Abstract: A computer-implemented method for correction of digital maps, in particular for vehicle navigation. The method includes: using existing sensor data, i.e., from vehicles, which were acquired using an acquisition unit, creating a digital map by combining the sensor data with existing map data; iteratively training a machine AI model based on the initial digital map and the sensor data; applying the trained AI model to identify deviations between the digital map and the actual conditions of the environment of at least one vehicle; suggesting corrections of the identified deviations in the digital map based on the results of the machine AI model and displaying these suggestions for review by a human user and/or the trained AI model; wherein the correction of the digital map is carried out by a manual check by the user, and/or the correction of the digital map is carried out automatically using the AI model.

Abstract: A method for recognizing horizontal road markings and determining the course thereof. The method includes the steps of capturing an image of a road, and dividing a central region of the image into a plurality of vertically superimposed cells and assigning to each cell predefined lines that are variously aligned around a horizontal direction. In a further step, at least one probability value for the presence of a road marking and displacement values of the line to the road marking are calculated for each line of each cell. The probability values and the displacement values are subsequently entered into a calculation function, and at least one line is output. The course of the horizontal road marking is determined from the at least one line and the displacement values.

Abstract: The invention relates to a method for detecting line structures (14) in image data and determining their progression. The method comprises the steps of capturing (30) image data by means of an image sensor, dividing (32) the entire image data into a plurality of cells (18) and assigning at least one predefined and oriented line (22) of fixed length to each cell (18), and dividing (34) the at least one line (22) into a predetermined number of line segments (38).

Abstract: An ordered data structure for the computer-readable representation of a physical linear driving surface boundary in the surroundings of a vehicle as a virtual linear driving surface boundary is disclosed. The data structure is stored in a memory or communicated via a data interface and is either a component of a training data set for training an environment recognition system. The training data set further includes environment detection data which is a sensory representation of a vehicle environment in which the at least one physical linear driving surface boundary is located; OR a component of an output data set of an environment detection system. The ordered data structure defines a line object with a list of support points. The line object is assigned a semantic line property that characterizes the entire line object as a virtual linear driving surface boundary. At least one of the support points is assigned a separate semantic point property.

Abstract: A method for training a classifier for map verification. A plurality of surroundings data sets, each of which represents respective surroundings of motor vehicles during respective trips of the motor vehicles through the same geographic region, are used to create a digital map of the geographic region, wherein in particular information or data resulting from the mapping pipeline are used to train the classifier or for map verification. A method for map verification, a device, a computer program, and a machine-readable storage medium are also described.

Abstract: A computer-implemented method for training a machine learning system for transferring images of a source domain into a target domain. The method includes: ascertaining source patches based on source images of a source domain and target patches based on target images of a target domain, the source patches and the target patches each being assigned pixel-by-pixel pieces of meta-information; ascertaining tuples, each including one source patch and at least one target patch which characterizes a neighbor of the source patch nearest to k according to a similarity measure, k being a hyperparameter of the method and the similarity measure characterizing a similarity between a source patch and a target patch based on the pixel-by-pixel meta-information of the source patch and of the target patch; training the machine learning system based on the source patches of the tuples and on the target patches of the tuples.

Abstract: A method for determining at least one anchor for an anchor-based lane line recognition and/or roadway marking recognition in a digital image representation on the basis of sensor data that are obtained from at least one surroundings sensor of a system. The method includes at least the following steps: a) receiving a digital image representation, b) setting at least one row or one column of possible anchors in at least one area of the digital image representation, the row or column of possible anchors being situated at a distance from at least the upper and lower or left and right edge of the area of the digital image representation.

Abstract: Method and device for providing a description of a comparison of a surrounding area with a digital map. The method includes a step of acquiring surrounding area data value sets that represent the surrounding area of a vehicle, wherein this surrounding area includes static and/or dynamic objects, a step of reading in the digital map, wherein the digital map represents a digital image of the surrounding area of the vehicle, a step of creating the comparison of the surrounding area data value sets with the digital map, a step of creating the description of the comparison by means of a large language model, wherein the description represents at least a degree of correspondence of the static and/or dynamic objects with the digital image, and a step of providing the description.

Abstract: A method for determining a lane marking of a first lane for a vehicle. The method includes: providing measurement data from a monitoring of the surroundings of the vehicle; feeding the measurement data to at least one machine learning model; evaluating a course of the lane marking using the at least one machine learning model; evaluating a width of the lane marking using the at least one machine learning model. A method is also described for training at least one machine learning model for use in the above-mentioned method.

Abstract: A method for training a convolutional neural network. For each of a multiplicity of training input images, the method includes processing of the training input image by the convolutional network; processing of a scaled version of the training input image by the convolutional network; determining a pair of convolutional layers of the convolutional network so that a convolutional layer of the pair generates a first feature map for the training input image which has the same size as a second feature map which is generated by the other convolutional layer of the pair for the scaled version of the training input image; and calculating a loss between the first feature map and the second feature map; and training the convolutional neural network to reduce an overall loss which includes the calculated losses.

Abstract: A method for generating training datasets for training an evaluation algorithm using which an alignment of two map datasets can be evaluated to determine navigation information for a mobile device that is moving or is to move in an environment. The method includes, for each of a plurality of training datasets: providing two input feature datasets, wherein underlying map datasets each contain environmental information that has been acquired using a sensor of the mobile device; providing a transformation dataset generated during an alignment of the two input feature datasets; providing a reference transformation dataset as ground truth; determining a correlation dataset based on the two input feature datasets and/or the transformation dataset; determining a quality measure depending on an accuracy of a match between the transformation dataset and the reference transformation dataset; and providing the training dataset which includes the correlation dataset and the quality measure.

Abstract: A method for creating map data for the operation of an automated and/or assisting system of a vehicle or robot. A map for a traffic route portion of a traffic route is compared with a reference map for the traffic route portion on an automated basis. The map is used exclusively for creating the map data on an automated basis when the map does not deviate from the reference map by a predefined extent with regard to at least one predefined quality measure.

Abstract: A method for generating training datasets for training an evaluation algorithm using which an alignment of two map datasets can be evaluated to determine navigation information for a mobile device that is moving or is to move in an environment. The method includes, for each of a plurality of training datasets: providing two input feature datasets, wherein underlying map datasets each contain environmental information that has been acquired using a sensor of the mobile device; providing a transformation dataset generated during an alignment of the two input feature datasets; providing a reference transformation dataset as ground truth; determining a correlation dataset based on the two input feature datasets and/or the transformation dataset; determining a quality measure depending on an accuracy of a match between the transformation dataset and the reference transformation dataset; and providing the training dataset which includes the correlation dataset and the quality measure.

Abstract: A method and device for creating a digital map. The method includes: receiving first training data sets; generating second training data sets by means of a generative neural network by determining a second, perturbed image of the same surrounding area in each case for each first image, wherein each second training data set represents the second image of the corresponding first image; training a further neural network; receiving surrounding area data sets, which in each case represent a surrounding area image of a vehicle surrounding area, wherein these surrounding area data sets comprise a position description of the corresponding vehicle surrounding area; creating the digital map by merging the surrounding area images, depending on the position description, by means of a scan matching method, wherein the scan matching method includes at least the further neural network, and a step of providing the digital map.

Abstract: A method for aligning two map datasets. The method includes: providing two map datasets, each containing environmental information, wherein the environmental information in the two map datasets has been detected by a sensor of a mobile device, and at least one of the two map datasets is a sparse map dataset; providing the two map datasets as input feature data or determining input feature data based on the two map datasets; carrying out an alignment of the two map datasets using a machine learning algorithm based on sparse convolution, wherein output data including information about a transformative relation between the two map datasets are generated from the input feature data, via intermediate feature data in one or more intermediate layers.

Abstract: A method for aligning a first map section of a digital road map with a second map section of the digital road map that at least partially overlaps the first map section. The method includes: determining that a first relative rotation between the two map sections cannot be unambiguously ascertained; ascertaining a second relative rotation between a third map section of the digital road map and a fourth map section of the digital road map that at least partially overlaps the third map section, wherein the third map section and the fourth map section are adjacent to the first map section and to the second map section; aligning the first map section with the second map section; wherein the alignment includes ascertaining a relative rotation between the first map section and the second map section based on the ascertained second relative rotation.

Abstract: A method for recognizing horizontal road markings and determining the course thereof. The method includes the steps of capturing an image of a road, and dividing a central region of the image into a plurality of vertically superimposed cells and assigning to each cell predefined lines that are variously aligned around a horizontal direction. In a further step, at least one probability value for the presence of a road marking and displacement values of the line to the road marking are calculated for each line of each cell. The probability values and the displacement values are subsequently entered into a calculation function, and at least one line is output. The course of the horizontal road marking is determined from the at least one line and the displacement values.

Abstract: A method for recognizing an object from input data is disclosed. Raw detections are carried out in which at least one attribute in the form of a detection quality is determined for each raw detection. At least one further attribute for each raw detection is determined. A temporally or spatially resolved distance measure is determined for at least one attribute of the raw detections. Raw detections of a defined distance measure are combined to form a group of raw detections. The object is recognized from a group with at least one raw detection with the smallest distance measure of the at least one attribute in comparison with another raw detection, or from a group with at least one raw detection which were combined by combining at least two raw detections with the smallest distance measure of the at least one attribute to form said one raw detection.

Abstract: A method for training a convolutional neural network. For each of a multiplicity of training input images, the method includes processing of the training input image by the convolutional network; processing of a scaled version of the training input image by the convolutional network; determining a pair of convolutional layers of the convolutional network so that a convolutional layer of the pair generates a first feature map for the training input image which has the same size as a second feature map which is generated by the other convolutional layer of the pair for the scaled version of the training input image; and calculating a loss between the first feature map and the second feature map; and training the convolutional neural network to reduce an overall loss which includes the calculated losses.

Abstract: A computer-implemented method for training a machine learning system for transferring images of a source domain into a target domain. The method includes: ascertaining source patches based on source images of a source domain and target patches based on target images of a target domain, the source patches and the target patches each being assigned pixel-by-pixel pieces of meta-information; ascertaining tuples, each including one source patch and at least one target patch which characterizes a neighbor of the source patch nearest to k according to a similarity measure, k being a hyperparameter of the method and the similarity measure characterizing a similarity between a source patch and a target patch based on the pixel-by-pixel meta-information of the source patch and of the target patch; training the machine learning system based on the source patches of the tuples and on the target patches of the tuples.