DE102017218430A1 - Method for identifying parameterizable geometric curves - Google Patents

Abstract

A method of recognizing parametrisable geometric curves having at least a first parameter and a second parameter, comprising the steps of: providing at least one edge image with pixels (M1); Setting a set of curves to be recognized in the edge image with at least a first parameter and a second parameter (M2); Performing a Hough transformation for several pixels of the edge image into a Hough space (M3); Detecting a curve candidate if a value at a point in the Hough space is greater than a predetermined first threshold (M4.1) and / or recognizing a curve candidate if a value at a point in the Hough space is greater than a second predetermined threshold and Sum of values in a specified neighborhood of the point is greater than a third threshold, the second threshold being less than the first threshold and the third threshold being at least equal to the first threshold (M4.2); Grouping curve candidates into a group if the curve candidates have at least one parameter that deviates no more than a predetermined tolerance value from the parameter of another curve candidate (M5); Estimate a curve using the grouped curve candidates (M9).

Description

FIELD OF THE INVENTION

The present invention relates to a method for the detection of parametrisierbaren geometric curves and a driver assistance system.

TECHNICAL BACKGROUND

The US Pat. No. 6,807,286 B1 shows a method for detecting objects in image data from surveillance cameras.

SUMMARY OF THE INVENTION

Against this background, the invention has the object to provide a real-time capable method for the detection of parametrisierbaren geometric curves.

"Real time" means that the process is simultaneous with reality.

Parameterizable curves are curves or surfaces that can be described as a function of one or more parameters.

According to the invention, this object is achieved by a method having the features of patent claim 1 and by a driver assistance system having the features of patent claim 17.

• - ein Verfahren zur Erkennung von parametrisierbaren geometrischen Kurven mit wenigstens einem ersten Parameter und einem zweiten Parameter mit den folgenden Schritten:
  • Bereitstellen wenigstens eines Kantenbildes mit Pixeln;
  • Festlegen einer Menge von in dem Kantenbild zu erkennenden Kurven mit wenigstens einem ersten Parameter und einem zweiten Parameter;
  • Durchführen einer Houghtransformation für mehrere Pixel des Kantenbildes in einen Houghraum;
  • Erkennen eines Kurvenkandidaten, wenn ein Wert an einem Punkt in dem Houghraum größer ist als ein festgelegter erster Schwellwert und/oder Erkennen eines Kurvenkandidaten, wenn ein Wert an einem Punkt in dem Houghraum größer ist als ein zweiter festgelegter Schwellwert und eine Summe von Werten in einer festgelegten Nachbarschaft des Punktes größer ist als ein dritter Schwellwert, wobei der zweite Schwellwert kleiner ist als der erste Schwellwert und der dritte Schwellwert wenigstens gleichgroß wie der erste Schwellwert ist;
  • Gruppieren von Kurvenkandidaten zu einer Gruppe, wenn die Kurvenkandidaten wenigstens einen Parameter aufweisen, der nicht mehr als einen vorbestimmten Toleranzwert von dem Parameter eines weiteren Kurvenkandidaten abweicht;
  • Schätzen einer Kurve mittels der gruppierten Kurvenkandidaten; sowie
• - ein Fahrerassistenzsystem mit einem Bildsensor, einer Datenspeichereinheit, einer Recheneinheit sowie einer Kommunikationseinheit, wobei das Fahrerassistenzsystem eingerichtet ist, Verkehrsschilder zu erkennen, indem Bilddaten des Bildsensors von der Recheneinheit zu einem Kantenbild verarbeitet werden und Konturen von Verkehrsschildern in dem Kantenbild mittels einer modifizierten Houghtransformation des Kantenbildes erkannt werden, wobei die Kommunikationseinheit eingerichtet ist, eine Meldung über ein Verkehrsschild auszugeben, nachdem ein Verkehrsschild erkannt wurde.

Accordingly, it is provided:

• a method for recognizing parameterizable geometric curves having at least a first parameter and a second parameter, comprising the following steps:
  • Providing at least one edge image with pixels;
  • Defining a set of curves to be recognized in the edge image with at least a first parameter and a second parameter;
  • Performing a Hough transformation for several pixels of the edge image into a Hough space;
  • Detecting a curve candidate if a value at a point in the Hough space is greater than a predetermined first threshold and / or detecting a curve candidate if a value at a point in the Hough space is greater than a second predetermined threshold and a sum of values in one predetermined neighborhood of the point is greater than a third threshold, wherein the second threshold is less than the first threshold and the third threshold is at least equal to the first threshold;
  • Grouping curve candidates into a group if the curve candidates have at least one parameter that deviates no more than a predetermined tolerance value from the parameter of another curve candidate;
  • Estimating a curve using the grouped curve candidates; such as
• a driver assistance system having an image sensor, a data storage unit, a computing unit and a communication unit, wherein the driver assistance system is configured to detect traffic signs by processing image data of the image sensor from the arithmetic unit into an edge image and contours of traffic signs in the edge image by means of a modified Hough transformation of the Edge image are detected, wherein the communication unit is adapted to issue a message on a traffic sign after a traffic sign has been detected.

An edge image is calculated from a digital image by so-called edge detection by analyzing adjacent pixels (pixels) for changes in their color values or gray values.

The Hough transformation is a method of recognizing parametrisable geometric figures in an edge image. In order to detect geometric objects, a Hough space is created in which possible parameters of the curve to be recognized are entered in the Hough space for each point in the edge image which lies on one edge. Each point in the Hough space thus corresponds to a geometric object in the edge image. In a straight line z. As the slope or the y-axis section, in a circle the center and the radius. If there is an accumulation at one point in the Hough space, a corresponding curve can be recognized in the edge image.

A neighborhood of a point in this application means an environment of the point containing points whose distance to the point is less than a predetermined value.

A curve candidate is a curve that is determined during the process as a possible parameterized representation of a geometric object in the edge image. Typically, the method determines a plurality of curve candidates for representing a geometric object in the edge image and selects a curve from the curve candidates.

Color value difference is the distance between two color values. Color values can be express as at most three-dimensional vectors. In an edge image, the color value is often designed as a gray value.

Thus, the method according to the invention recognizes a parameterisable geometric object in a camera recording starting from an edge image of the camera recording. It determines which curves are visible in the edge image.

The set amount of the curves to be recognized may include, for example, all circles or lines in the edge image. Based on the set amount of curves to be recognized, a Hough transform is applied to the edge image. The Hough transformation can vary for different set amounts. For example, Hough transformations are different for lines, circles, and ellipses. The result of the Hough transformation or the Hough space is often referred to as an accumulator and is a matrix whose dimension corresponds to that of the parameter combination. For example, for a circle of variable center and variable radius to be detected, the Hough space is a three-dimensional matrix. For a circle of fixed radius to be recognized, the Hough space is a two-dimensional matrix.

The result rough space is a matrix in which all parameter combinations that can lead to an edge pixel are entered. If the pixels in the edge image represent a curve with a certain combination of parameters, an increased value is entered in the corresponding location in the matrix, since there many pixels of the curve for the parameter combination vote.

Accordingly, the curve is recognized with this combination of parameters if the corresponding value in the matrix exceeds a specified threshold.

However, for example, image noise or defects in the camera image can lead to points for a certain combination of parameters being distributed over several similar parameter combinations. Therefore, the inventive method also considers the sum of the matrix entries in a predetermined neighborhood of a parameter combination.

Frequently, several curve candidates are detected for an object of a camera recording to be recognized. This can result, for example, from the fact that an object has several contour lines that are close together in the camera image. In order to assign an object in a camera recording to several curve candidates, curve candidates with similar parameter combinations are grouped in a group. For grouping, cluster algorithms can be used.

For example, it can be provided that all curve candidates with a parameter that deviates from a parameter of another curve candidate no more than a specified value are grouped into a group. Alternatively it can be provided that a new group is created if a parameter of a curve candidate deviates by more than a fixed value from the average of the parameter of the existing group.

Based on the group of curve candidates, a parameter combination of a curve representing the object in the camera image is estimated. For the estimation, different estimation methods, e.g. Means or weighted averaging of the parameters. Alternatively, the group of curve candidates can be further selected.

A message from a communication unit of a driver assistance system can be directed, for example, to a driver or to a control unit which controls a vehicle. Thus, the message can be configured for example as a warning signal or as a control signal.

Advantageous embodiments and further developments will become apparent from the other dependent claims and from the description with reference to the figures of the drawing.

According to a preferred embodiment of the invention, the method is configured to detect circles, wherein the first parameter is designed as a radius and the second parameter as the center. By restricting the method to circles, computing time or arithmetic operations can be saved.

Furthermore, circling is particularly relevant in some real-time applications. For example, in driver assistance systems, it is desirable to detect traffic signs, in particular round traffic signs, since these, for example, specify a maximum speed to be maintained or inform about the cancellation of a maximum speed to be maintained.

According to a preferred development of the invention, the method further comprises the following steps: forming a color value difference between at least two points of a curve candidate for a plurality of curve candidates of the group; Determining at least one parameter of a curve candidate having the least chrominance difference; Estimating at least one curve by means of the at least one determined parameter.

Thus, the group of curve candidates is further selected to further enhance the estimation of the parameters of the curve representing the object. Thus, one or more parameters can be estimated more accurately, especially if the averaging of the parameters of curve candidates of a group leads to an unsatisfactory result.

The determination of a parameter of a curve candidate with the lowest color value difference is based on the assumption that pixels lying on a curve have an identical color value. Accordingly, it is possible to check whether a curve candidate matches an object globally or at certain points.

It is understood that it is also possible to sum color value differences over several points of a curve candidate and to subsequently determine the parameter of a curve candidate with the lowest sum of the color value differences.

According to a preferred embodiment of the invention, the color value difference is designed as a gray value difference. Edge images are often stored in a black-and-white representation or in a black-and-white-gray representation. In this case, arithmetic operations can be saved if a gray value difference is determined instead of a color value difference of fully occupied color vectors. Full color values are three-dimensional vectors, whereas gray values are a number.

It is expedient if a color value difference between two symmetrical, in particular point-symmetrical points is formed. If the color value difference is formed for a circular line, the points of the color value difference can be, for example, opposite points of a circular line with respect to a circle diagonal.

Point symmetry is preserved in many camera shots, even if the camera shots are distorted. For example, in the automotive sector, distorted camera shots are frequently used, in which point symmetries are preserved.

According to a preferred embodiment of the invention, an average value for the first and / or for the second parameter for curve candidates of the group is determined. In many cases, the average value of a parameter of curve candidates of a group is already a sufficiently good estimate.

However, in methods intended to meet increased safety requirements, it is still appropriate to specify a parameter estimate by determining a color value difference for curve candidates having a parameter that is within a predetermined search range around the average value of the parameter of the group.

This embodiment of the invention is based on the assumption that the best estimate is frequently found in an environment of the average value of a parameter of a group.

Thus, the result is further refined compared to averaged parameters, saving computing power compared to determining color value differences for each curve candidate of a group.

In this case, the predetermined search range can be determined by the average of the parameter application-specific. Typically, an optimal predetermined search range grows with the scatter of the curve candidates around a curve to be determined.

It is further advantageous if several edge images are provided with different resolutions and, depending on the average of the parameter, a resolution of the edge image from which the color value difference is formed is selected.

Often, systems that perform methods of detecting objects in image data process camera shots in multiple resolutions. Thus, this additional information already present in a system can also be used to improve the accuracy of the curve parameterization detection method. It is expedient to choose a high resolution if a surface-related parameter is small, whereas for larger objects a lower resolution may be sufficient. Thus, memory and arithmetic operations can be reduced with almost constant accuracy of the method.

Optimal resolutions for area-related parameters can be stored in a table. Area-related parameters are parameters that influence the limited area of a curve.

It is advantageous if the color value difference is determined for two subsections of a curve candidate. In this way, the computational effort compared to a global comparison of the color values of a curve can be reduced.

According to a preferred development of the invention, a parameter range to which the first and / or the second parameter of the curve to be detected is restricted is defined. Thus, previously known knowledge, for example about an approximate position or size of an object to be recognized, can be used to save arithmetic operations.

In a method that is set up to detect traffic signs, it may be provided, for example, that only contours of signs are searched for in a specific area of a camera recording or, for example, a radius is restricted in advance to a specific area.

It is advantageous if a resolution of the edge image, for which the Hough transformation is performed, is selected as a function of the specified parameter range.

Analogous to the choice of the resolution of an edge image for determining the color value difference, the accuracy of the method can be further improved if, for small area-related parameters, a large resolution of an edge image is selected for performing the Hough transformation. Thus, a more accurate result can be achieved for the Houghraum.

According to a preferred development of the invention, a first curve candidate with a highest value of the parameter among the curve candidates of the group is determined for at least one first parameter, and the parameter of the curve to be recognized is estimated by means of the first parameter of the first curve candidate. For clarity, the first curve candidate is hereinafter referred to as the "maximum curve candidate".

An estimate based on a maximum value of a parameter of a group of curve candidates is advantageous in the estimation of area-related parameters. In this way, no information is truncated if an area-related parameter is underestimated.

It is also advantageous to determine a second curve candidate (touching curve candidate) that touches the maximum curve candidate, wherein the difference of a second parameter of the maximum and the touching curve candidate is determined and the first parameter of the curve to be detected by means of the first parameter of the maximum Curve candidates and the difference of the second parameter of the maximum and the touching curve candidate is estimated.

This refinement of the method is based on the assumption that a curve to be recognized touches the maximum curve candidate of a group. The estimate of the area-related parameter can thus be further improved by comparing a second parameter, for example a location-related parameter, of the maximum curve candidate with the corresponding parameter of the touching curve candidate. The first parameter of the curve is then more accurately estimated by means of the maximum value of the first parameter and the difference of the other parameters between the maximum and the touching curve candidate.

For example, it can be provided that the first (surface-related) parameter is designed as a radius of a circle or as a half-axis length of an ellipse and the second (location-related) parameter is formed as the center of the circle or the ellipse. If a curve candidate with the largest radius of a group of curve candidates (maximum curve candidate) and a touching curve candidate is determined, the radius of the circle can be estimated more precisely by subtracting the centers of the two curve candidates by subtracting the distance of the center points from the radius of the maximum curve candidate becomes.

list of figures

• 1 ein schematisches Blockbild gemäß einer Ausführungsform der Erfindung;
• 2 eine schematische Ansicht eines Verfahrensschrittes einer Ausführungsform der Erfindung;
• 3 eine schematische Ansicht eines Verfahrensschrittes einer Ausführungsform der Erfindung;
• 4 eine schematische Ansicht eines Verfahrensschrittes einer Ausführungsform der Erfindung;
• 5 ein schematisches Blockbild gemäß einer Ausführungsform der Erfindung;

The present invention will be explained in more detail with reference to the exemplary embodiments indicated in the schematic figures of the drawings. It shows:

• 1 a schematic block diagram according to an embodiment of the invention;
• 2 a schematic view of a method step of an embodiment of the invention;
• 3 a schematic view of a method step of an embodiment of the invention;
• 4 a schematic view of a method step of an embodiment of the invention;
• 5 a schematic block diagram according to an embodiment of the invention;

The accompanying drawings are intended to provide further understanding of the embodiments of the invention. They illustrate embodiments and, together with the description, serve to explain principles and concepts of the invention. Other embodiments and many of the stated advantages will become apparent with reference to the drawings. The elements of the drawings are not necessarily shown to scale to each other.

In the figures of the drawings are the same, functionally identical and equivalent elements, features and components - unless otherwise stated - each provided with the same reference numerals.

DESCRIPTION OF EMBODIMENTS

1 shows the steps M1 - M9 a method for detecting circles. Unless otherwise stated in the following description, no fixed order is provided for the sequence of processes.

M1 shows providing at least one edge image with pixels. M2 shows the setting of the amount of curves to be recognized in the edge image with a radius between 9 and 17 Pixels and any center. M3 shows performing a Hough transformation for the pixels of the edge image into a Hough space.

M4.1 Figure 11 shows the detection of a curve candidate when a value at a point in the Hough space is greater than a predetermined first threshold. M4.2 Figure 11 shows the detection of a curve candidate when a value at a point in the gross space is greater than a second predetermined threshold and a sum of values in a specified neighborhood of the point is greater than a third threshold, the second threshold being less than the first threshold and the third threshold is at least equal to the first threshold. The steps M4.1 and M.4.2 can be performed sequentially, with no order set, or concurrently.

M5 shows the grouping of curve candidates into a group when the curve candidates have a radius that deviates at most 40% from the average value of the radius of the existing group and is at most 30% of the average value of the radius from an average center point.

M6 shows the formation of a gray value difference between two circle segments which face each other with respect to a circle diagonal. In this case, the difference is formed from the gray values of two opposing pixels and the differences of all pixels of the circle segments are summed.

M7 shows the determination of the center by means of a least sum of the gray value differences.

M8.1 shows the determination of the curve candidate with the largest radius of a group of curve candidates and the determination of a curve candidate that touches the curve candidate with the largest radius. The method estimates the radius of the circle by subtracting from the maximum radius the distance between the centers of the two curve candidates.

M9 FIG. 12 shows the estimation of a curve by means of the grouped curve candidates, by the curve being determined by means of the determined parameter configuration according to the steps M8.1 for the radius and M7 is estimated for the midpoint.

The 2 shows a detailed view of the steps on 4.1 and 4.2 , In the 2 are several clippings of a result gross space 10 . 11 and 12 shown. At the Houghräumen 10 . 11 and 12 each are two-dimensional sections of an accumulator matrix. The method is configured such that a district candidate is generated when an entry in the Houghräumen 10 . 11 or 12 greater than 15. Accordingly, a circle is generated with parameters such as the address in the matrix of the entry 17 with the value 16 correspond. Accordingly, the step generates 4.1 in the clippings 10 and 11 the Houghräume no district candidate. At the excerpts from the Houghräumen 10 . 11 and 12 These are excerpts from three-dimensional matrices. Accordingly, each entry points in the clippings 10 . 11 and 12 a three-dimensional address that stores the radius with one dimension and the center with two dimensions.

The method is also configured such that a circle is generated when an entry of the matrix is larger than 5 and the sum of the entry with its neighborhood is greater than 17 , This is true for the neighborhood 13 to. Accordingly, in the clipping 10 of the Houghraums in the second line also generates a circle with the address of the entry corresponding parameter combination. For the neighborhood 14 however, no circle is generated because the sum of the entries of the neighborhood 14 is smaller than 17 , Further, in the neighborhood 15 of the clipping 11 of the Houghraums no circle generated, since the sum of the entries of the neighborhood 15 is smaller than 17 , Further, in the clipping 12 of the Houghraum in the neighborhood 16 no circle is generated because the sum of the entries of the neighborhood 16 is smaller than 17 ,

The 3 shows an embodiment of the method steps M6 and M7 , The 3 shows an object 20 in an edge picture 30 with the resolution 760 × 475 pixels. The circle 21 represents the circle with the average radius and the center of its group. It can be seen that the radius of the circle 21 Although approximately to the object 20 fits, but the center 23 of the circle 21 opposite the center 28 of actual circle 22 is moved. The procedure presents in the step M6 by forming the gray value difference between two circle segments of the circle 21 notice that the circle 21 the object 20 not described with satisfactory accuracy.

To find a circle containing the object 20 in more detail, the method also forms gray scale differences for circles having a center within the search area 24 lies. The search area 24 corresponds to a neighborhood of 5x5 pixels of the center 23 of the circle 21 ,

In the step M7 it is determined that the circle 22 the lowest gray value difference over the circle segments 25 and 26 in the search area 24 having contained circles.

The gray value difference is determined by the gray values of pixels on the circle segment 25 from the gray values of the pixels in the circle segment 26 be subtracted. It is the gray value of a pixel in the circle segment 25 the gray value of the opposite pixel in the circle segment 26 subtracted. In the 3 are exemplary the circle diagonals 27a . 27b and 27c between two opposite pixels. Accordingly, the method estimates the contour of the object 20 through the circle 22 around the center 28 from.

The 4 shows a curve 42 together with an embodiment of the method steps M8.0 or. M8.1 to estimate the radius 43 the curve 42 , For this purpose, first the circle or curve candidate 40 with the largest radius 41 determined his group of curve candidates. This ensures that no information that is outside a circle of too small a radius is lost. The following section looks for a circle that contains the curve candidate 40 touched. Thus, the circle becomes the center 44 determined. The following is the distance of the midpoint 44 to the center 45 of the curve candidate 40 determined and the radius of the curve 42 estimated by the radius 41 of the curve candidate 40 the distance between the centers 44 and 45 is subtracted.

The 5 shows a driver assistance system 50 with an image sensor 51 , a data storage unit 52 , a computing unit 53 as well as a communication unit 54 , The driver assistance system 50 is set up to detect traffic signs by using image data from the image sensor 51 from the arithmetic unit 53 to an edge picture 30 be processed and contours of traffic signs in the edge image 30 by means of a modified Hough transformation of the edge image 30 be recognized. Here is the communication unit 54 set up to issue a message via a recognized traffic sign.

LIST OF REFERENCE NUMBERS

10

Detail of a Houghroom

11

Detail of a Houghroom

12

Detail of a Houghroom

13

neighborhood

14

neighborhood

15

neighborhood

16

neighborhood

17

entry

20

object

21

circle

22

circle

23

Focus

24

search area

25

circular segment

26

circular segment

27a-c

circle diagonal

28

Focus

30

edge image

40

curve candidate

41

radius

42

Curve

43

radius

44

Focus

45

Focus

M1-M9

individual process steps

50

Driver assistance system

51

image sensor

52

Data storage unit

53

computer unit

54

communication unit

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

• US 6807286 B1 [0002]

Claims (17)

Method for identifying parameterizable geometric curves having at least a first parameter and a second parameter with the following steps: - providing at least one edge image (30) with pixels (M1); Setting a set of curves to be recognized in the edge image with at least a first parameter and a second parameter (M2); - performing a Hough transformation for several pixels of the edge image into a Hough space (10; 11; 12) (M3); Detecting a curve candidate (21; 22; 40) if a value (17) at a point in the Hough space is greater (10; 11; 12) than a defined first threshold value (M4.1) and / or recognition of a curve candidate ( 21; 22; 40) if a value at a point in the Hough space is greater than a second predetermined threshold and a sum of values in a predetermined neighborhood (13; 14; 15) of the point is greater than a third threshold; second threshold is less than the first threshold and the third threshold is at least equal to the first threshold (M4.2); Grouping curve candidates (21; 22; 40) into a group if the curve candidates (21; 22; 40) have at least one parameter which does not exceed a predetermined tolerance value of the parameter of another curve candidate (21; 22; 40) deviates (M5); Estimation of a curve by means of the grouped curve candidates (21, 22, 40) (M9). Method according to Claim 1 which detects circles (21; 22), wherein the first parameter is designed as a radius (41; 43) and the second parameter as a center point (23; 28; 44; 45). Method according to one of the preceding claims, which further comprises the following steps: - forming a color value difference between at least two points of a curve candidate (21; 22; 40) for a plurality of curve candidates (21; 22; 40) of the group (M6); - determining at least one parameter of a curve candidate (21; 22; 40) having the lowest color value difference (M7) formed; Estimation of at least one curve by means of the at least one determined parameter (M9). Method according to Claim 3 , wherein the color value difference is designed as a gray value difference. Method according to Claim 3 or 4 , wherein the color value difference between two symmetrical, in particular point-symmetrical points is formed. Method according to one of the preceding claims, wherein an average for the first and / or second parameter for curve candidates (21; 22; 40) of the group is determined. Method according to Claim 6 in which the color value difference for curve candidates (21; 22; 40) is determined with a parameter lying within a predetermined search range (24) around the average of the parameter. Method according to Claim 6 or 7 in which a plurality of edge images (30) are provided with different resolutions and, depending on the average of the parameter, a resolution of the edge image (30) from which the chrominance difference is formed is selected. Method according to one of the preceding Claims 3 - 8th in which the color value difference is determined for two subsections (25; 26) of a curve candidate (21; 22; 40). Method according to one of the preceding claims, wherein a parameter range to which the first and / or second parameter of the curves to be detected is restricted is determined. Method according to Claim 10 in which a resolution of the edge image (30) for which the Hough transformation is performed is selected as a function of the specified parameter range. Method according to one of the preceding claims, wherein for at least a first parameter a first curve candidate (21; 22; 40) with the highest value of the parameter among the curve candidates (21; 22; 40) of the group is determined and the first parameter of the one to be recognized Curve (42) is estimated by means of the first parameter of the first curve candidate (21; 22; 40) (M8.0). Method according to Claim 12 in which a second curve candidate (21; 22; 40) is determined which touches the first curve candidate (21; 22; 40), wherein the difference of a second parameter of the first and second curve candidates (21; 22; 40) is determined and the first parameter of the curve (42) to be detected is estimated by means of the first parameter of the first curve candidate (21; 22; 40) and the difference of the second parameter of the first and second curve candidates (21; 22; 40) (M8.1) , Method according to Claim 13 , wherein the first parameter is formed as a radius (41; 43) of a circle (21; 22; 40) or as a half-axis length of an ellipse and wherein the second parameter is the center (23; 28; 44; 45) of the circle (21; 22 ; 40) or the ellipse is formed. Arithmetic unit for the detection of traffic signs in a digital image, wherein the arithmetic unit is set up, the method according to one of Claims 1 - 14 perform. Computer program with program code means for performing the functions according to one of Claims 1 - 14 perform when the computer program is executed on a computing unit. Driver assistance system (50) with an image sensor (51), a data storage unit (52), a computing unit (53) and a communication unit (54), wherein the driver assistance system (50) is adapted to recognize traffic signs by image data of the image sensor (51) of processing the arithmetic unit (53) into an edge image (30) and recognizing contours of traffic signs in the edge image (30) by means of a modified Hough transformation of the edge image (30), wherein the communication unit (54) is set up, a message about a recognized traffic sign issue.

Images