according to the invention is implemented; and Fig. Figure 8 shows different scenarios in which a rule-based sign can be assigned to a lane.

[0036] To better understand the principles of the present invention, embodiments of the invention are explained in more detail below with reference to the figures. It is understood that the invention is not limited to these embodiments and that the described features may also be combined or modified without departing from the scope of the invention as defined in the appended claims.

[0037] Fig. 1 schematically shows a method for training a neural network for assigning a regulatory sign to a lane. The neural network is preferably a logic tensor network. In a first step, input variables are provided 10 for the neural network. The input variables comprise at least a representation of lanes and regulatory signs, e.g., in the form of a tensor, as well as prior knowledge. The prior knowledge can be provided, for example, in a real logic syntax. The neural network is then trained 11 with the input variables. Training is terminated 12 when a predetermined criterion is met. Preferably, the predetermined criterion is the maximization of the degree of fulfillment of the prior knowledge. The trained neural network can then be made available for use 13.

[0038] Fig. 2 shows a simplified schematic representation of a first embodiment of a device 20 for training a neural network N for assigning a regulatory sign to a lane. Preferably, the neural network N is a logic tensor network. The device 20 has an input 21 for receiving data. An input module 22 is configured to provide input variables E for the neural network N. The input variables E comprise at least one representation R of lanes and regulatory signs, e.g., in the form of a tensor, as well as prior knowledge V. The prior knowledge V can be provided, for example, in a real logic syntax. A training module 23 is configured to train the neural network N with the input variables E and to terminate the training when a predetermined criterion K is met. Preferably, the predetermined criterion K is the maximization of the degree of fulfillment of the prior knowledge.An output module 24 is configured to provide the trained neural network N for use via an output 27 of the device 20.

[0039] The input module 22, the training module 23, and the output module 24 can be controlled by a control module 25. Settings of the input module 22, the training module 23, the output module 24, or the control module 25 can be changed via a user interface 28. The data generated in the device 20 can be stored in a memory 26 of the device 20 if necessary, for example, for later evaluation or for use by the components of the device 20. The input module 22, the training module 23, the output module 24, and the control module 25 can be implemented as dedicated hardware, for example, as integrated circuits. Of course, they can also be partially or completely combined or implemented as software running on a suitable processor, for example, a CPU or a GPU.The input 21 and the output 27 can be implemented as separate interfaces or as a combined bidirectional interface.

[0040] Fig. 3 shows a simplified schematic representation of a second embodiment of a device 30 for training a neural network for assigning a regulatory sign to a lane. The device 30 has a processor 32 and a memory 31. For example, the device 30 is a computer, a workstation, or a distributed system. Instructions are stored in the memory 31 which, when executed by the processor 32, cause the device 30 to carry out the steps according to one of the described methods. The instructions stored in the memory 31 thus embody a program executable by the processor 32 which implements the method according to the invention. The device 30 has an input 33 for receiving data. Data generated by the processor 32 are provided via an output 34. In addition, data can be stored in the memory 31.The input 33 and the output 34 can be combined to form a bidirectional interface.

[0041] The processor 32 may include one or more processor units, such as microprocessors, digital signal processors, or combinations thereof.

[0042] The memories 26, 31 of the described embodiments can have both volatile and non-volatile memory areas and can comprise a wide variety of storage devices and storage media, for example hard disks, optical storage media or semiconductor memories.

[0043] Fig. 4 schematically shows a method for assigning a regulatory sign to a lane. In a first step, the surroundings of a means of transport are recorded 40, e.g., based on sensor data from a sensor system of the means of transport. Subsequently, a representation of lanes and regulatory signs of the recorded surroundings is created 41, e.g., in the form of a tensor. The representation is processed 42 using a neural network according to the invention. An output of the neural network can then be provided as an input for a driving function 43. Alternatively or additionally, an output of the neural network can be provided as an input for a map update 44.

[0044] Fig. 5 shows a simplified schematic representation of a first embodiment of a device 50 for assigning a regulatory sign to a lane. The device 50 has an input 51, via which, for example, sensor data SD from a sensor system 71 of a means of transport can be received. A detection module 52 is configured to detect an environment of the means of transport, for example based on the sensor data SD. A processing module 53 is configured to create a representation R of lanes and regulatory signs of the detected environment, for example in the form of a tensor. A neural network N according to the invention is configured to process this representation R. An output module 54 can be configured to provide an output A of the neural network N via an output 57 of the device 50 as an input variable for a driving function or as an input variable for updating a map.

[0045] The acquisition module 52, the processing module 53, and the output module 54 can be controlled by a control module 55. Settings of the acquisition module 52, the processing module 53, the output module 54, or the control module 55 can be changed via a user interface 58. The data generated in the device 50 can be stored in a memory 56 of the device 50 if necessary, for example, for later evaluation or for use by the components of the device 50. The acquisition module 52, the processing module 53, the output module 54, and the control module 55 can be implemented as dedicated hardware, for example, as integrated circuits. Of course, they can also be partially or completely combined or implemented as software running on a suitable processor, for example, a CPU or a GPU.The input 51 and the output 57 can be implemented as separate interfaces or as a combined bidirectional interface.

[0046] Fig. 6 shows a simplified schematic representation of a second embodiment of a device 60 for assigning a regulatory sign to a lane. The device 60 has a processor 62 and a memory 61. For example, the device 60 is a computer, a control unit, or an embedded system. Instructions are stored in the memory 61 which, when executed by the processor 62, cause the device 60 to carry out the steps according to one of the described methods. The instructions stored in the memory 61 thus embody a program executable by the processor 62 which implements the method according to the invention. The device 60 has an input 63 for receiving data, e.g., sensor data from a sensor system of a means of transport. Data generated by the processor 62 are provided via an output 64. In addition, data can be stored in the memory 61.The input 63 and the output 64 can be combined to form a bidirectional interface.

[0047] The processor 62 may include one or more processor units, such as microprocessors, digital signal processors, or combinations thereof.

[0048] The memories 56, 61 of the described embodiments can have both volatile and non-volatile memory areas and can comprise a wide variety of storage devices and storage media, for example hard disks, optical storage media or semiconductor memories.

[0049] Fig. 7 schematically illustrates a means of transport 70 in which a solution according to the invention is implemented. In the example shown, the means of transport 70 is a motor vehicle. The motor vehicle has a sensor system 71 for detecting environmental information, such as cameras, radar sensors, lidar sensors or ultrasonic sensors. A neural network N is configured to assign a control sign detected by the sensor system 71 to a lane. The neural network N can be implemented, for example, in a computer 72 of the motor vehicle. An output A of the neural network N can be made available as an input variable to an assistance system 73, e.g. as the basis for a driving function, as the basis for outputting a warning on a display device 74 or as the basis for updating a map stored in a memory 75 of the motor vehicle.A data transmission unit 76 can be used to establish a connection to a backend 80, e.g., to transmit the output A of the neural network N, to retrieve an updated map, or to retrieve updated software for components of the motor vehicle. Data exchange between the various components of the motor vehicle takes place via a network 77.

[0050] In the following, an embodiment of the invention will be described with reference to Fig. 8 will be explained.

[0051] Fig. Figure 8 shows various scenarios in which the assignment of a regulatory sign Z to a lane S may be necessary. The arrows indicate the direction of travel of the respective lanes S. The regulatory signs Z are partly traffic signs, specifically speed limits in this case, and partly traffic lights. The scenarios depicted are those that may occur in practice in Germany. Fig. 8 a) there are two lanes S which are used in the same direction and for which different speed limits apply, which are indicated by traffic signs on both sides. Fig. 8 b) there are two lanes S which are used in the same direction and for which a uniform speed limit applies, which is indicated by traffic signs on both sides. Fig. 8 c) there are two lanes S which are used in the same direction and for which a single speed limit applies, which is indicated by a traffic sign on the right-hand side. Fig. 8 d) there are three lanes S which are used in the same direction and for which a uniform speed limit applies, which is indicated by traffic signs on both sides. Fig. 8 e) there are three lanes S which are used in the same direction and for which different speed limits apply, which are indicated by traffic signs located above the lanes. Fig. 8 f) there are two lanes S which are used in different directions. The speed limit for the right lane is indicated by a traffic sign on the right side. The speed limit for the left lane is indicated by a traffic sign on the left side, the content of which is not visible from the right lane. Fig. 8 g) there are two lanes S which are used in different directions. For the right lane S, a traffic light is located on the right side. In Fig. 8 h) there are two lanes S which are used in the same direction and for which traffic lights arranged above the lanes apply.

[0052] Using a real logic syntax, the prior knowledge reflected in the scenarios can be formulated as follows: • ∀Z: (ThereIs(S, Assignment(S, Z))) “For every rule-giving sign Z, there is at least one lane S with a valid assignment.” • ∀S, Z: (HasVerticalLine of Sight(Z, S)) → Assignment(S, Z) “For each pair of lane S and regulatory sign Z, the following applies: They only form a valid assignment if the regulatory sign has a vertical line of sight to the lane, ie is located either to the right or left of the lane or above the lane.” • VS, Z: (IsVisibleFor(Z, S)) → Assignment(S, Z) “For each pair of lane S and regulatory sign Z, the following applies: They only form a valid assignment if the regulatory sign for the lane is recognizable, ie the regulatory sign must have the correct orientation.” • VS1, S2, Z: Assignment(S1, Z) IF (1) IsNearestNeighborOf(S1, Z) OR (2) IsDirectNeighborLaneOf(S1, S2) AND NOT IsNearestNeighborOfAny(S1, Z) "For every group of three consisting of lane S1, lane S2 and regulatory sign Z, the following applies: Lane S1 and regulatory sign Z form a valid assignment if either (1) lane S1 is the nearest neighbor of the regulatory sign Z, or (2) lane S1 and lane S2 are directly adjacent lanes and lane S1 is not already the nearest neighbor of any regulatory sign."

[0053] With the above rule (1), the scenarios in Fig. 8 a), b), e) and h) are fully covered, the scenario in Fig. 8 c) for the right lane and the scenario in Fig. 8 d) for the two outer lanes.

[0054] The above rule (2) covers the special cases that are Fig. 8 c) and d), where a lane is not the nearest neighbor of a regulatory sign. In these scenarios, a valid assignment requires that there is an adjacent lane and that no other regulatory sign is the nearest neighbor.

[0055] Depending on the applicable traffic regulations or the applicable rules for the placement of regulatory signs, the logic for describing the prior knowledge may need to be adapted. For example, in the scenario of left-hand traffic, the Fig.8 c) the traffic sign is located on the left side. List of reference symbols 10 Providing input variables 11 Training the neural network 12 Ending the training 13 Providing the trained neural network 20 Device 21 Entrance 22 Input module 23 Training module 24 Output module 25 Control module 26 storage 27 Exit 28 User interface 30 Device 31 storage 32 processor 33 Entrance 34 Exit 40 Capturing an environment 41 Creating a representation 42 Processing the representation by a neural network 43 Providing an output of the neural network as input for a driving function 44 Providing a neural network output as input for a map update 50 device 51 Entrance 52 Recording module 53 Processing module 54 Output module 55 Control module 56 storage 57 Exit 58 User interface 60 device 61 storage 62 processors 63 Entrance 64 Exit 70 means of transport 71 Sensor technology 72 computers 73 Assistance system 74 Display device 75 storage 76 Data transmission unit 77 Network 80 backend E Input variable K criterion N Neural Network R Representation S lane SD sensor data V Prior knowledge Z Regulatory symbol QUOTES CONTAINED IN THE DESCRIPTION

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

[0000] US 2021 / 0 404 841 A1

[0007] US 2022 / 0 266 824 A1

[0008] US 2023 / 0 343 109 A1

[0009]

Claims

[1] Method for training a neural network (N) for assigning a regulatory sign (Z) to a lane (S), comprising the steps: - Providing (10) input variables (E) for the neural network (N), wherein the input variables (E) comprise at least one representation (R) of lanes (S) and regulatory signs (Z) as well as prior knowledge (V); - training (11) the neural network (N) with the input variables (E); and - Terminating (12) the training when a predetermined criterion (K) is met. [2] Method according to claim 1, wherein the predetermined criterion (K) is the maximization of the degree of fulfillment of the prior knowledge (V). [3] Method according to claim 1 or 2, wherein the representation (R) of lanes (S) and regulatory signs (Z) is a representation (R) in the form of a tensor. [4] Method according to one of the preceding claims, wherein the prior knowledge (V) is provided in a real logic syntax (10). [5] Method according to one of the preceding claims, wherein the neural network (N) is a logic tensor network. [6] Computer program with instructions which, when executed by a computer, cause the computer to carry out the steps of a method according to one of claims 1 to 5 for training a neural network (N) for assigning a control symbol (Z) to a lane (S). [7] Device (20) for training a neural network (N) for assigning a control sign (Z) to a lane (S), with - an input module (22) for providing (10) input variables (E) for the neural network (N), wherein the input variables (E) comprise at least one representation (R) of lanes (S) and regulatory signs (Z) as well as prior knowledge (V); and - a training module (23) for training (11) the neural network (N) with the input variables (E), wherein the training module (23) is configured to terminate the training (12) when a predetermined criterion (K) is met. [8] Neural network (N) for assigning a regulatory sign (Z) to a lane (S), wherein the neural network (N) was trained by means of a method according to one of claims 1 to 5. [9] Method for assigning a regulatory sign (Z) to a lane (S), comprising the steps: - detecting (40) an environment of a means of transport (70); - creating (41) a representation (R) of lanes (S) and regulatory signs (Z) of the detected environment; and - processing (42) the representation (R) with a neural network (N) according to claim 8. [10] Method according to claim 9, wherein the representation (R) of lanes (S) and regulatory signs (Z) is a representation (R) in the form of a tensor. [11] Method according to claim 9 or 10, wherein an output (A) of the neural network (N) is provided as an input variable for a driving function (43). [12] Method according to one of claims 9 to 11, wherein an output (A) of the neural network (N) is provided as an input for an update of a map (44). [13] Computer program with instructions which, when executed by a computer, cause the computer to carry out the steps of a method according to one of claims 9 to 12 for assigning a regulatory sign (Z) to a lane (S). [14] Device (50) for assigning a regulatory sign (Z) to a lane (S), comprising: - a detection module (52) for detecting (40) an environment of a means of transport (70); - a processing module (53) for creating (41) a representation (R) of lanes (S) and regulatory signs (Z) of the detected environment; and - a neural network (N) according to claim 8 for processing (42) the representation (R). [15] Means of transport (70), wherein the means of transport (70) comprises a device (50) according to claim 14 or is arranged to carry out a method according to one of claims 9 to 12 for assigning a regulatory sign (Z) to a lane (S).

Images