DE102016120066A1 - A computer implemented method for controlling an object recognition system - Google Patents

Abstract

To enhance driving safety, a computer-implemented method (V) is provided for controlling and / or testing an object recognition system (31), such as a vehicle. a driving assistance system, comprising: an object recognition method (1) and a teaching method (2) or a recognition method (2 ') performing a preprocessing method (5) for determining at least one image section (7) in which image information is contained and / or, based on at least one processing criterion (6), performing a feature extraction method (8) by determining graphic properties of the at least one image detail and merging them into at least one feature vector (11) and setting an algorithm for a classification method (20), in which the at least one feature vector (11) is assigned the underlying image information and the resulting algorithm is stored as a classifier (12, 16).

Description

The invention relates to a computer-implemented method for controlling and / or testing an object recognition system, such as e.g. a driver assistance system, and a driver assistance system for carrying out the computer-implemented method according to the preamble of claim 14.

Numerous object recognition systems are known in the prior art, especially in the field of autonomous driving. For example, from the DE 10 2011 087 791 A1 a maneuvering assistance system for vehicles, in which a situation detection can be carried out by means of an environmental sensor of the vehicle. In this case whistleblowers who describe the current situation can be perceived by means of the environmental sensor system. Objects that can serve as whistleblowers in turn are according to the DE 10 2011 087 791 A1 eg parking ticket machines, curbs, curbs, lane markings, barriers or the like. In this case, to implement a driver assistance system, computer devices with computer programs provided for control, which are based on algorithms provided specifically for the object recognition, are used. These algorithms are used to evaluate the data supplied by the sensor system, for example image data of a front camera, by comparing these with stored templates.

The object of the invention is to provide a computer-implemented method for controlling or testing an object recognition system and a driver assistance system, which can offer a higher degree of reliability and thus safety.

The object is achieved on the basis of a computer-implemented method or an assistance system of the type mentioned initially by the features of claim 1 or by the characterizing features of claim 14.

The measures mentioned in the dependent claims advantageous embodiments and modifications of the invention are possible.

The computer-implemented method according to the invention serves to control or test an object recognition system. In the object recognition system, sensory, for example, with a camera, situations or sections of an environment of a region to be monitored are detected and evaluated in such a way that individual objects are extracted or recognized therefrom. The objects to be recognized are relevant for assessing the corresponding situations. In road traffic this can be e.g. to pedestrians, traffic signs, road markings or the like, which may require a change or adjustment of the driving behavior. If such an object has been detected, a corresponding output of the object recognition system can take place; If the object recognition system is, for example, a driver assistance system, warnings to the driver or else interference in the driving behavior can be carried out, for example a braking maneuver, switching on the ABS, etc.

With the computer-implemented method according to the invention, such an object recognition system can be tested, i. be checked or taught; but in principle it can also be used to control it during operation, i. be used to control or regulate the object recognition system.

First of all, two image data sets are provided, namely a first and a second image data set. Both contain at least one image, each image being based on image information. The underlying image information may include, for example, in the image contained, relevant to the detection of the situation information, so if, for example, a traffic sign or a pedestrian can be seen on the image. However, it is also conceivable that the image contains no relevant image information, i. the picture information underlying the picture is thus the empty set. Unlike the first image data set, the second image data set contains at least one of the underlying image information, i. In the second image data set, the information which image information is relevant and what it means is also explicitly and readable.

For performing the computer-implemented method or for carrying out sub-methods comprised therein, a computing unit is provided. This arithmetic unit contains i.d.R. a processor and / or microcontroller, an FPGA (abbr. field programmable gate array) a graphics card or the like.

A sub-method of the method to be carried out is the object recognition method. In this object recognition method, the second image data record is used, which also contains the underlying image information or at least one of the underlying image information. The second image data record can advantageously be configured very informatively, so that as many reference data as possible are available to the object recognition system later in order to be able to carry out reliable and reliable object recognition.

First, within the object recognition method, a preprocessing method is carried out, in which at least one image detail is determined in which image information is or can be contained. In order to operate this evaluation within the preprocessing method, at least one processing criterion is necessary, which is applied to the corresponding image of the second image data set in order to be able to determine the at least one image detail. This image section is also referred to as Region of Interest (ROI). With this pre-processing, a data reduction can take place, ie the large amount of possibly existing information is limited to the areas in which potentially an object relevant for the object recognition system can be present.

Such preprocessing criteria with which the relevant image sections are determined in the preprocessing method can, for example, work in such a way that an edge detection, e.g. with the so-called canny-edge method and e.g. in addition, a circle recognition by means of Hough transformation takes place. In this way, a particular road sign (e.g., for speed limit) can be recognized. Using the Canny Edge method, only the outlines and edges are filtered out of the picture. However, this filtering alone is not enough, since in a picture of a typical street here are still some objects are shown, which may be of little relevance, for example. The edges in the environment of existing houses, trees, shrubs or the like. However, if e.g. Searching for round traffic signs, a round object can be recognized with the Hough transformation.

As a next sub-method, a feature extraction is then carried out by determining graphic properties of the at least one image section or the region of interest and combining them into at least one feature vector. This can be done in different ways. The region of interest may be e.g. again divided into individual sections. It is conceivable to represent and depict the color gradient along certain directions. In particular, e.g. for each pixel, the gradients are determined and evaluated. In this sub-method, one or more feature vectors are to be generated, with which it can be determined later on whether the particular region of interest or the corresponding image section actually contains a relevant object with situation-important information or not.

In order to establish the still missing bridge between the feature vector and the underlying image information, an algorithm for a classification method is now set or adjusted in which the at least one feature vector is assigned the underlying image information and the resulting algorithm is stored as a classifier. If, as in the present case, the second image data set is used, this underlying image information is already known. The feature vector contains a collection of properties of the image detail that were previously obtained in the feature extraction method, for example, the color arrangements in spatial resolution in certain directions, histograms of the gradient directions, etc. All of this information is summarized in the feature describing or image detailing feature vector.

In order to set the algorithm, it is still necessary to allocate what in the present case this feature vector or this summary of image detail information means. For example, In other words, in the case of a certain feature vector, if appropriate within a certain range to be tolerated, the searched object is a specific traffic sign. Thus, via the classifier, the feature vector becomes a label, i. associated with the underlying image information.

Possibly. the described object recognition method is performed again with another image of the second image data set. This is particularly advantageous if, on the one hand, the image already read in did not contain all possibly relevant objects that are to be recognized in principle. It should also be noted that the same object may be used in different situations, e.g. at different daylight levels, precipitation, recorded under other angles or under other environmental conditions, also graphically not always shown in the image is consistent. Nevertheless, as a result, the object should always be recognized as such. In this way, the resulting algorithm or the classifier may still be adapted or modified.

As a further sub-method, a recognition method is carried out according to the invention, in which now at least one of the pictures of the first picture data record is read into the arithmetic unit. This image of the first image data set thus does not contain the underlying image information. Analogous to the object recognition method, a preprocessing method is carried out for determining at least one image section in which image information is contained or may be contained. For this purpose, the processing criterion used is the same as that which has already been used in the previously performed object recognition method.

If a picture or a region of interest has been filtered out, the feature extraction method is carried out accordingly by determining graphic properties of the picture detail and in turn combining them into at least one feature vector.

Now, the classifier already created in the object recognition method is used in the so-called assignment method, in which the result image information is assigned to the at least one feature vector on the basis of the classifier. Subsequently, in principle, the recognition method can be carried out with a further image of the first image data set.

After performing the object recognition process, the basis for the object recognition is created.

For testing the object recognition system, in an advantageous development of the invention, a learning procedure can be carried out in connection with the recognition method, wherein a comparison of the at least one result image information from the recognition process with the underlying image information of the first image data set is carried out and wherein the image information of the first image data set is, for example determined and compared via a separate test unit. The algorithm can thus be adjusted or trained via the learning procedure. If the correct or matching image information has been assigned, no change of the resulting algorithm or classifier is necessary. However, if the result image information does not match the underlying image information, a corresponding change of the classifier is necessary and is performed accordingly.

Without further testing, the recognition process can be performed for object recognition.

It is also conceivable that the evaluated image contains no information. In this case the procedure is analogous. The result image information is thus an empty set or a value that expresses that no information was found. If this agrees with the underlying image information, then the classifier need not be changed, otherwise an adjustment is necessary.

In particular, the computer-implemented method according to the invention has the advantage that, in comparison to previously known systems, it involves less effort in order to capture the objects to be detected to a sufficient extent. It can thus be reduced in an advantageous manner costs. The reason for this is, in particular, that a small number of images must be identified with the underlying image information and much of it can be done by providing essential features of the objects to be recognized, so to speak, through teaching. A corresponding computer-implemented method according to the invention can basically be used for various applications. However, as already stated above, it is particularly suitable in combination with driver assistance systems, especially in connection with systems for autonomous or semi-autonomous driving. In such an embodiment, the object recognition method may be used to recognize in particular traffic signs and / pedestrians as objects. The traffic signs can basically be signs, but also barriers, barriers, pylons or lane markings.

A significant advantage of the computer-implemented method according to the invention is that the cost can be substantially reduced by this training. In order to realize this, a machine-learning method can be used, i. an artificial generation of knowledge from experience is used. During the object recognition process, the computer-implemented method learns from examples of the second image data set and recognizes patterns and laws in the learning data that is still unknown in the recognition or learning process. It is therefore a supervised or partially supervised learning. In contrast to previously known methods, therefore, not only pre-labeled images or situations can be detected on the basis of given templates.

To reduce the amount of data to be processed and thus u.a. In order to be able to save time and computing power as well, an edge detection can be carried out in the preprocessing method, in particular by means of canny edge detection or circle recognition by means of Hough transformation. The Canny Edge process performs a convolution that leaves only the edges of the image being processed left over. This also allows objects to be reduced to their recognizable features. Geometric shapes can be recognized by means of the Hough transformation; in particular, circle recognition for round traffic signs can be performed. In this way, simplified and automated recognition-relevant features within an image that identify an object can be detected.

In embodiments of the invention, in the feature extraction method, for example, a histogram of orientated gradients may be created to obtain image-describing information or an image-describing feature vector. The starting point is a selected image section, ie a corresponding region of interest. This image section is rasterized into cells, for example. For each cell, the strongest gradient direction should be determined. Furthermore, the gradient is calculated for each pixel. Subsequently, the gradients of each cell are discretized and the gradient amounts are summed, resulting in histograms of the gradient directions (Histogram of Oriented Gradients, HOG). All of these histograms are combined into a feature vector describing the image section.

For classification, the object recognition method can also use a support vector machine (SVM) method whereby a set of objects is subdivided into classes so that the widest possible area remains free of objects around the class boundaries. The classification then takes place via the division of separation areas, e.g. a hyperplane can be fitted to serve as the interface between the classes.

In embodiments, in the area of the machine learning method, a deep learning method can be used, in which, so to speak, artificial neural networks are generated in order to build algorithms and learn from abstractions and also to be able to process unknown input data.

In order to be able to obtain a better determination of the strongest gradient direction, in the feature extraction method a screening of the image section into at least two cells can be carried out, each cell in particular having the size 8 × 8 pixels. Such a size is particularly advantageous in terms of processing.

As already stated, in a development of the invention, a local distribution of the gradient orientations or a determination of the gradient strength can be used as graphical properties in the feature extraction method. In this way, image structures can be filtered out in a characteristic manner and compared in a reliable manner, so that a simplified evaluation and assignment of the image information becomes possible.

It has already been shown above that in one embodiment of the invention a support vector machine can be used to determine the classifier. Accordingly, in one embodiment of the invention in the classification method the classifier at least one hyperplane can be assigned, which separates the feature vectors of feature vectors associated with a specific image information, which are assigned to another image information. The summarized properties, which are in the form of a feature vector, can be entered as points in a generally multi-dimensional space. Ideally, an object would always have the same properties, so basically an object could be identified by a single point. But since the individual images are recorded in different situations (different ambient brightness, different viewing angles, soiling of the object, etc.) i.d.R. not just a single point per object, but the points are somewhat scattered in this generally multi-dimensional space. However, the properties of another object or its image properties are different from those of the aforementioned object in that it forms a different collection of points within the generally multi-dimensional space. The respective accumulations of points can be separated by a hyperplane. Mathematically, this problem can be considered as an optimization problem, e.g. Solved with the Lagrange method can be solved. By means of such a hyperplane, multi-class problems can advantageously be solved, whereby mathematically a clean classification into individual object classes, e.g. a division into several different speed limit signs can be made. The speed limit signs are similar to e.g. basically in their geometry and their structure, only the numbers for indicating the maximum speed are different. If this image-describing information is combined in a feature vector, the individual different speed limit signs can be reliably distinguished accordingly in this geometric representation by hyperplanes. Accordingly, such an SVM method or a separation by hyperplanes in space can also be used in an assignment method. In particular, an optimization method can be used.

Since the computer-implemented invention enables computer-aided automation of object recognition, it is particularly suitable for use in driver assistance systems of vehicles. In this case, one or more sections of the surroundings of the vehicle can be recorded accordingly and be available as an image data record. With these pictures, e.g. while driving, the detection process can be performed. The recording of the image may e.g. in the direction of travel; however, rear or side areas of the vehicle can also be accommodated if these are relevant for driving, e.g. when maneuvering.

In a further development of the invention, the driver assistance system can be coupled as such with a control unit for controlling the vehicle. As a result, for example, commands to the system controlling the vehicle are delivered. Furthermore, basically driving data can be provided with the method, if necessary also for display to the driver via the dashboard, a head-up display, by loudspeaker announcement or the like. Furthermore, corresponding data or result image information can also be transmitted to the control unit as a control or regulation parameter. For example. It is conceivable that a particular road sign is detected, and this information is displayed to the driver (in addition), so that this does not overlook the changed speed specifications on the road.

It is also conceivable, however, to take another step towards autonomous driving, that this information is immediately implemented by the control unit and the vehicle adapts its speed accordingly. In particular, when pedestrians are detected on the road, it is conceivable that the control unit does not first notify the driver and wait, but automatically initiates a braking operation. It is also conceivable, however, that appropriate information be used to make adjustments in the vehicle, for example to change driver assistance programs, ABS or the like to different situations, for example. In difference between the driving environment outside a village or within a town.

In a particularly preferred development of the invention, a test unit is used which contains or determines comparison information from the images of the first image data set and performs a comparison between the at least one result image information from the recognition process and the comparison information. In this way, a partially supervised learning can be realized, i. a separate test unit is used to verify the recognition processes learned during the object recognition process during the recognition process. In this way, a higher level of security can be ensured. In this way, only the information-rich images can now be additionally tested, while already in the object recognition process a corresponding basis for the detection of such situations has been created. As a rule, it suffices to render the images of the second image data set, which can be less informative and contain less recognizable objects in the images, with image information that is specified in the object recognition process for later recognition. As a result, in particular the complexity and the cost of the system can be reduced. In addition, time is saved, because an at least partially independent learning is possible.

In one embodiment of the invention, a test unit may be used, which comprises a measurement device for measuring or gathering driving data while the vehicle is traveling, which uses measured driving data as comparison information and a comparison between the at least one result data. Image information and the recognition method and the comparison information is performed. In this way, the test unit introduces a "second instance" for checking the object recognition system, which can provide a higher level of security.

Accordingly, the above-mentioned advantages of the computer-implemented method according to the invention or the embodiments of the invention can be used in a driver assistance system. In one embodiment of the invention, an interface or a memory unit for reading, reading or storing images, image information or classifiers can be provided so that they are permanently used and used during the journey, but may also be available for review in the workshop ,

A vehicle may directly have a control unit for controlling the vehicle, for example via an on-board computer, which is coupled to corresponding driver assistance systems, in particular an electronic system which controls or regulates the engine and acts on the driving characteristics of the vehicle.

An embodiment of such a driver assistance system according to the invention may comprise at least one sensor for measuring or gathering driving data. This may, for example, be a camera. But it can also be provided other sensors, for example. Sensors for detecting the precipitation, so that this information can be used in turn in the image analysis. For example. can be expected at appropriate precipitation with visual disabilities, which are taken into account when corresponding objects such as pedestrians, traffic signs or the like to be detected.

embodiments

Embodiments of the invention are illustrated in the drawings and are explained in more detail below, giving further details and advantages.

• 1: eine schematische Darstellung des Objekterkennungsverfahrens bzw. des Erkennungsverfahrens,
• 2: ein Schema des gesamten computer-implementierten Verfahrens gem. der Erfindung zur Kontrolle bzw. zum Testen eines Objekterkennungssystems,
• 3: eine schematische Darstellung des Klassifikationsverfahrens, sowie
• 4: eine schematische Darstellung eines Kraftfahrzeugs mit einem Fahrassistenzsystem gem. der Erfindung.

In detail show:

• 1 : a schematic representation of the object recognition method or the recognition method,
• 2 a scheme of the entire computer implemented method acc. the invention for controlling or testing an object recognition system,
• 3 : a schematic representation of the classification method, as well
• 4 : A schematic representation of a motor vehicle with a driving assistance system acc. the invention.

1 shows a schematic representation of the object recognition process 1 or the learning procedure 2 , Part of the learning process 2 is the recognition method 2 ' (see. 2 ). First, a picture 3 read in, for example, was taken by a camera of a driver assistance system in a vehicle. This image contains various image information and shows, for example, the course of the road, a house and a road sign 4 namely a speed limit of 30 km / h. First, here is a preprocessing process 5 carried out, including at least one processing criterion 6 , This will be a picture section 7 determines, namely in this case the road sign 4 , For this purpose, first the preprocessing process 5 an edge detection by means of Canny edge detection performed, ie the original color image 3 is first converted so that only the edges of the objects imaged thereon can still be seen. Thereafter, a circle detection is performed by Hough transformation, which allows the speed limit in the form of the round shield 4 to recognize and separate.

If, in the present case, this circle recognition does not lead to a result, for example, other geometric shapes could also be determined, eg triangles, diamonds or the like (eg priority roadsigns, right of way signs, etc.). Subsequently, the feature extraction method 8th carried out. This first comprises a screening in cells, in which the image section 7 is divided. Subsequently, the strongest gradient direction is determined for each cell, the gradient is calculated for each pixel, and then the gradients of each cell are discretized and the gradient magnitudes summed up. The summation of these gradient amounts finally gives histograms of the gradient directions. These formations of the histograms of the gradient directions, which together form an image-describing feature vector, are denoted by the reference symbol 10 described. The result of the feature extraction method 8th is thus at least one feature vector 11 ,

When using the recognition method, the feature vector II becomes a classifier 12 to a result or a label 13 containing a result image information; ie it becomes the assignment procedure Z applied. In carrying out the object recognition process 1 the algorithm underlying the classifier is adapted on the basis of image information. This additionally provided in the object recognition method image information is dashed with the reference numeral 14 characterized.

2 shows a schematic of the entire computer-implemented method V to control or test an object recognition system. The starting point is the second image data set B2 from which an image is read into the arithmetic unit (eg image 3 ). First of all, the object recognition method is applied to this picture 1 applied. If several pictures are read in, then this repetition loop becomes 15 applied to this. The result is a resulting algorithm in the form of a classifier 16 who then goes for the learning procedure 2 or the detection method 2 ' is used. Also as input for the learning process 2 serve images of the first image data set B1 , With these two input information, namely the images from the first image data set without image information and the classifier 16 , becomes the detection method 2 carried out.

If there are several pictures from the first picture data set, this will be a repeat loop 17 executed so that they can be processed one after the other. The result image information 18 Finally, by means of a test unit T compared with comparative information.

3 shows an implementation of a classification method 20 using a classifier using a Support Vector Machine. The individual feature vectors can be identified by dots 21 respectively. 22 in a room 23 , which is usually multidimensional, can be represented. The points 21 and 22 For example, represent different classes, for example, different speed limit signs. Due to the different environments in which pictures are taken in a traffic situation, these respective feature vectors do not completely match each other, resulting in a collection of dots each 21 respectively. 22 comes. The points 21 however, which stand for a particular road sign, are close to each other and are clearly separable from the accumulation of points 22 , This separation is made by a hyperplane 24 , This solves an optimization problem. The dashed lines 25 and 26 are just chosen to have extreme points like the points 21 ' and 22 ' just recording. Through mathematical optimization can be over the hyperplane 24 Thus, a distinction of the road signs are made, each by the point accumulations 21 and 22 being represented.

4 shows a motor vehicle 30 with a driver assistance system 31 , The sensor includes this driver assistance system 31 first a camera 32 as well as a computing unit 33 , In the arithmetic unit 33 become the object recognition method 1 as well as the recognition method 2 or the complete computer-implemented method V carried out. About the exit 34 For example, the result image information can be displayed directly to the driver via a head-up display 35. In addition, over the output 36 the corresponding information also to a control unit 37 of the motor vehicle 30 be forwarded so that they may intervene in the ride, for example, controls the anti-lock braking system or initiates a braking operation when a dangerous situation exists.

LIST OF REFERENCE NUMBERS

1

Object detection method

2

learning method

2 '

detection methods

3

original camera image

4

road sign

5

preprocessing

6

Vorverarbeitungskriterium

7

image section

8th

Feature extraction method

9

Zellenrasterung

10

Creation of the feature vector with HOG

11

feature vector

12

classifier

13

label

14

image information

15

repeat loop

16

classifier

17

repeat loop

18

Result image information

20

classification methods

21

feature vectors

22

feature vectors

23

Multidimensional space

24

hyperplane

25

parting plane

26

parting plane

21 '

boundary point

22 '

boundary point

30

motor vehicle

31

Driving Assistance System

32

camera

33

computer unit

34

output

35

Head-Up Display

36

output

37

control unit

B1

First image data set

B2

Second image data set

T

test unit

V

Computer-implemented method

Z

allocation method

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 102011087791 A1 [0002]

Claims (18)

A computer implemented method (V) for controlling and / or testing an object recognition system (31), e.g. a driver assistance system comprising: Providing a first image data set (B1) with at least one image, wherein the respective image is based on image information, Providing a second image data set (B2) with at least one image, wherein the respective image is based on image information and wherein the image data set contains at least one of the underlying image information, Providing a computing unit (33) for carrying out the following sub-procedures, - Implementation of an object recognition method (1) with at least the following method steps: Reading at least one of the images of the second image data set (B2) into the arithmetic unit (33), Carrying out a preprocessing method (5) for determining at least one image detail (7) in which image information is contained and / or can be contained, based on at least one processing criterion (6), Carrying out a feature extraction method (8) by determining graphic properties of the at least one image section and combining them into at least one feature vector (11), Setting an algorithm for a classification method (20), in which the at least one feature vector (11) is assigned the underlying image information and the resulting algorithm is stored as a classifier (12, 16), In particular renewed execution (15) of the object recognition process with another of the images of the second image data set (B2), • Carrying out a recognition process (2 '), with at least the following process steps: o reading at least one of the images of the first image data set (B1) into the arithmetic unit (33), carrying out the preprocessing method (5) for determining at least one image detail in which image information is contained and / or can be contained, based on the at least one processing criterion (6) used in the object recognition method (1), carrying out the feature extraction method (8) by determining graphic properties of the image detail and combining them into at least one feature vector (11), o carrying out an assignment method (Z) in which the result image information is assigned to the at least one feature vector (11) on the basis of the classifier (12, 16), o in particular renewed execution (17) of the recognition process (1) with another of the images of the first image data set (B1), Computer-implemented method (V) according to one of the preceding claims, characterized in that a teaching process (2) is performed, wherein a comparison of the at least one result image information from the recognition process (2 ') with the underlying image information of the first image data set ( B1), and a change of the classifier (12, 16) is performed if, in the comparison, one of the result image information does not match the underlying image information. Computer-implemented method (V) according to one of the preceding claims, characterized in that the object recognition method (1) is used to detect traffic signs (4) and / or pedestrians as objects. Computer-implemented method (V) according to any one of the preceding claims, characterized in that as a learning method (2), a machine-learning method is performed. Computer-implemented method (V) according to one of the preceding claims, characterized in that in the preprocessing method (5) as processing criterion (6) or processing criteria at least one of the following criteria is used: - an edge detection, in particular by means of canny edge detection and / or - a circle recognition by Hough transformation. Computer-implemented method (V) according to one of the preceding claims, characterized in that in the feature extraction method (8) a screening (9) of the image section is performed in at least two cells, each cell in particular having the size 8x8 pixels. Computer-implemented method (V) according to any one of the preceding claims, characterized in that the feature extraction method (8) as a graphical properties a local distribution of the gradient orientations and / or a gradient strength is determined. Computer-implemented method (V) according to one of the preceding claims, characterized in that in the classification method (20) the classifier (12, 16) is assigned at least one hyperplane (24) which maps the feature vectors (11, 21, 22) of feature vectors to be associated with another image information. Computer-implemented method (V) according to one of the preceding claims, characterized in that in the assignment method a Optimization method is applied to make the assignment. Computer-implemented method (V) according to one of the preceding claims, characterized in that the method is carried out in a driver assistance system (31) of a vehicle (30), wherein a picture of at least a portion of the environment of the vehicle (30) and subsequently the recognition process (2 ') is performed. Computer-implemented method (V) according to one of the preceding claims, characterized in that as a driving assistance system (31) such with a control unit (37) for controlling the vehicle (30), in particular for issuing commands to the vehicle (30) controlling Systems and / or for providing driving data and / or for displaying driving data to the driver, is used and the result image information is transmitted as control and / or control parameters to the control unit (37) Computer-implemented method (V) according to one of the preceding claims, characterized in that a test unit (T) is used, which contains comparison information, and performs a comparison between the at least one result image information from the recognition method (2 ') and the comparison information , Computer-implemented method (V) according to any one of the preceding claims, characterized in that a test unit (T) is used, which comprises a measuring and / or detection device for measuring and / or detecting driving data while driving the vehicle, the measured and / or detected driving data used as comparison information, and performs a comparison between the at least one result image information from the recognition method (2 ') and the comparison information. Driver assistance system (31) for assisting control of a vehicle (30) by outputting driving data, driving instructions and / or by performing controls and / or regulations of the vehicle for carrying out a computer-implemented method (V) according to one of the preceding claims. Driver assistance system (31) according to any one of the preceding claims, characterized in that an interface and / or a memory unit for reading, reading and / or storing images, image information and / or classifiers is provided. Driver assistance system (31) according to one of the preceding claims, characterized in that a control unit (37) for controlling the vehicle (30) is provided. Driver assistance system (31) according to one of the preceding claims, characterized in that at least one sensor (32) is provided for measuring and / or detecting driving data. Driver assistance system (31) according to any one of the preceding claims, characterized in that a test unit (T) is provided to make a comparison between the at least one result image information from the recognition method (2 ') and the comparison information.

Images