Method for the autonomous or partially autonomous guiding of a motor vehicle in relation to an external object, motor vehicle and method for training a neural network

Abstract

The invention relates to a method for the autonomous or partially autonomous guiding of a vehicle. Using at least one neural network, edge end points (7), edge intermediate points (5) and local tangent gradients (9) or bend curvatures (14) are determined. Edge end points (7) mark a predefined end region of an object, edge intermediate points (5) mark linearly geometric shapes of the object and by way of the local information relating to the tangent gradient and / or the bend curvature, regression determination is possible in order to ascertain as accurately as possible the object structure as a whole and thus to enable optimum longitudinal and / or transverse guiding of a motor vehicle to be carried out by means of a control device.

Description

[0001] DESCRIPTION

[0002] Method for autonomous or semi-autonomous driving of a motor vehicle with respect to an external object, motor vehicle and method for training a neural network

[0003] The invention relates to a method for autonomously or semi-autonomously driving a motor vehicle with respect to an external object, comprising the steps of: providing image data that describe at least one image of the vehicle's surroundings with the object located therein, by at least one sensor device, wherein the sensor device is preferably part of the vehicle; feeding the image data into at least one neural network that is trained to determine feature data, wherein the feature data relate to (or comprise) predetermined features (characteristics) with regard to basic geometric shapes (reference shapes) and / or colors of the object, wherein the feature data is determined in a first stage, and wherein, in a second stage, one or more object structures of the object are determined by the neural network based on this feature data.The object structure is transmitted to a control device of the vehicle (provided to the vehicle), and the vehicle is guided longitudinally and / or transversely using control commands from the control device on the basis of the provided object structure.

[0004] The object detection used here is a subfield of image processing that focuses on identifying individual objects in images. The term "object" refers, for example, to traffic islands and / or road markings and / or lane lines and / or another vehicle when it comes to object detection for driver assistance systems. However, conventional methods, such as the use of bounding boxes or image recognition, have limitations, as they can only detect objects of a certain, fixed shape as a whole. Providing the correct assignment and / or identification and / or ordering of pixels or pixel groups, especially when detecting connected lines, therefore represents a challenging task.

[0005] To ensure an advantageous design of such object recognition systems, it is therefore desirable to determine an object, an object structure, or a centerline of an object of any shape. "Object structure" here refers to the basic shape (e.g., a straight line segment) and / or the composition of the object's optical appearance from such basic shapes or characteristics, e.g., a polyline shape.

[0006] An object detection system specifically designed for lane detection typically only considers highway scenarios where lane markings and road edges can be described by straight lines. However, such an object detection system fails in urban scenarios, for example, where objects such as road edges and / or traffic islands can have any geometric shape, such as the aforementioned polyline shape. A polyline shape, or polyline, refers to an open or closed sequence of connected lines and / or arc segments and / or line segments that, taken together, no longer form a single straight line.

[0007] The paper entitled “A Keypoint-based Global Association Network for Lane Detection,” authored by Jinsheng Wang et al., published on April 15, 2022, and available in November 2023 at https: / / arxiv.org / abs / 2204.07335, discloses the use of regression in lane detection.

[0008] From the paper entitled "Focus on Local: Detecting Lane Marker from Bottom Up via Key Point", authored by Zhan Qu et al., published on May 28, 2021, and available in November 2023 at https: / / arxiv.org / abs / 2105.13680, a two-stage use of local information in pre-detection by neural networks is known: first the rough detection of intermediate points, then a more precise position detection to these.

[0009] The invention is based on the object of being able to control an autonomous or semi-autonomous vehicle more reliably by better recognizing an object structure on the basis of image data.

[0010] The object is achieved by the method having the features of claim 1, the motor vehicle having the features of claim 6, and the methods for training the neural network of the motor vehicle according to the invention according to claim 9 and claim 10, respectively. Advantageous developments of the invention are described by the dependent claims, the following description, and the figures.

[0011] The method according to the invention for autonomously or semi-autonomously driving a vehicle with respect to an external object is based on the method of the generic type as described above. Determining the object structure by the at least one neural network comprises: a) dividing the at least one image into image cells (the image cells are preferably all the same size, particularly preferably provided in a checkerboard pattern. For example, the image cells can be formed from an equal number of pixel rows and pixel columns; thus, the pixels from four columns in four rows can be combined to form one image cell, i.e., 4 x 4 = 16 pixels. Other dimensions are possible. The image cells can also be identical to pixels, i.e., image points.); b) determining image cells with edge endpoints of the object structure by applying an edge endpoint detector of the at least one neural network to the feature data;c) Determining image cells with intermediate edge points of the object structure by applying an intermediate edge point determiner of the at least one neural network to the feature data; and determining the object structure further comprises: d) for each image cell with intermediate edge points, determining a tangent gradient of the object structure within the respective image cell by applying a tangent gradient determiner to the feature data and / or e) determining a curve curvature of the object structure within the respective image cells by applying a curve curvature determiner to the feature data; and f) performing a regression determination based on the determined tangent gradients and / or curve curvatures by means of a regression unit of the neural network to determine the object structure as a whole.

[0012] With regard to the feature data, it should be noted that this can be determined using a so-called backbone, wherein the backbone is designed as or is comprised in the at least one artificial neural network, wherein the backbone can, for example, comprise at least one residual neural network and / or densely connected convolutional network pre-trained on predetermined features with regard to basic geometric shapes and / or colors of the object. The at least one neural network can additionally or alternatively be trained on data describing polygonal shapes and / or structures. The feature data can, for example, be comprised or provided as a feature vector. In other words, the at least one neural network can be designed to receive image data or visual data from at least one sensor device and to determine and / or extract feature data therefrom.The at least one neural network can comprise a convolutional neural network (CNN) and / or a graph neural network (GNN). Additionally or alternatively, the at least one neural network can be configured as an encoder-decoder network. Overall, the at least one neural network can thus perform semantic segmentation.

[0013] A so-called feature map can be generated from the at least one neural network. The at least one neural network can comprise a plurality of convolutional layers, whereby the feature map comprises a receptive field (e.g., the detection of 3 x 3 to 200 x 200 image points or pixels) that can, for example, detect objects such as traffic islands and / or road edges and / or lane markings as a whole.

[0014] In the invention, at least three and preferably four different image processing functionalities are connected downstream of the at least one neural network, which can process the features from the feature map on a pixel-based basis, at least in groups, independently of one another.

[0015] The at least one neural network comprises, as a first image processing functionality, a so-called edge endpoint detector, which is designed to determine and / or mark edge endpoints or endpoints of the object structure of the object. The edge endpoints each mark a predetermined object end region (in particular a line end) of the object. The feature data are subjected to a binary classification by means of the edge endpoint detector, wherein the binary classification can include that feature vectors of the feature data are normalized and a normalized value, i.e. a classification value, is representative of having or representing an edge endpoint of an object structure above a predetermined threshold value. An edge endpoint can be recognized by the fact that an image region is identified in the image data as a component of a line shape (e.g.a curb) is detected and, starting from this edge endpoint, a remainder of the line leads away in only one direction (instead of in two different directions). For example, an edge endpoint can be detected and / or set at an expiring lane line and / or lane marking. In other words, for each pixel of an image, it can be checked whether an edge endpoint is present and / or represented.

[0016] Using the edge endpoint detector, a so-called seed point map can be generated, whereby the seed point map has or indicates marked edge endpoints or so-called seed points. For example, if the image data shows a white lane arrow on a gray road surface, the two ends of the lane arrow (tip and opposite end) can be identified as edge endpoints. The edge endpoint detector can be trained to identify shapes with a size ranging from 20x20cm to 75x75cm and that end in one spatial direction or do not extend any further in this direction. In this case, the object ends in this direction, and an edge endpoint can be set there. Using so-called seed point detection, one-dimensional points (edge ​​endpoints or seed points) are detected.Therefore, start and / or end points can be learned, which can be recognized as edge end points.

[0017] The at least one neural network comprises, as a second image processing functionality, an edge intermediate point detector configured to determine and / or mark edge intermediate points and / or keypoints and / or connection points of the object structure of the object to the feature data. The edge intermediate points mark linear geometric shapes of the object, wherein a linear geometric shape is a part of the object structure of the object. The linear geometric shapes can be formed from straight lines and / or arcs and / or a mixture of both and / or polygonal surfaces.

[0018] The feature data are subjected to a further binary classification, different from the first, by means of the edge intermediate point determiner, wherein this second binary classification may comprise that feature vectors of the feature data are normalized and a normalized value, i.e. a classification value, is representative of having line-geometric shapes of an object structure above a predetermined threshold value.

[0019] The feature vectors are therefore preferably each subjected to a binary classification, whereby the feature vectors can comprise parameterizable (numerical) properties of a geometric shape and / or a pattern of the object in a vectorial manner. Different features characteristic of the pattern can form different dimensions of the feature vectors. Using the feature vectors, the subsequent binary classification can be facilitated because they greatly reduce the number of properties to be classified (for example, instead of an entire image, only a feature vector consisting of N numbers needs to be considered, where the number N is smaller than the number of pixels in the image).

[0020] In the above example of the lane arrow, for example, individual line segments or line sections of the boundary line formed by the transition between the white lane arrow and the gray road surface can each be identified as an edge intermediate point. The edge intermediate point detector can be trained to identify sections with a length in a range of 20 cm to 75 cm that belong to a continuous line running in two directions in the surrounding area as an edge intermediate point.

[0021] Additionally or alternatively, it can be provided that by means of non-maximum suppression (NMS) only edge end points and / or edge intermediate points are determined and / or imaged which have an intensity or brightness value above a predetermined threshold value.

[0022] As part of the method according to the invention, it is provided that, as a further image processing functionality, the determination of a tangent gradient of the object structure within the relevant image cell is carried out by applying a tangent gradient determiner to the feature data and / or the determination of a curve curvature of the object structure within the relevant image cells is carried out by applying a curve curvature determiner to the feature data and then a regression is carried out. The tangent gradient can also be referred to as a "gradient" and is a numerical value that relates to local properties of the object structure. If the object structure represents a lane marking (or an instance of the object structure represents a lane marking), this tangent gradient indicates a change as to whether the lane marking is straight orin which direction the straight line runs, or whether a shape deviates from a straight line. The tangent slope / gradient indicates the first derivative of the object structure, i.e., in abstract terms, the derivative of the geometric representation of the object. The curve curvature indicates the second such derivative and could typically be determined manually by applying an inner circle to the curve structure. The regression determination can involve applying a regression criterion or regression network of the at least one neural network to the determined data (tangent slopes and / or curve curvatures) output by the tangent slope determiner and / or the curve curvature determiner. A curve curvature determiner can be a part of an artificial neural network trained to determine curve curvatures.Its training can be carried out using so-called labeled data, which on the one hand indicate lines of different curvature and, in each case, the true or actual curvature value resulting from the curvature (“ground truth”).

[0023] The four image processing functions mentioned above determine whether an edge endpoint and / or intermediate edge point is present in an image cell, and in particular, if an intermediate edge point is present, which tangent gradient and / or curve curvature the object structure has there. The first two image processing functions mentioned above (edge ​​endpoint determination and intermediate edge point determination) can be performed independently of one another, thus independently yielding the aforementioned corresponding information or estimates. Furthermore, determining the tangent gradient and determining the curve curvature can themselves be performed independently of one another, thus independently yielding the information on the tangent gradient and the information on the curve curvature.

[0024] According to an advantageous embodiment of the invention, a plurality of object structures may be recognized during the determination process (based on at least one image). It is then helpful to check for possible duplications (a "non-maximum suppression" approach). As already explained, a lane marking is advantageously represented as the object structure or an instance of the object structure. The motor vehicle according to the invention comprises a control device for enabling autonomous or semi-autonomous control of the motor vehicle (by the control device applying control commands to corresponding actuators for performing longitudinal and / or lateral control, namely actuators for the engine, brakes, steering, and the like). The motor vehicle comprises a sensor device that provides or delivers digital image data on objects in the environment of the motor vehicle.The motor vehicle further comprises at least one neural network having a backbone unit configured to determine feature data relating to predetermined features with respect to basic geometric shapes and / or colors of an object from digital image data. The neural network further comprises: an edge endpoint determiner for determining image cells with edge endpoints of an object structure of an object; an edge intermediate point determiner for determining image cells with intermediate edge points of the object structure; and furthermore: a tangent gradient determiner for determining a tangent gradient of the object structure within an image cell with intermediate edge points; and / or a curve curvature determiner for determining a curve curvature of the object structure within an image cell with intermediate edge points.

[0025] Furthermore, there is a regression unit in the neural network for carrying out a regression determination based on the determined tangent gradients and / or curve curvatures in order to determine the object structure as a whole, and the neural network is designed to transmit a recognized object structure (or data for describing the same) (internally if necessary) to the control device, which is then designed to output the control commands to the actuators according to its design.

[0026] The motor vehicle serves to implement the method according to the invention internally. Alternatively, it would be possible to outsource units such as the neural network to external devices (“edge computing,” outsourcing computing tasks to local computing devices that the vehicle passes by).

[0027] Advantageously, the control device has a so-called computer vision function. The functions are thus combined in the control device. In particular, the neural network can be part of the control device, which leads to high efficiency due to its compactness.

[0028] According to an advantageous embodiment, the tangent gradient determiner and the curve curvature determiner are designed to work independently of each other.

[0029] The inventive method for training the neural network of the motor vehicle of the type according to the invention (and / or the neural network used within the method) involves providing images of lanes with numerical data on local tangent gradients for training the tangent gradient detector and inputting these data (i.e., so-called "ground truth") into the tangent gradient detector. A tangent gradient detector can thus be a part of an artificial neural network trained to determine tangent gradients. Its training can be carried out using so-called labeled data, which, on the one hand, indicate lines of varying curvature and, in each case, the true or actual tangent gradient resulting from the curvature ("ground truth").

[0030] The method according to the invention for training the neural network of the motor vehicle, the method for training the neural network of the motor vehicle, or the neural network used in the method, alternatively or additionally comprises, for training the curve curvature determiner, providing images of lanes with numerical data on local curve curvatures and inputting them (likewise as "ground truth") into the curve curvature determiner. An advantageous embodiment provides that the determination of the object structure of the object is trained by means of error feedback (backpropagation), wherein training data is used that has a label of the object structure. As a label of a line or polyline, for example, it can be provided that a line position and / or line thickness and / or line length and / or line orientation and / or line color and / or line type are known.Additionally or alternatively, as already described, one or more points can be marked on a line, with the label previously known being what percentage of the arc length of this line has been reached up to a respective marked point.

[0031] The at least one neural network is therefore trained using supervised learning. During backpropagation, an input, such as image data, which is transformed into input vectors, is propagated through the at least one neural network. The resulting output of the at least one neural network is compared with the desired output. The difference between the two values ​​is considered the error of the neural network, and the error is then propagated back via the output layer to the input layer. In the process, the weights of the neuron connections of the at least one neural network are changed depending on their influence on the error. This guarantees an approximation of the desired output when the input is applied again. Using backpropagation, the at least one neural network can be corrected. The parameters of the at least one neural network can thus be optimized or improved.With the thus improved parameters of the at least one neural network, the at least one neural network is suitable in the application phase to determine meaningful output vectors (outputs) from input vectors (inputs) that deviate from the originally learned input vectors of the training cases.

[0032] For use cases or application situations that may arise during the procedure and which are not explicitly described here, it may be provided that, in accordance with the procedure, an error message and / or a request for user feedback is issued and / or a

[0033] Default setting and / or a predetermined initial state is set.

[0034] The invention also includes the control device for the motor vehicle. The control device can have a data processing device or a processor device that is configured to carry out an embodiment of the method according to the invention. For this purpose, the processor device can have at least one microprocessor and / or at least one microcontroller and / or at least one FPGA (Field Programmable Gate Array) and / or at least one DSP (Digital Signal Processor). In particular, a CPU (Central Processing Unit), a GPU (Graphical Processing Unit), or an NPU (Neural Processing Unit) can be used as the microprocessor. Furthermore, the processor device can have program code that is configured to carry out the embodiment of the method according to the invention when executed by the processor device. The program code can be stored in a data memory of the processor device.The processor device can be based, for example, on at least one circuit board and / or on at least one SoC (System on Chip).

[0035] The invention also includes further developments of the motor vehicle according to the invention that have features already described in connection with the further developments of the method according to the invention. For this reason, the corresponding further developments of the motor vehicle according to the invention are not described again here.

[0036] The motor vehicle according to the invention is preferably designed as a motor vehicle, in particular as a passenger car or truck, or as a passenger bus or motorcycle.

[0037] As a further solution, the invention also encompasses a computer-readable storage medium comprising program code which, when executed by a computer or computer network, causes the computer to carry out an embodiment of the method according to the invention. The storage medium can be provided at least partially as a non-volatile data memory (e.g. as a flash memory and / or as an SSD - solid state drive) and / or at least partially as a volatile data memory (e.g. as a RAM - random access memory). The storage medium can be arranged in the computer or computer network. However, the storage medium can also be operated on the Internet, for example, as a so-called app store server and / or cloud server. The computer or computer network can provide a processor circuit with, for example, at least one microprocessor.The program code may be provided as binary code and / or as assembly code and / or as source code of a programming language (e.g. C) and / or as a program script (e.g. Python).

[0038] The invention also encompasses combinations of the features of the described embodiments. The invention therefore also encompasses implementations that each comprise a combination of the features of several of the described embodiments, unless the embodiments are described as mutually exclusive.

[0039] Exemplary embodiments of the invention are described below. Shown are:

[0040] Fig. 1 is a schematic representation of an architecture of at least one artificial neural network for implementing an embodiment according to the invention;

[0041] Fig. 2 is a sketch illustrating the determination of tangent gradients to a curve and the subsequent filtering; and

[0042] Fig. 3 is a schematic representation of a motor vehicle according to an embodiment of the invention.

[0043] The exemplary embodiments explained below are preferred embodiments of the invention. In the exemplary embodiments, the described components of the embodiments each represent individual features of the invention that can be considered independently of one another, each of which also develops the invention independently of one another. Therefore, the disclosure is intended to encompass combinations of the features of the embodiments other than those shown. Furthermore, the described embodiments can also be supplemented by further features of the invention already described.

[0044] In the figures, the same reference symbols designate elements with the same function.

[0045] Fig. 1 shows a schematic representation of an architecture of at least one artificial neural network 100 for implementing an embodiment of the disclosure. First, image data 1 can be provided, which describes at least one image of an environment and an object located therein and which was received from at least one sensor device, e.g., during a journey in a motor vehicle.

[0046] For this purpose, it can be provided that sensor data from at least one sensor device, e.g., from a camera system and a radar system, are previously weighted (multiplied) by a respective sensor-specific weighting factor, and a sum or a square sum can be calculated from the thus weighted sensor data, thereby generating a sum value. Thus, a sensor fusion can be performed. If the sum value is greater than a threshold value, image data 1 can be generated that signals that an object is within the detection range of the sensors. If the sum value is less than the threshold value, image data 1 can be generated that signals that no object is within the detection range. In this case, it can be provided that a message is output indicating that no object has been detected. If an object has been detected, the method can continue as follows:

[0047] The image data 1 can then be fed into at least one neural network, in particular into a backbone 2 of the at least one neural network, which is trained to determine feature data, wherein the feature data comprises predetermined features regarding basic geometric shapes and / or colors of the object. This backbone 2 can have a self-attention technique 11 and / or a feature pyramid network technique 12 and / or be configured to implement such a technique. This feature data can then be represented on a feature map 3.

[0048] In a first image processing step, edge endpoints 7 of the object structure of the object can be determined by applying an edge endpoint determiner 6 of the at least one neural network to the feature data, wherein edge endpoints 7 each mark a predetermined object end region of the object.

[0049] In a second image processing step, intermediate edge points 5 of the object structure can be determined by applying an intermediate edge point determiner 4 of the at least one neural network to the feature data, wherein intermediate edge points 5 mark line-geometric shapes of the object, wherein a line-geometric shape is part of the object structure of the object.

[0050] In a first further image processing step, tangent gradients 9 to individual image cells or line points of the object structure can be determined by applying a tangent gradient determiner 8 of the at least one neural network to the feature data.

[0051] In a second further image processing step, curve curvatures 14 for individual image cells or line points of the object structure can be determined by applying a curve curvature determiner 13 of the at least one neural network to the feature data.

[0052] Units 4, 6, 8, and 13 can operate independently of one another, although units 8 and 13 preferably rely on the results of determining the intermediate edge points 5 of the object structure, so that tangent gradients and / or curve curvatures are not unnecessarily determined for points that do not actually belong to the object structure. In the map merging unit 15 of the neural network, the output data 5, 7, 9, and 14 are merged to produce a graph of object points (seed point graph). Unit 16 can, in particular, perform regression. A neural network does not perform a calculation in the strict sense, but rather applies a regression criterion. Unit 16 of the neural network then searches for polylines and outputs them as the result, thus generating the actual graph as the object structure. Unit 17 checks for possible duplications.Since an actual line marking in a digital image contains several adjacent pixels, it is possible that several parallel lines are detected during image processing where only one was visible in reality. Unit 17 thus outputs the final object structure. The object structure is thus determined using the edge endpoints 7, the edge intermediate points 5, and also the tangent gradients 9 and / or the curve curvatures 14.

[0053] Fig. 2 shows a sketch to illustrate a filtering on an output of the neural network, such as can be carried out in unit 16.

[0054] On the left side, image cells are shown, which can represent individual pixels or a combination of multiple pixels. The real curve K of the object structure runs through several of these image cells. While image cells 20a and 20f are not traversed by curve K, the curve does pass through image cells 20b, 20c, as well as 20d and 20e.

[0055] The tangent gradient finder determines the local gradient, given as a number between 0 and 1, where 0 stands for a horizontal gradient and 1 for a vertical gradient, thus differing by 90° (TT).

[0056] The tangent gradient is shown on the left side of Fig. 2 as Sb for image cell 20b, as Sc for image cell 20c, as Sd for image cell 20d, and as Se for image cell 20e. After filtering, this is simplified by, for example, applying a threshold criterion to assign upward arrows if the gradient is greater than 0.5 and right or left arrows if the gradient is less than 0.5. In the present example, the new tangent gradients S'b, S'c, S'd, and S'e resulting from the filtering are assigned to the tangent gradients Sb, Sc, Sd, and Se.

[0057] The motor vehicle 200 shown in Fig. 3 is intended to be semi-autonomously or autonomously controlled by a control device 210. For this purpose, the control device 210 issues control commands to actuators 216. In Fig. 3, the actuator 216 can symbolically represent a plurality of actuators relating to the engine control, brakes, steering, and possibly the like. An image sensor 214, depicted here as an optical camera, displays data about the surroundings to the control device 210. The camera captures images, in particular, including an object, which in the example may include a road marking 300.

[0058] In the example case, the control device 210 of the motor vehicle 200 comprises the neural network 100 from Fig. 1.

[0059] The motor vehicle can use the object structure provided by the neural network to utilize a computer vision function, for example, to recognize lane 300 as mentioned above and subsequently perform longitudinal and / or lateral guidance. By providing the specifically local information regarding the tangent gradient and / or the curve curvature, such object structures can be detected particularly precisely, resolved as polylines, and thus reliable semi-autonomous or autonomous control of the vehicle can also be achieved.

[0060] Overall, the examples show how objects can be recognized that are recognizable by polylines in a given camera image, whereby the neural network has various units (special heads, English: “dedicated heads”) that provide in particular purely local geometric information.

Claims

PATENT CLAIMS 1 . A method for autonomously or semi-autonomously driving a vehicle with respect to an external object, comprising the steps: Providing image data (1 ) which describe at least one image of an environment of the vehicle with the object located therein, by at least one sensor device; Feeding the image data (1) into at least one neural network which, in a first stage, determines feature data, wherein the feature data relate to predetermined features with regard to basic geometric shapes and / or colors of the object; Determining an object structure of the object by the neural network in a second stage based on the feature data; Providing the object structure to a control device of the vehicle; Carrying out longitudinal and / or transverse guidance of the vehicle using control commands from the control device on the basis of the provided object structure; wherein the determination of the object structure by the at least one neural network comprises: a) dividing the at least one image into, preferably equally sized, image cells, b) determining image cells with edge endpoints (7) of the object structure by applying an edge endpoint determiner (6) of the at least one neural network to the feature data; c) determining image cells with intermediate edge points (5) of the object structure by applying an intermediate edge point determiner (4) of the at least one neural network to the feature data; wherein the determination of the object structure further comprises: d) for each image cell with intermediate edge points, determining a tangent gradient (9) of the object structure within the respective image cell by applying a tangent gradient determiner (8) to the feature data and / or e) determining a curve curvature (14) of the object structure within the respective image cells by applying a curve curvature determiner (13) to the feature data; and f) performing a regression determination based on the determined tangent gradient and / or curve curvatures by means of a regression unit of the neural network to determine the object structure as a whole.

2. Method according to claim 1, in which the tasks of determining according to b) and c) are carried out independently of one another.

3. Method according to claim 1 or 2, wherein the tasks of determining according to d) and e) are carried out independently of each other.

4. Method according to one of the preceding claims, in which, when several object structures are determined on the basis of at least one image, a comparison is carried out for possible duplications.

5. Method according to one of the preceding claims, characterized in that a lane marking is represented by an object structure.

6. Motor vehicle with a control device (210) for enabling autonomous or semi-autonomous driving of the motor vehicle and with a sensor device (214) which supplies digital image data on objects in an environment of the motor vehicle, and with at least one neural network (100) which has a backbone unit (2) which is designed to determine feature data from digital image data which relate to predetermined features with regard to basic geometric shapes and / or colors of the object, wherein the neural network further comprises: - an edge endpoint detector (6) for determining image cells with edge endpoints of an object structure of an object; - an edge intermediate point detector (4) for determining image cells with edge intermediate points of the object structure; and furthermore: - a tangent gradient determiner (8) for determining a tangent gradient (9) of the object structure within an image cell with intermediate edge points; and / or - a curve curvature detector (13) for determining a curve curvature (14) of the object structure within an image cell with intermediate edge points; and furthermore: - a regression unit (16) for carrying out a regression determination based on the determined tangent gradients and / or curve curvatures, wherein the neural network is designed to transmit a recognized object structure to the control device (210) and the control device (210) is designed to initiate longitudinal and / or transverse guidance of the motor vehicle based on a recognized object structure.

7. Motor vehicle according to claim 6, wherein the control device has a computer vision function.

8. Motor vehicle according to claim 6 or 7, wherein the tangent gradient determiner (8) and the curve curvature determiner (13) are designed to operate independently of one another.

9. A method for training the neural network of the motor vehicle according to one of claims 6 to 8, wherein, for training the tangent gradient determiner (8), images of lanes are provided with numerical information on local tangent gradients and are input into the tangent gradient determiner (8).

10. Method for training the neural network of the motor vehicle according to one of claims 6 to 8, in particular method according to claim 9, in which, for training the curve curvature detector (13), images of lanes are provided with numerical information on local curve curvatures and are input into the curve curvature detector.

Images