EP3884427A1 - Verfahren sowie system zur bestimmung eines fahrkorridors - Google Patents
Verfahren sowie system zur bestimmung eines fahrkorridorsInfo
- Publication number
- EP3884427A1 EP3884427A1 EP19812933.0A EP19812933A EP3884427A1 EP 3884427 A1 EP3884427 A1 EP 3884427A1 EP 19812933 A EP19812933 A EP 19812933A EP 3884427 A1 EP3884427 A1 EP 3884427A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- neural network
- pixels
- artificial neural
- motor vehicle
- driving corridor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/94—Hardware or software architectures specially adapted for image or video understanding
- G06V10/95—Hardware or software architectures specially adapted for image or video understanding structured as a network, e.g. client-server architectures
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2420/00—Indexing codes relating to the type of sensors based on the principle of their operation
- B60W2420/40—Photo, light or radio wave sensitive means, e.g. infrared sensors
- B60W2420/403—Image sensing, e.g. optical camera
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W60/00—Drive control systems specially adapted for autonomous road vehicles
- B60W60/001—Planning or execution of driving tasks
Definitions
- the invention relates to a method for determining a driving corridor, a control device for a system for determining a driving corridor, a system for determining a driving corridor, a computer program for carrying out the method and a computer-readable data carrier with such a computer program.
- driver assistance systems that control a subsystem or several subsystems of the motor vehicle at least partially in an automated manner.
- An example of this are active lane keeping systems that detect an unwanted departure from a lane and counteract this by actively intervening in vehicle steering and / or vehicle acceleration. These driver assistance systems therefore control a longitudinal and / or transverse movement of the motor vehicle at least partially in an automated manner.
- Such driver assistance systems are often based on at least one camera that generates images of the surroundings of the motor vehicle.
- the images are processed automatically by means of image processing algorithms and the driver assistance system controls the motor vehicle based on the processed images at least partially in an automated manner, in particular in a fully automated manner. For example, based on the processed images, a driving corridor is determined for the lane departure warning system that corresponds to a lane to be kept.
- the object of the invention is therefore to provide a method and a system for determining a driving corridor of a motor vehicle, which ensure reliable and robust detection of the driving corridor in a large number of different situations.
- the object is achieved according to the invention by a method for determining a driving corridor for a motor vehicle.
- the motor vehicle has at least one camera and a control unit, the camera being designed to generate images from a front area in front of the motor vehicle and to forward them to the control unit.
- the control device comprises a machine learning module with an artificial neural network.
- the method comprises the following steps: an image of the front area in front of the motor vehicle is obtained from the at least one camera; characteristic image features of the image are extracted by means of the artificial neural network; and by means of the same artificial neural network, image points are determined based on the extracted characteristic image features, which delimit the driving corridor of the motor vehicle and / or at least one driving corridor adjacent to the driving corridor.
- both the characteristic image features of the image and the driving corridor are determined by the same neural network. It has been found that a particularly reliable and robust determination of the driving corridor and / or neighboring driving corridors is achieved in this way.
- a “driving corridor” is to be understood here and in the following as an area on a road, on a path or the like in which the motor vehicle can move safely. This is, for example, a marked lane or an area of an unmarked street or one Path that corresponds to a lane. Accordingly, there is always a driving corridor in which the motor vehicle is currently located, which is referred to below as the “current driving corridor”. Furthermore, for example on roads with several lanes per direction of travel there are sections in which at least one further possible driving corridor for the motor vehicle runs adjacent to the current driving corridor of the motor vehicle.
- those pixels are determined which limit the current driving corridor of the motor vehicle and / or at least one driving corridor adjacent to the current driving corridor of the motor vehicle.
- the pixels represent lateral boundaries of the corresponding driving corridor. From the point of view of a driver of the motor vehicle, the pixels each form part of a left-hand or a right-hand border of the corresponding driving corridor.
- at least one possible driving corridor preferably a plurality of possible driving corridors for the motor vehicle, is determined, since two mutually opposite lateral boundaries define the corresponding driving corridor.
- the same artificial neural network preferably also determines a number of existing driving corridors, in particular a number of driving corridors that are adjacent to the current driving corridor of the motor vehicle.
- the artificial neural network is preferably a neural network which, in addition to an input and an output layer, has a plurality of intermediate layers.
- the artificial neural network is accordingly trained with methods of "deep learning”.
- One aspect of the invention provides that the extraction of the characteristic image features corresponds to a classification of image areas of the image into one or more categories in each case.
- the individual image areas are grouped into one or more of the following categories: road, no road, drivable path, impassable terrain, lane marking, lane limitation, traffic signs, motor vehicle, cyclists, pedestrians, other road users, etc.
- At least those image areas are identified in this step and categorized that for Determination of the at least one driving corridor are relevant.
- road markings are an important indicator of the existence of a driving corridor, but cannot necessarily be used as the only feature for this, since they are faulty or simply missing in some road sections, for example in the area of construction sites.
- the image areas are classified in particular by means of image recognition methods by the artificial neural network. Accordingly, the artificial neural network is trained to group the individual image areas into one or more of the categories described above.
- the pixels comprise a height coordinate and a latitude coordinate, the value of one of the coordinates, in particular the height coordinate, being set to at least one search value and only the value of the other of the coordinates, in particular the latitude coordinate, being determined for determining the pixels .
- the pixels are along certain search lines, i.e. all points at which the corresponding coordinate corresponds to the search value are determined, which run in an image area relevant for determining the at least one driving corridor.
- the search lines can be predetermined or determined by the artificial neural network itself. Accordingly, one of the coordinates of the pixels is already fixed and the artificial neural network only has to determine the other of the two coordinates of the pixels, which saves computing time.
- the pixels are determined along four search lines. It has been found that the determination of four pixels per lateral limitation of the corresponding driving corridor provides a good compromise between robust detection of the driving corridor and the shortest possible computing time.
- At least two pixels are preferably determined for each search value, the pixels corresponding to a left boundary of at least one of the driving corridors and a right boundary of the same driving corridor. If there is more than one existing corridor accordingly more pixels per search value determined. For example, with two existing driving corridors, three pixels per search value are determined, since the two driving corridors share those pixels that separate the two driving corridors from one another. In general, with n existing driving corridors (n + 1), pixels are determined per search value, where n is a natural number greater than zero.
- the set of specific pixels can contain pairs of a pixel which is assigned to the left boundary of a driving corridor and a pixel which is assigned to the right boundary of the same driving corridor, the pixels of a pair having the same coordinate.
- the same coordinate is in particular the height coordinate.
- the artificial neural network has at least a first subnetwork and at least a second subnetwork, the first subnetwork, which is constructed in particular as a convolutional network, extracting the characteristic image features and / or the second subnetwork, which is constructed in particular as a network for regression is determined the pixels.
- the artificial neural network thus comprises two interconnected sub-networks, the first sub-network specializing in extracting the characteristic image features, and the second sub-network specializing in determining the pixels based on the characteristic image features extracted by the first sub-network, in particular by means of a regression analysis. It has been found that a combination of the two subnetworks enables a particularly reliable and robust determination of the pixels.
- At least one result curve is preferably determined, which laterally delimits the determined driving corridor, in particular wherein the result curve is a cubic spline or a Bezier spline.
- the result curve is a cubic spline or a Bezier spline.
- lateral boundary lines for the corresponding driving corridor are determined.
- the image points determined serve as support points for interpolating the at least one result curve.
- the artificial neural network can also determine the result curve.
- the result curve can also be used take place outside the machine learning module, for example be carried out by a corresponding computer program that runs on the control device, since the interpolation of the result curves can also take place without machine learning, in particular without deep learning, based on the known coordinates of the specific pixels.
- the machine learning module or rather the artificial neural network with the pixels, provides the bases for the interpolation of the at least one result curve, while the actual result curve is then not determined by the artificial neural network.
- the artificial neural network is trained with target training data, the target training data comprising images generated by the camera and information on actual characteristic image features present in the images and / or actual image points delimiting at least one driving corridor , with the following training steps:
- the weighting factors of the neural network are preferably changed in such a way that the error is minimized.
- "no error” can also be determined as an error.
- the artificial neural network is trained to extract or determine the characteristic features and / or the pixels with the smallest possible error.
- Euclidean distance squares between the determined pixels and the actual pixels and / or between points on the at least one result curve and points on a corresponding actual result curve are used to determine the error, in particular wherein the respective Euclidean distance squares are added up and / or be averaged.
- the actual result curve corresponds to the curve that results from interpolation of the actual pixels.
- the Euclidean distance square between the ascertained image points and the actual image points or between the points on the at least one result curve and the points on the corresponding actual result curve which, as described above, lie on a search line is preferably determined in each case.
- the target training data further preferably contain an actual number of existing driving corridors, so that the artificial neural network is additionally trained to determine the number of driving corridors. In particular, the number of driving corridors adjacent to the current driving corridor of the motor vehicle is determined.
- the artificial neural network in particular the first sub-network, is initially trained in a first training step with target training data which contain images generated by the camera and information on actual characteristic image features present in the images, and that the artificial neural network, in particular the second subnetwork, is subsequently trained in a second training step with target training data which contain images generated by the camera and corresponding actual pixels delimiting at least one of the driving corridors.
- the artificial neural network is initially trained only to extract the characteristic image features.
- the artificial neural network is then trained in the second training step to determine the pixels based on the extracted characteristic image features, in particular by means of a regression analysis.
- the two training steps are not independent of one another, since the second training step also has an influence on the image recognition capabilities of the artificial neural network, more precisely on the extraction of the characteristic image features. It has been found that an artificial neural network trained in this way delivers particularly robust and reliable results for the image points.
- the first or in an additional training step it can also be limited to a certain characteristic image feature, e.g. on the image feature "street - no street” to ensure particularly reliable detection.
- the determined pixels, the one or more result curves and / or further information based on the determined pixels about the determined driving corridor are transmitted to at least one driver assistance system of the motor vehicle.
- the driver assistance system then controls the motor vehicle at least partially automatically based on the transmitted data and / or information.
- the driver assistance system is, for example, a brake assistant, a lane departure warning system and / or a lane change assistance system.
- control device for a system for determining a driving corridor for a motor vehicle, with a machine learning module which comprises an artificial neural network, the control device being designed to carry out a method described above.
- the control device can be connected in a signal-transmitting manner to a camera of the motor vehicle and can receive images from the camera.
- the object is also achieved according to the invention by a system for determining a driving corridor for a motor vehicle, with a control device described above and at least one camera which is designed to generate images from a front area in front of the motor vehicle and to the Forward control unit.
- the method described above for determining a driving corridor for a motor vehicle is therefore carried out automatically by the control unit, more precisely by the machine learning module of the control unit.
- the object is also achieved according to the invention by a computer program with program code means in order to carry out the steps of a method described above if the computer program is executed on a computer or a corresponding computing unit, in particular on a computing unit of a control device described above.
- the machine learning module described above in the context of the method and / or the artificial neural network described above in the context of the method form or forms at least part of the computer program, in particular the entire computer program.
- Program code means are and are to be understood in the following as computer-executable instructions in the form of program code and / or program code modules in compiled and / or in uncompiled form, which can be in any programming language and / or in machine language.
- the object is also achieved according to the invention by a computer-readable data carrier on which a computer program described above is stored.
- the data carrier can be an integral part of the control device described above or can be formed separately from the control device.
- the data carrier has a memory in which the computer program is stored.
- the memory is any suitable type of memory that is based, for example, on magnetic and / or optical data storage.
- FIG. 1 schematically shows a motor vehicle with a system according to the invention for determining a vehicle corridor
- FIG. 2 is a schematic flow diagram of the steps of an inventive method for determining a vehicle corridor
- Figure 3 shows schematically an image processed according to the method of Figure 2;
- Figure 4 shows schematically a structure of an artificial neural network used for the method of Figure 2;
- FIG. 5 schematically shows a flow diagram of the training steps for the artificial neural network of Figure 4.
- FIG. 6 shows an illustration of the error determination during the training steps in FIG. 5.
- FIG. 1 schematically shows a motor vehicle 10 which is located on a multi-lane road 12 which has a first lane 14, a second lane 16 and a third lane 18.
- a direction of travel of motor vehicle 10 is indicated in FIG. 1 by an arrow 19.
- the first lane 14 is delimited on the right side by a lane marking RR and on the left side by a lane marking R.
- the second lane 16 is delimited on the right side by the lane marking R and on the left side by a lane marking L.
- the third lane 18 is delimited on the right side by the lane marking L and on the left side by a lane marking LL.
- the second lane 16 is the current lane of the motor vehicle 10, while the first lane 14, viewed in the direction of travel, runs to the right of the second lane 16 and the third lane 18 to the left of the second lane 16.
- the motor vehicle 10 has a system 20 for determining a driving corridor of the motor vehicle 10 with a camera 22 and a control unit 24, which is connected to the camera 22 in a signal-transmitting manner.
- a machine learning module 28 with an artificial neural network is implemented on the control unit 24, the functioning of which will be explained in more detail later.
- the artificial neural network represents a computer program that is stored on a data carrier 29 of the control device 24.
- the motor vehicle 10 has a further control device 26, which is connected to the control device 24 in a signal-transmitting manner and is designed to control the motor vehicle 10 at least partially in an automated manner, in particular in a completely automated manner.
- a driver assistance system is implemented on the further control device 26, which can control a transverse movement and / or a longitudinal movement of the motor vehicle 10 at least partially automatically, in particular fully automatically.
- the two control devices 24, 26 can also be partial control devices of a single control device.
- the driver assistance system which in the present case is, for example, a lane departure warning system or a lane change assistant, requires information about the surroundings of the motor vehicle 10 in order to control the motor vehicle 10 at least partially in an automated manner.
- the system 20 is designed to use the camera 22 to create images of a front area 30 in front of the motor vehicle 10, to process these images by means of the control device 24 and thus to determine a current driving corridor of the motor vehicle 10 and / or at least one of the current driving corridor of the vehicle To determine motor vehicle 10 adjacent driving corridor.
- a “driving corridor” is to be understood here and in the following as an area on the road 12, on a path or the like in which the motor vehicle 10 can move safely. This is, for example, one of the lanes 14, 16, 18 or an area of an unmarked road or a path that corresponds to a lane. Accordingly, the “current driving corridor” is to be understood as the driving corridor in which the motor vehicle 10 is currently located and in which the motor vehicle 10 can safely move on.
- the driving corridor is that area of the road 12 or a path which is recognized by the control unit 24 as a safe area to be driven on, while the tracks 14, 16, 18 are those of the Road markings RR, R, L, LL denote limited areas of road 12.
- the driving corridors and at least one of lanes 14, 16, 18 can coincide, but this is not necessarily the case.
- Characteristic image features of the individual images are now extracted using the machine learning module 28, more precisely using the artificial neural network (step S2).
- individual image areas of the individual images are grouped into one or more of the following categories: road, no road, drivable path, impassable terrain, lane marking, lane limitation, traffic signs, motor vehicle, cyclists, pedestrians, other road users, etc. .
- step S2 at least those image areas that are relevant for determining the at least one driving corridor are identified and categorized.
- the road markings RR, R, L, LL shown in FIG. 1 are an indication of the presence of a driving corridor. However, they are not necessarily used as the only feature for determining the at least one driving corridor, since the road markings RR, R, L, LL can be faulty or simply missing in some road sections, for example in the area of construction sites.
- step S3 pixels 32 are now determined which laterally limit the at least one driving corridor.
- FIG. 3 shows a camera image of a situation that is very similar to the situation shown in FIG. 1.
- the motor vehicle 10 is in the middle, second lane 16.
- Left and the third track 18 and the first track 14 run to the right of the second track 16.
- the pixels 32 are determined along four search lines 34 which run through the image along the width direction b of the image with a constant height coordinate h kl -h k4 .
- the pixels 32 are determined based on four search values, the search values being the height coordinate h of the respective search line 34. Any other number of search lines 34 is of course also possible.
- a pair of two pixels 32 is determined along each of the search lines 34.
- One of the pixels 32 of the pair represents the left-hand boundary of the driving corridor and the other of the pixels 32 of the pair represents the right-hand boundary of the driving corridor at the level of the respective search line 34.
- step S3 at least eight pixels 32 are thus determined, which limit the current lane of the motor vehicle 10.
- pixels 32 are shown in FIG. 3 which limit the current driving corridor of the motor vehicle 10 corresponding to the second lane 16.
- a determination of further pixels that limit further driving corridors is of course possible analogous to the case described above.
- adjacent driving corridors share those pixels 32 which separate the adjacent driving corridors from one another.
- 16 pixels would therefore be determined.
- At least two result curves 36 which laterally limit the current driving corridor of the motor vehicle 10, are then determined based on the determined pixels 32 (step S4).
- One of the result curves 36 thus delimits the current driving corridor of the motor vehicle 10 on the left-hand side and is determined by interpolating those four pixels 32 which limit the current driving corridor on the left-hand side.
- the other of the two result curves 36 limits the current driving corridor of the motor vehicle 10 on the right-hand side and is determined by interpolation of those four pixels 32 that limit the current driving corridor on the right-hand side.
- the two result curves 36 are determined based on the image coordinates, that is to say the respective height coordinate h and the respective latitude coordinate b of the pixels 32.
- the specific pixels 32 thus represent support points for the result curves 36.
- the result curves 36 are each a cubic spline, a Bezier spline or another type of curve suitable for the interpolation of the pixels 32.
- the determination of the result curves 36 can also be carried out by the artificial neural network. Alternatively, however, the result curves 36 can also be determined outside of the machine learning module 28, for example from a corresponding module that runs on the control unit 24.
- step S5 The specific pixels 32, the result curve 36 and / or further information based on the specific pixels 32 about the current vehicle corridor or adjacent vehicle corridors are now transmitted to the further control unit 26 (step S5).
- the motor vehicle 10 can be controlled differently by the further control device 26.
- the information is sufficient for a lane keeping assistant, but is inadequate for a lane change assistant.
- additional information must be available about at least one driving corridor adjacent to the current driving corridor.
- At least steps S2 and S3 described above, in particular also step S4, are therefore carried out by the same artificial neural network.
- the artificial neural network Basically, there are various suitable architectures for the artificial neural network that can be trained to carry out the method described above.
- An exemplary architecture of the artificial neural network is shown in FIG.
- the artificial neural network has a first sub-network 38 and a second sub-network 40, at least the first sub-network 38 being designed as a convolutional neural network (CNN), in particular as a deep convolutional neural network.
- CNN convolutional neural network
- the first subnetwork 38 and the second subnetwork 40 each have a plurality of layers 42.
- the layers 42 within the respective subnetworks 38, 40 are completely networked with their respective neighboring layers 42, so they are so-called “fully connected” (FC) layers.
- FC fully connected
- the two outer layers 42 of the two subnetworks 38, 40 are also completely networked with one another, which lie next to one another at the boundary between the two subnetworks 38, 40.
- an output layer of the first sub-network 38 is completely networked with an input layer of the second sub-network 40.
- the artificial neural network thus comprises two interconnected subnetworks 38, 40.
- the first subnetwork 38 is trained to extract the characteristic image features, that is to say perform step S2
- the second subnetwork 40 is trained thereon to determine the pixels 32 on the basis of the characteristic image features extracted from the first subnetwork 38, in particular by means of a regression analysis.
- the first subnetwork 38 therefore executes step S2
- the second subnetwork 40 executes step S3 and possibly step S4.
- Steps S1 and S5 can also be carried out by these two subnetworks 38, 40 or further parts of the artificial neural network.
- the neural network In order to ensure optimal results when determining the at least one driving corridor, the neural network must of course be trained before use in the system 20. For this purpose, the following training steps described with reference to FIG. 5 are provided before use.
- the first sub-network 38 is trained to extract the characteristic image features in the images generated by the camera 22 (step T1), that is to carry out step S2 described above.
- the artificial neural network in particular only the first sub-network 38, is fed with target training data, which contain images generated by the camera 22 and information on actual characteristic image features present in the images.
- the artificial neural network in particular the first sub-network 38, then extracts the characteristic image features from the target training data, more precisely from the images contained therein.
- an error or rather a deviation between the determined characteristic image features and the actual characteristic image features is determined and the artificial neural network is fed back with the error. Weighting factors of the artificial neural network are adjusted in such a way that the error or the deviation are minimized.
- the artificial neural network can also be trained to determine a number of existing driving corridors.
- the training steps T1 and T2 can also contain several sub-steps or phases.
- the artificial neural network pre-trained in training step T1 is now trained in a second training step to determine the pixels 32 based on the extracted characteristic image features.
- the artificial neural network is fed with target training data, which contain images generated by the camera 22 and information about the actual pixels 32 present in the images.
- the first sub-network 38 extracts the characteristic image features from the target training data, more precisely from the images contained therein, and transfers them to the second sub-network 40.
- the second Subnetwork 40 determines the image points 32 from the target training data based on the extracted characteristic image features.
- a measure of the error at each of the pixels 32 is the Euclidean distance of the pixel 32 determined from the respectively associated actual pixel 32 '.
- Xi (i is a natural number) is the latitude coordinate b of the actual i-th pixel 32 'and x t is the latitude coordinate of the associated pixel 32 determined by the neural network, the error associated with this pixel 32 is determined as (x * - x) 2 . Only the latitude coordinate of the pixel 32 is taken into account, since the elevation coordinate is fixed to the respective search value.
- the total error is calculated as the sum or as a weighted average of the errors of all pixels 32, ie the error results
- N denotes the number of relevant boundary lines. In the case of a single driving corridor, N is 2.
- an error function can also be used which is based on the Euclidean distance between points on the determined result curves 36 and the corresponding actual result curves 36 '.
- the result curve 36 and the actual result curve 36 ' are initially determined based on the pixels 32 and the actual pixels 32'.
- the Euclidean distances between the points corresponding to one another are then calculated on the result curves 36 or 36 ', points which are different from the pixels 32, that is to say which have a height coordinate different from the pixels 32, can also be considered.
- the artificial neural network is now fed backwards with the error. Weighting factors of the second sub-network 40, but also of the first sub-network 38, are adjusted in such a way that the error or the deviation are minimized.
- the artificial neural network trained in this way is then set up to carry out steps S1 to S5.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102018129485.9A DE102018129485A1 (de) | 2018-11-22 | 2018-11-22 | Verfahren sowie System zur Bestimmung eines Fahrkorridors |
| PCT/EP2019/082060 WO2020104579A1 (de) | 2018-11-22 | 2019-11-21 | Verfahren sowie system zur bestimmung eines fahrkorridors |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| EP3884427A1 true EP3884427A1 (de) | 2021-09-29 |
Family
ID=68733022
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| EP19812933.0A Withdrawn EP3884427A1 (de) | 2018-11-22 | 2019-11-21 | Verfahren sowie system zur bestimmung eines fahrkorridors |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US12283088B2 (de) |
| EP (1) | EP3884427A1 (de) |
| CN (1) | CN113168516A (de) |
| DE (1) | DE102018129485A1 (de) |
| WO (1) | WO2020104579A1 (de) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102021133089A1 (de) * | 2021-12-14 | 2023-06-15 | Cariad Se | Vorrichtung zur Ermittlung einer Topographie einer Fahrzeugumgebung, Fahrzeug und Verfahren |
| DE102024104238A1 (de) * | 2024-02-15 | 2025-08-21 | Volkswagen Aktiengesellschaft | Ermittlung von Fahrstreifen |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102011056671A1 (de) * | 2011-12-20 | 2013-06-20 | Conti Temic Microelectronic Gmbh | Bestimmung eines Höhenprofils einer Fahrzeugumgebung mittels einer 3D-Kamera |
| JP6185418B2 (ja) * | 2014-03-27 | 2017-08-23 | トヨタ自動車株式会社 | 走路境界区画線検出装置 |
| US9286524B1 (en) * | 2015-04-15 | 2016-03-15 | Toyota Motor Engineering & Manufacturing North America, Inc. | Multi-task deep convolutional neural networks for efficient and robust traffic lane detection |
| US9710714B2 (en) * | 2015-08-03 | 2017-07-18 | Nokia Technologies Oy | Fusion of RGB images and LiDAR data for lane classification |
| US10528824B2 (en) * | 2017-12-11 | 2020-01-07 | GM Global Technology Operations LLC | Artificial neural network for lane feature classification and localization |
-
2018
- 2018-11-22 DE DE102018129485.9A patent/DE102018129485A1/de active Pending
-
2019
- 2019-11-21 EP EP19812933.0A patent/EP3884427A1/de not_active Withdrawn
- 2019-11-21 WO PCT/EP2019/082060 patent/WO2020104579A1/de not_active Ceased
- 2019-11-21 CN CN201980077273.7A patent/CN113168516A/zh active Pending
- 2019-11-21 US US17/295,503 patent/US12283088B2/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020104579A1 (de) | 2020-05-28 |
| CN113168516A (zh) | 2021-07-23 |
| US20220004781A1 (en) | 2022-01-06 |
| US12283088B2 (en) | 2025-04-22 |
| DE102018129485A1 (de) | 2020-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020048669A1 (de) | Verfahren zum bestimmen einer spurwechselangabe eines fahrzeugs, ein computerlesbares speichermedium und ein fahrzeug | |
| DE102016203723A1 (de) | Verfahren und System zum Bestimmen der Pose eines Fahrzeugs | |
| AT521647B1 (de) | Verfahren und System zur Datenaufbereitung, zum Trainieren eines künstlichen neuronalen Netzes und zum Analysieren von Sensordaten | |
| WO2020127010A1 (de) | Verfahren sowie system zum steuern eines kraftfahrzeugs | |
| DE102018111935A1 (de) | Bildverarbeitungssystem, Bildverarbeitungsverfahren, Informationsverarbeitungsvorrichtung und Aufzeichnungsmedium | |
| DE102019214558A1 (de) | Projektionsinformations-erkennungsvorrichtung auf basis eines künstlichen neuronalen netzwerks und verfahren derselben | |
| DE102024129386A1 (de) | Verfahren und vorrichtung zur fahrerabsichtsprognose | |
| DE102017127600A1 (de) | Verfahren zum automatischen Parken für ein Kraftfahrzeug mit einem Fahrunterstützungssystem | |
| EP3785169A1 (de) | Verfahren und vorrichtung zur umsetzung eines eingangsbildes einer ersten domäne in ein ausgangsbild einer zweiten domäne | |
| DE102019209463A1 (de) | Verfahren zur Bestimmung eines Vertrauenswertes eines Objektes einer Klasse | |
| DE102024201831A1 (de) | Verfahren zum Erstellen von hochaufgelösten Umgebungskarten für ein Fahrzeug mit einer autonomen Fahrfunktion | |
| WO2024240427A1 (de) | Verfahren zur erstellung einer digitalen karte | |
| EP3884427A1 (de) | Verfahren sowie system zur bestimmung eines fahrkorridors | |
| EP3576013A1 (de) | Abschätzen eines verlaufs eines schienenpfads | |
| DE102021114935A1 (de) | Verfahren zum Bestimmen von zumindest einer Fahrstreifenbegrenzung einer Fahrbahn in einem Bild mittels eines Assistenzsystems, omputerprogrammprodukt, computerlesbares Speichermedium sowie Assistenzsystem | |
| EP4526842B1 (de) | Verfahren zum steuern einer robotervorrichtung | |
| DE102022126904A1 (de) | Nachbarfahrspurdetektion auf der Basis von Fahrspurmodellen | |
| DE102021124680A1 (de) | Verfahren zur Erfassung von Straßenbegrenzungen durch eine Erfassungsvorrichtung eines Fahrzeugs, Erfassungsvorrichtung und Fahrzeug | |
| WO2023247356A1 (de) | Verfahren und assistenzsystem zum unterstützen einer fahrzeugführung basierend auf einem fahrschlauch und einer begrenzungsschätzung und kraftfahrzeug | |
| DE102024100647B4 (de) | Verfahren zum autonomen oder teilautonomen Führen eines Kraftfahrzeugs in Bezug auf ein externes Objekt sowie Kraftfahrzeug | |
| EP3772017A1 (de) | Bahnsignalerkennung für autonome schienenfahrzeuge | |
| DE102020208421A1 (de) | Verfahren sowie Steuergerät zum Steuern eines Kraftfahrzeugs | |
| DE102020200875A1 (de) | Verfahren zum Bereitstellen von Sensordaten durch eine Sensorik eines Fahrzeugs | |
| DE102019115788A1 (de) | Automatische Identifizierung einer Ego-Fahrspur und Anpassung einer Fahrtrajektorie für eine Ego-Fahrzeug in Abhängigkeit von der Ego-Fahrspur | |
| DE102023211080A1 (de) | Computerimplementiertes Verfahren zum Verarbeiten von Sensordaten in einem Kraftfahrzeug |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
| PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
| 17P | Request for examination filed |
Effective date: 20210521 |
|
| AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
| DAV | Request for validation of the european patent (deleted) | ||
| DAX | Request for extension of the european patent (deleted) | ||
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: EXAMINATION IS IN PROGRESS |
|
| 17Q | First examination report despatched |
Effective date: 20230419 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN |
|
| 18D | Application deemed to be withdrawn |
Effective date: 20231031 |