Methods for autonomous or semi-autonomous driving of a motor vehicle in relation to an external object and motor vehicle

Abstract

The invention relates to a method for the autonomous or semi-autonomous control of a vehicle. By using at least one neural network, edge endpoints (7), intermediate edge points (5), and local tangent slopes (9) or curve curvatures (14) are determined. Edge endpoints (7) mark a predefined end region of an object, intermediate edge points (5) mark line geometric shapes of the object, and the local information regarding the tangent slope and / or the curve curvature enables regression analysis to determine the object structure as a whole as precisely as possible, thus enabling optimal longitudinal and / or lateral control of a motor vehicle by means of a control unit.

Description

[0001] The invention relates to a method for autonomous or semi-autonomous driving of a motor vehicle with respect to an external object, comprising the steps of: providing image data describing at least one image of the vehicle's environment with the object located therein, by means of 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 trained to determine feature data, wherein the feature data relates to (or comprises) predetermined features (values) with respect 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 unit of the vehicle (provided to this), and longitudinal and / or lateral guidance of the vehicle is carried out based on control commands from the control unit based on the provided object structure.

[0002] The object recognition 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, lane markings, lane lines, or another vehicle when it comes to object recognition for driver assistance systems. However, conventional methods, such as the use of bounding boxes or image recognition, have limitations because they can only recognize objects of a specific, fixed shape as a whole. Therefore, providing a correct assignment, identification, and / or sequence of pixels or pixel groups, especially when determining connected lines, presents a challenging task.

[0003] To ensure the advantageous design of such object recognition systems, it is therefore desirable to identify an object, an object structure, or the center line 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 variations, i.e., a polyline shape.

[0004] An object recognition 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 recognition 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, is defined as an open or closed sequence of connected lines and / or arc segments and / or line segments that together no longer form a single straight line.

[0005] The publication 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, reveals the use of regression in lane detection.

[0006] From the publication 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.

[0007] From US 2018 / 0 285 659 A1 it is known to determine a tangent slope or curve curvature for image cells.

[0008] The publication by A. Monroy et al. entitled “Beyond straight lines - object detection using curvature”, in: IEEE International Conference on Image processing, 2011, IEEE, 2011, pages 3561-3564, ISBN 978-1-4577-1303-3 teaches the departure from the exclusive use of straight curve segments.

[0009] The topic of optical lane recognition is also addressed in DE 10 2021 208 204 A1.

[0010] The invention is based on the objective of enabling more reliable control of an autonomously or semi-autonomously guided vehicle by better recognizing the object structure of an object based on image data.

[0011] The object is achieved by the method with the features according to claim 1 and the motor vehicle with the features according to claim 4. Advantageous further developments of the invention are described by the dependent patent claims, the following description and the figures.

[0012] The inventive method for autonomously or semi-autonomously guiding a vehicle with respect to an external object is based on the method of the type 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, especially preferably arranged in a checkerboard pattern. For example, the image cells can be formed from an equal number of rows and columns of pixels, i.e., 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 finder of at least one neural network to the feature data; c) Determining image cells with edge intercepts of the object structure by applying an edge intercept finder of the at least one neural network to the feature data, wherein the determination tasks according to d) and e) are performed independently of each other; and the determination of the object structure further comprises: d) For each image cell with edge interpoints, determine the tangent slope of the object structure within the respective image cell by applying a tangent slope finder to the feature data and / or e) Determining the curvature of the object structure within the relevant image cells by applying a curvature analyzer to the feature data; and f) Performing a regression analysis based on the determined tangent slopes and / or curve curvatures using a regression unit of the neural network to determine the object structure as a whole.

[0013] In the invention, a plurality of object structures may be detected during the determination process (based on at least one image). It is then helpful to perform a check for possible duplicates (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. Because the determination of the tangent slope and the determination of the curve curvature are carried out independently of each other, the information on the tangent slope and the information on the curve curvature are obtained independently of each other.

[0014] Regarding the feature data, it should be noted that this can be determined using a so-called backbone, where the backbone is designed as, or includes, at least one artificial neural network. This backbone can, for example, comprise at least one residual neural network and / or densely connected convolutional network pre-trained on predefined features relating 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 provided as a feature vector. In other words, the at least one neural network can be configured to receive image data or visual data from at least one sensor device and to determine and / or extract feature data from it.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 designed as an encoder-decoder network. Overall, the at least one neural network can therefore perform semantic segmentation.

[0015] A feature map can be generated from at least one neural network. This neural network can comprise a large number of convolutional layers, allowing the feature map to encompass a receptive field (such as capturing 3x3 to 200x200 pixels) that can capture objects like traffic islands, road edges, and / or lane markings as a whole.

[0016] 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 pixel-based at least in groups independently of each other.

[0017] At least one neural network includes, as its first image processing functionality, a so-called edge endpoint finder, which is designed to identify and / or mark edge endpoints or endpoints of the object's structure. The edge endpoints each mark a predefined object endpoint region (in particular, a line end) of the object. The feature data is subjected to binary classification by the edge endpoint finder. This binary classification can involve normalizing feature vectors of the feature data, 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. An edge endpoint can be identified by the fact that an image area in the image data is part of a line shape (e.g., a line).An edge endpoint (e.g., a curb) is detected, and from this edge endpoint, the remainder of the line extends in only one direction (instead of two different directions). For example, an edge endpoint can be detected and / or set at the end of a lane line and / or lane marking. In other words, it can be checked for each pixel in an image whether an edge endpoint is present and / or represented.

[0018] Using the edge endpoint detector, a so-called seed point map can be generated, which contains or specifies marked edge endpoints, or 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 between 20x20cm and 75x75cm that end in a specific spatial direction or do not extend further in that direction. In this case, the object ends in that direction, and an edge endpoint can be set there. Thus, one-dimensional points (edge ​​endpoints or seed points) are determined using seed point detection.Therefore, start and / or end points can be learned that can be recognized as edge endpoints.

[0019] At least one neural network includes, as a second image processing functionality, an edge-intercept finder designed to identify and / or mark edge-intercepts, keypoints, and / or connection points of the object's structure on the feature data. The edge-intercepts mark line-geometric shapes of the object, where a line-geometric shape is part of the object's structure. These line-geometric shapes can consist of straight lines, arcs, a mixture of both, and / or polygonal surfaces.

[0020] The feature data are subjected to a further binary classification, different from the first, using the edge interpoint determiner, whereby this second binary classification may include normalizing the feature vectors of the feature data and ensuring that a normalized value, i.e., a classification value, is representative of exhibiting line-geometric forms of an object structure above a predetermined threshold.

[0021] The feature vectors are therefore preferably subjected to a binary classification, whereby the feature vectors can encompass parameterizable (numerical) properties of a geometric shape and / or pattern of the object in a vectorial manner. Different features characteristic of the pattern can form different dimensions of the feature vectors. The subsequent binary classification can thus be facilitated by the feature vectors, as they greatly reduce the properties to be classified (for example, instead of a complete 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).

[0022] In the above example of the lane arrow, individual line segments or 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 point. The edge point detector can be trained to signal sections with a length measured in the range of 20 cm to 75 cm, belonging to a continuous line running in two directions in the surrounding area, as edge points.

[0023] Additionally or alternatively, it may be provided that, by means of non-maximum suppression (NMS), only edge endpoints and / or intermediate edge points are determined and / or mapped that have an intensity or brightness value above a predetermined threshold.

[0024] As part of the method according to the invention, it is provided that, as a further image processing functionality, the determination of the tangent slope of the object structure within the relevant image cell is carried out by applying a tangent slope determiner to the feature data, and / or the determination of the curvature of the object structure within the relevant image cells is carried out by applying a curvature determiner to the feature data, followed by a regression. The tangent slope 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, for example, a lane marking (or an instance of the object structure represents a lane marking), this tangent slope indicates whether the lane marking is straight or curved.The direction of the straight line, or whether it deviates from a straight line, is determined. The tangent slope / gradient represents the first derivative of the object's structure, that is, abstractly, the derivative of the object's geometric representation. The curve curvature represents the second such derivative and could typically be determined manually by applying an incircle to the curve's structure. Regression analysis can involve applying a regression criterion or regression network from at least one neural network to the 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 done using so-called labeled data, which specifies lines of different curvature and, for each of these, the true or actual curvature value ("ground truth") resulting from the curvature.

[0025] The four image processing functionalities mentioned above determine whether an edge endpoint and / or an intermediate edge point is present in an image cell, and, in particular, if an intermediate edge point is present, what the tangent slope and / or curve curvature of the object structure is at that point. The first two image processing functions mentioned (edge ​​endpoint determination and intermediate edge point determination) can be performed independently of each other, so that the aforementioned corresponding information or estimates are obtained independently of one another.

[0026] The motor vehicle according to the invention comprises a control unit for enabling autonomous or semi-autonomous driving of the motor vehicle (by the control unit supplying control commands to corresponding actuators for longitudinal and / or lateral control, namely actuators for the engine, the brakes, the steering, and the like). The motor vehicle comprises a sensor device that provides or delivers digital image data of objects in the environment of the motor vehicle. The motor vehicle further comprises at least one neural network having a backbone unit designed to determine feature data from digital image data relating to predefined features concerning basic geometric shapes and / or colors of an object. The neural network further comprises: - an edge endpoint finder for determining image cells with edge endpoints of an object structure of an object; - an edge-intercept finder for determining image cells with edge intercepts of the object structure; and furthermore: - a tangent slope calculator for determining the tangent slope of the object structure within an image cell with intermediate edge points; and - a curve curvature determiner for determining a curve curvature of the object structure within an image cell with edge interpoints, wherein the tangent slope determiner and the curve curvature determiner are designed to operate independently of each other.

[0027] Furthermore, the neural network includes a regression unit for performing regression analysis based on the determined tangent slopes and / or curve curvatures in order to determine the object structure as a whole. The neural network is designed to transmit a recognized object structure (or data describing it) (possibly internally) to the control unit, which is then configured to issue control commands to the actuators. Additionally, the neural network is designed to perform a comparison for possible duplicates when identifying multiple object structures from at least one image.

[0028] The motor vehicle serves for the internal execution of the method according to the invention. 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 drives past).

[0029] Advantageously, the control unit features a so-called computer vision function. The functions are thus integrated within the control unit. In particular, the neural network can be part of the control unit, which leads to high efficiency due to its compact design.

[0030] A method for training the neural network of the motor vehicle of the inventive type (and / or the neural network used within the framework of the method) that can be used to realize the invention involves providing images of lanes with numerical data on local tangent slopes for training the tangent slope detector and inputting this data (so-called "ground truth") into the tangent slope detector. A tangent slope detector can thus be a part of an artificial neural network trained for determining tangent slopes. Its training can be carried out using so-called labeled data, which specifies lines of different curvatures and, for each line, the true or actual tangent slope ("ground truth") resulting from the curvature.

[0031] The procedure for training the neural network of the motor vehicle, or the neural network used in the procedure, includes alternatively or additionally providing images of lanes with numerical data on local curve curvatures for training the curve curvature determiner and entering them (also as "ground truth") into the curve curvature determiner.

[0032] An advantageous embodiment provides that the determination of the object's structure is trained using backpropagation, employing training data that includes a label representing the object's structure. For example, the label for a line or polyline can be a known line position, line thickness, line length, line orientation, line color, and / or line type. Additionally or alternatively, as previously described, one or more points on a line can be marked, with the label indicating the percentage of the arc length of that line reached at each marked point.

[0033] The at least one neural network is trained using supervised learning. In backpropagation, an input, such as image data 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 an error in the neural network, and this error is then propagated back to the input layer via the output layer. During this process, the weights of the neural connections in the at least one neural network are adjusted based on their influence on the error. This ensures that when the input is re-introduced, the output approximates the desired output. Backpropagation allows the at least one neural network to be corrected. Optimization or improvement of the parameters of the at least one neural network can thus be performed.With the improved parameters of at least one neural network, the at least one neural network is suitable in the application phase to determine meaningful output vectors from input vectors that differ from the originally learned input vectors of the training cases.

[0034] For use cases or application situations that may arise during the procedure and are not explicitly described here, it may be provided that, according to the procedure, an error message and / or a request for user feedback is issued and / or a default setting and / or a predetermined initial state is set.

[0035] The invention also includes the control device for the motor vehicle. The control device can comprise a data processing device or a processor unit configured to carry out an embodiment of the method according to the invention. For this purpose, the processor unit can comprise 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 (Graphics Processing Unit), or an NPU (Neural Processing Unit) can be used as the microprocessor. Furthermore, the processor unit can comprise program code configured to carry out the embodiment of the method according to the invention when executed by the processor unit. The program code can be stored in a data memory of the processor unit.The processor setup can be based on at least one circuit board and / or at least one SoC (System on Chip).

[0036] The invention also includes further developments of the motor vehicle according to the invention, which have features as 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.

[0037] 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.

[0038] As a further solution, the invention also includes a computer-readable storage medium comprising program code which, when executed by a computer or a computer network, causes it to execute an embodiment of the method according to the invention. The storage medium can be provided at least partially as a non-volatile data storage medium (e.g., as flash memory and / or as an SSD - solid state drive) and / or at least partially as a volatile data storage medium (e.g., as RAM - random access memory). The storage medium can be located within the computer or computer network. However, the storage medium can also be operated, for example, as an app store server and / or cloud server on the internet. The computer or computer network can provide a processor circuit with, for example, at least one microprocessor.The program code can 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).

[0039] The invention also includes combinations of the features of the described embodiments. The invention therefore also includes realizations that each exhibit a combination of the features of several of the described embodiments, provided that the embodiments have not been described as mutually exclusive.

[0040] The following are exemplary embodiments of the invention described. This is illustrated by: Fig. 1 a schematic representation of an architecture of at least one artificial neural network for implementing an embodiment according to the invention; Fig. 2. A sketch illustrating the determination of tangent slopes to a curve and the subsequent filtering; and Fig. 3 a schematic representation of a motor vehicle according to an embodiment of the invention.

[0041] The exemplary embodiments described below are preferred embodiments of the invention. In these exemplary embodiments, the described components each represent individual features of the invention, which can be considered independently of one another and each further develops the invention independently. Therefore, the disclosure is intended to include combinations of features of the embodiments other than those shown. Furthermore, the described embodiments can also be supplemented by further features of the invention already described.

[0042] In the figures, identical reference symbols denote functionally equivalent elements.

[0043] Fig. Figure 1 shows a schematic representation of an architecture of at least one artificial neural network 100 for realizing an embodiment of the disclosure. First, image data 1, which describe at least one image of an environment and an object located therein and which were received from at least one sensor device, e.g., during a journey in a motor vehicle, can be provided.

[0044] 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 first weighted (multiplied) by a respective sensor-specific weighting factor, and a sum or a sum of squares can be calculated from the weighted sensor data in this way, thereby generating a total value. Sensor fusion can thus be performed. If the total value is greater than a threshold, image data 1 can be generated, indicating that an object is within the detection range of the sensors. If the total value is less than the threshold, image data 1 can be generated, indicating that no object is within the detection range. In this case, it can be provided that a message is issued indicating that no object was detected. If an object has been detected, the process can continue as follows: 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 predefined features relating to basic geometric shapes and / or colors of the object. This backbone 2 can employ a self-attention technique 11 and / or a feature pyramid network technique 12 and / or be trained to perform such a technique. This feature data can then be represented on a feature map 3.

[0045] In a first image processing step, edge endpoints 7 of the object structure of the object can be determined by applying an edge endpoint finder 6 of the at least one neural network to the feature data, whereby edge endpoints 7 each mark a predefined object end area of ​​the object.

[0046] In a second image processing step, edge intermediate points 5 of the object structure can be determined by applying an edge intermediate point finder 4 of the at least one neural network to the feature data, wherein edge intermediate points 5 mark line geometric shapes of the object, where a line geometric shape is a part of the object structure of the object.

[0047] In a first further image processing step, tangent slopes 9 to individual image cells or line segments of the object structure can be determined by applying a tangent slope determiner 8 of at least one neural network to the feature data.

[0048] In a second further image processing step, curve curvatures 14 to individual image cells or line segments of the object structure can be determined by applying a curve curvature detector 13 of at least one neural network to the feature data.

[0049] Units 4, 6, 8 and 13 can operate independently of each other, although units 8 and 13 preferably build upon the results of determining the edge intermediate points 5 of the object structure, so that tangent slopes and / or curve curvatures are not unnecessarily determined to points that do not belong to the object structure.

[0050] In map merging unit 15 of the neural network, the input data points 5, 7, 9, and 14 are fused to produce a seed point graph. Unit 16 can perform regression. A neural network doesn't perform a calculation in the strict sense, but rather applies a regression criterion. Unit 16 of the neural network searches for polylines and outputs them, thus generating the actual graph as an object structure. Unit 17 checks for possible duplications. Since, for example, an actual line marking in a digital image contains several adjacent pixels, it's possible that the image processing might detect multiple parallel lines where only one was visible in reality. Unit 17 therefore outputs the final object structure.Thus, the object structure is determined using the edge endpoints 7, the edge intermediate points 5 and furthermore the tangent slopes 9 and / or the curve curvatures 14.

[0051] Fig. Figure 2 shows a sketch illustrating filtering on an output of the neural network, as can be done in Unit 16.

[0052] The diagram on the left shows image cells that can represent single pixels or multiple pixels in combination. The actual curve K of the object structure passes 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, 20d, and 20e.

[0053] The tangent slope finder determines the local slope, given as a number between 0 and 1, where 0 represents a horizontal slope and 1 represents a vertical slope, thus differing by 90° (π).

[0054] The tangent slope is on the left side of Fig. 2 is represented 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.

[0055] After filtering, this is simplified by applying a threshold criterion, for example, assigning upward arrows if the slope is greater than 0.5 and assigning right or left arrows if the slope is less than 0.5. In the present example, the tangent slopes Sb, Sc, Sd, and Se are thus assigned the new tangent slopes S'b, S'c, S'd, and S'e resulting from the filtering.

[0056] The in Fig. The motor vehicle 200 shown is to be capable of being driven semi-autonomously or autonomously by means of a control unit 210, for which purpose the control unit 210 gives control commands to actuators 216, wherein the actuator 216 in Fig. 3 can symbolically represent a plurality of actuators that relate to the engine control, brakes, steering, and possibly other similar functions. Data about the environment is displayed to the control unit 210 by an image sensor 214, represented here as an optical camera. The camera captures images, in particular, which also show an object that, in this example, could be a lane marking 300.

[0057] In the example case, the control unit 210 of the motor vehicle 200 comprises the neural network 100. Fig. 1.

[0058] The 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 then to perform longitudinal and / or lateral guidance. By providing specifically local information regarding the tangent slope and / or the curve curvature, such object structures can be captured with exceptional precision, resolved as polylines, and thus enable reliable semi-autonomous or autonomous vehicle control.

[0059] Overall, the examples show how objects can be identified by polylines in a given camera image, where the neural network has different units (dedicated heads) that provide, in particular, purely local geometric information.

Claims

[1] Method for autonomously or semi-autonomously driving a vehicle with respect to an external object, comprising the steps: - Providing image data (1) describing at least one image of the vehicle's surroundings with the object located therein, by means of 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 predefined features with regard to basic geometric shapes and / or colors of the object; - Determining the object structure of the object by the neural network in a second stage using the feature data; - Providing the object structure to a control unit of the vehicle; - Performing longitudinal and / or lateral control of the vehicle based on control commands from the control unit using the provided object structure; where determining the object structure involves at least one neural network: 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 finder (6) of the at least one neural network to the feature data; c) Determining image cells with edge interpoints (5) of the object structure by applying an edge interpoint finder (4) of the at least one neural network to the feature data; where determining the object structure further includes: d) for each image cell with edge interpoints, determine a tangent slope (9) of the object structure within the respective image cell by applying a tangent slope determiner (8) to the feature data and e) Determining a curve curvature (14) of the object structure within the relevant image cells by applying a curve curvature determiner (13) to the feature data, wherein the tasks of determination according to d) and e) are carried out independently of each other; as well as f) Performing a regression analysis based on the determined tangent slope and / or curve curvature using a regression unit of the neural network to determine the object structure as a whole, whereby, when determining several object structures based on at least one image, a comparison is made for possible duplications. [2] Method according to claim 1, wherein the tasks of determining according to b) and c) are carried out independently of each other. [3] Method according to claim 1 or 2, characterized by , that a lane marking is represented by an object structure. [4] Motor vehicle with a control unit (210) for enabling autonomous or semi-autonomous driving of the motor vehicle and with a sensor unit (214) that provides digital image data of objects in an environment of the motor vehicle, and with at least one neural network (100) that has a backbone unit (2) designed to determine feature data from digital image data relating to predefined features relating to basic geometric shapes and / or colors of the object, the neural network further exhibits: - an edge endpoint finder (6) for determining image cells with edge endpoints of an object structure of an object; - an edge-intercept finder (4) for determining image cells with edge-intercepts of the object structure; and furthermore: - a tangent slope finder (8) for determining a tangent slope (9) of the object structure within an image cell with edge intermediate points; and - a curve curvature determiner (13) for determining a curve curvature (14) of the object structure within an image cell with edge interpoints, wherein the tangent slope determiner (8) and the curve curvature determiner (13) are designed to operate independently of each other; and furthermore: - a regression unit (16) for performing a regression analysis based on the determined tangent slopes and / or curve curvatures, wherein the neural network is designed to transmit a recognized object structure to the control unit (210) and the control unit (210) is designed to initiate longitudinal and / or lateral guidance of the motor vehicle based on a recognized object structure, wherein the neural network is designed to perform a comparison for possible duplicates when several object structures are detected based on at least one image. [5] Motor vehicle according to claim 4, wherein the control unit has a computer vision function.

Images