DE102013002889A1 - Method for modeling road edge of road for vehicle, involves determining road curvature using two curvature corrected inspection of search area in image, where inspection is corrected according to two different road curvature hypotheses - Google Patents

Abstract

The method involves determining a road curvature using two curvature corrected inspection of a search area in the image, where the curvature corrected inspection is corrected according to two different road curvature hypotheses. A lateral deposition is determined (5) by using an inspection of the search area. A yaw angle difference is determined (6) by using two yaw angle difference corrected inspection of another smaller search area in the image. An independent claim is included for a device for modeling a road edge of a road for a vehicle.

Description

The invention relates to a method and a device for modeling a roadway edge and to a vehicle having a device for modeling a roadway edge.

For lane keeping or assistance systems for vehicles and other applications, it can be useful to reliably detect a roadway edge of a roadway area ahead of the vehicle, in particular if this corresponds to the limit of the lane to be maintained.

A variety of sensors can be used individually or in combination. Typically, at least one camera is used, with grayscale cameras being preferred for better dynamic range and / or cost reasons. Especially in unfavorable light conditions Gaubildkameras are superior to color cameras.

An exclusively camera-based lane edge detection is carried out according to the prior art by means of a lane edge marking. If the marking is missing in part, the roadside edge can be extrapolated on the basis of the existing marking in the area of the missing marking. However, if there is no marking over longer distances, further or different sensor systems for reliable detection of the roadway edge are necessary according to the prior art.

A lane edge detection can in particular also be model-based. Taking advantage of the fact that lanes are represented as clothoids, can, as in DIN70000 3, model a lateral deposit of the roadway edge as a function of a distance x from an origin of a vehicle-borne and vehicle-centered coordinate system as a third-order polynomial: f (x) = a + b * x + c * x ² + d * x ³ (1)

A yaw angle difference between a vehicle longitudinal axis and a lane longitudinal axis corresponds to the change of the lateral tray and can therefore be modeled as follows: f '(x) = b + 2 * c * x + 3 * d * x ² (2)

A lane curvature corresponds to the change in the yaw angle difference and can therefore be modeled as follows: f '' (x) = 2 * c + 6 * d * x (3)

Finally, the curvature change can be modeled as a constant independent of x: f '''(x) = 6 · d (4)

Dean Pomerleau describes in "RALPH: Rapidly Adapting Lateral Position Handler" a method for vehicle control, which selects a trapezoidal search area from an image and parallelized by transformation into a low-resolution, rectangular image. To determine the curvature of the road, different curvature hypotheses are subtracted from the low-resolution image and tested how well the respective curvature hypothesis "straightened" the low-resolution image. By summing up the maximum absolute differences between gray sums of adjacent columns, the curvature hypothesis that best "straightens" the low resolution image can be determined.

The invention is based on the object of specifying a method and a device which make it possible to detect a roadway edge only by means of a grayscale camera even when the roadway edge is not marked over longer distances.

In a preferred embodiment of the invention it is provided that the roadway edge is modeled by a method according to claim 1. In this case, the roadway edge is modeled as a polynomial of at least second order using a gray-scale image which represents a region ahead of a vehicle in the direction of travel and is detected by means of a geometrically calibrated camera.

• – Bestimmen einer Fahrbahnkrümmung, einer lateralen Ablage und einer Gierwinkeldifferenz, und
• – Bestimmen eines Absolutglieds des Polynoms unter Verwendung der lateralen Ablage, eines linearen Glieds des Polynoms unter Verwendung der Gierwinkeldifferenz und eines quadratischen Glieds des Polynoms unter Verwendung der Krümmung.

The method is characterized in that it comprises the following steps:

• Determining a roadway curvature, a lateral shelf and a yaw angle difference, and
• - Determining an absolute term of the polynomial using the lateral shelf, a linear term of the polynomial using the yaw rate difference and a quadratic term of the polynomial using the curvature.

The road curvature is determined using at least two curvature-corrected views of a search area in the image. The lateral tray is determined using the search area in the image and the determined road curvature. And, the yaw angle difference is determined using at least two yaw angle difference corrected corrections of another, smaller search area in the image. The curvature-corrected surveys are according to at least two different Lane curvature hypotheses are corrected and the yaw angle difference corrected corrections are corrected according to at least two different yaw angle difference hypotheses.

This has the advantage that only a grayscale camera is required in order to reliably detect a roadway edge in real time even when the roadside edge is not marked. Further, the method allows for the elimination of curvature change modeling, which could compromise the robustness of the modeling, and has been found to be of little benefit in experiments for lane keeping methods in only a few scenarios and then only a little.

• – Glätten der lateralen Ablage, der Gierwinkeldifferenz und der Krümmung unter Verwendung einer lateralen Ablage, einer Gierwinkeldifferenz und einer Krümmung, die unter Verwendung eines unmittelbar zuvor erfassten Bildes bestimmt wurden, und
• – Bestimmen einer korrigierten lateralen Ablage unter Verwendung der geglätteten lateralen Ablage, der Bildrate und eines Querbeschleunigungsmaximalwerts, wobei
• – das Absolutglied unter Verwendung der korrigierten lateralen Ablage, das lineare Glied unter Verwendung der geglätteten Gierwinkeldifferenz und das quadratische Glied unter Verwendung der geglätteten Krümmung bestimmt wird.

In a further preferred embodiment of the invention, it is provided that the camera acquires images at a frame rate and the method further comprises the following steps:

• Smoothing the lateral shelf, the yaw angle difference and the curvature using a lateral shelf, a yaw angle difference and a curvature determined using an image captured immediately before, and
• Determining a corrected lateral offset using the smoothed lateral offset, the frame rate and a lateral acceleration maximum value, wherein
• The absolute term is determined using the corrected lateral offset, the linear term using the smoothed yaw rate difference and the square term using the smoothed curve.

• – Bestimmen einer Spurbreite unter Verwendung der lateralen Ablage und Informationen zur lateralen Ablage einer Fahrspurbegrenzungsmarkierung, Glätten der Spurbreite und Bestimmen einer geglätteten lateralen Ablage unter Verwendung der geglätteten Spurbreite.

Smoothing the lateral tray may include the following steps:

• Determining a track width using the lateral tray and information for later depositing a lane boundary marker, smoothing the track width and determining a smoothed lateral tray using the smoothed track width.

• – Erzeugen von mindestens zwei Spaltensummenvektoren durch Aufsummieren von Spalten der mindestens zwei krümmungskorrigierten Aufsichten und Bestimmen der Fahrbahnkrümmung unter Verwendung der Spaltensummenvektoren.

Determining the roadway curvature can in particular comprise the following steps:

• Generating at least two column sum vectors by summing columns of the at least two curvature-corrected planets and determining the roadway curvature using the column sum vectors.

• – Erzeugen von mindestens zwei anderen Spaltensummenvektoren durch Aufsummieren von Spalten der mindestens zwei gierwinkeldifferenzkorrigierten Aufsichten und
• – Bestimmen einer Gierwinkeldifferenz unter Verwendung der Spaltensummenvektoren.

The determination of the yaw angle difference may in particular comprise the following steps:

• Generating at least two other column sum vectors by summing columns of the at least two yaw angle difference corrected planets and
• Determining a yaw angle difference using the column sum vectors.

• – Erzeugen einer weiteren, gemäß der bestimmten Fahrbahnkrümmung krümmungskorrigierten Aufsicht des Suchbereichs,
• – Erzeugen eines weiteren Spaltensummenvektors durch Aufsummieren der Spalten der weiteren, krümmungskorrigierten Aufsicht,
• – Erzeugen noch eines weiteren Spaltensummenvektors durch Zuordnen, zu jedem Pixel der weiteren, krümmungskorrigierten Aufsicht, eines Wertes entsprechend einer Gradientenverteilung in einer Nachbarschaft des jeweiligen Pixels und spaltenweise Aufsummieren der zugeordneten Werte,
• – Bestimmen eines Suchintervalls innerhalb des weiteren Spaltensummenvektors unter Verwendung des noch einen weiteren Spaltensummenvektors,
• – Bestimmen eines Maximums innerhalb des Suchbereichs und
• – Bestimmen der lateralen Ablage unter Verwendung des bestimmten Maximums.

The determination of the lateral deposit can in particular comprise the following steps:

• Generating a further, according to the determined road curvature curvature-corrected supervision of the search area,
• Generating a further column sum vector by summing up the columns of the further, curvature-corrected plan view,
• Generating yet another column sum vector by assigning, to each pixel of the further, curvature-corrected plan view, a value corresponding to a gradient distribution in a neighborhood of the respective pixel and adding the assigned values column by column,
• Determining a search interval within the further column sum vector using the still further column sum vector,
• Determining a maximum within the search range and
• Determining the lateral shelf using the determined maximum.

The modeling is particularly good if the other, curvature-corrected supervision has a higher resolution than the at least two curvature-corrected views.

It has been shown, in particular, that the roadway edge can already be stably and reliably modeled when the search area comprises an area in front of the vehicle with a small depth, and the further, smaller search area comprises a further area in front of the vehicle with only half as much Depth includes.

In order to reliably determine the road curvature under real-time conditions, it is advantageous if the at least two different road curvature hypotheses are selected from one interval using a variable step size. In this case, information about a lane boundary marking curvature may be used to determine the interval.

In order to reliably determine the yaw angle difference under real-time conditions, it is advantageous if the at least two different yaw angle difference hypotheses are selected from a further interval. It can be an information to a Lane boundary marker yaw angle difference used to determine the further interval.

According to the invention, a device for modeling a roadway edge with the features of claim 10 is also proposed. In this case, the device comprises a geometrically calibrated camera which can be arranged in a vehicle so that it can detect grayscale images of a region lying in front of the vehicle in the direction of travel. The device is designed to model the lane edge as a polynomial of at least second order. The apparatus is characterized by comprising means adapted to determine a lane curvature, a lateral shelf and a yaw angle difference, and means adapted to place an absolute term of the polynomial using the lateral shelf a linear term of the polynomial Use the yaw angle difference and determine a quadratic term of the polynomial using the curvature.

In this case, the device is designed to determine the roadway curvature using at least two curvature-corrected views of a search area in the image. The curvature-corrected topographies are corrected according to at least two different roadway curvature hypotheses.

The apparatus is further configured to determine the lateral tray using the search area in the image and the determined roadway curvature.

Finally, the device is also designed to determine the yaw angle difference using at least two yaw angle difference corrected planets of another, smaller search area in the image, correcting yaw angle difference corrected planets according to at least two different yaw angle difference hypotheses.

According to the invention, a vehicle with the features of claim 11 is also proposed. The vehicle according to the invention comprises the device according to the invention.

Further preferred embodiments of the invention will become apparent from the remaining, mentioned in the dependent claims characteristics.

The invention will be explained below in embodiments with reference to the accompanying drawings. Show it:

1 a flowchart of an embodiment of the method according to the invention;

2 an exemplary grayscale image of a lying in front of a vehicle in the direction of travel area;

3 an exemplary gray level image of a lower resolution supervision of a search area in the exemplary gray level image 2 ;

4 an exemplary grayscale image of a lower resolution, corrected on the basis of one of the road curvature best corresponding roadway curvature hypothesis supervision of the search area in the exemplary grayscale image 2 ;

5 an exemplary grayscale image of a higher resolution, corrected on the basis of a certain road curvature supervision on the search area in the exemplary grayscale image 2 ;

6 an exemplary greyscale representation of a coherence matrix to the supervision 5 ;

7 an exemplary grayscale image of a top view of another, smaller search area in the exemplary grayscale image 2 ;

8th the grayscale image 2 with superposed polynomial of second order for modeling the roadway edge;

9 a flowchart of another embodiment of the method according to the invention and

10 a flowchart of yet another embodiment of the method according to the invention.

1 shows a flowchart of an embodiment of the method according to the invention. In this embodiment, a camera image is flowing 1 , Camera calibration data 2 and also referred to as Egomotion information about a vehicle own movement 3 in the process. The procedure comes without information 4 to a lane separation or a lane edge marker, but this can be in a in 9 use shown embodiment advantageous. 2 shows an exemplary camera image 1 how it can be used in the process.

The exemplary camera image 1 is recorded by a built-in near the interior mirror of the motor vehicle grayscale camera. The camera calibration data 2 allow to convert from image coordinates in pixels to real world coordinates in meters. In particular, a conversion into a carried by the vehicle Coordinate system possible. The information about the vehicle's own movement 3 includes at least vehicle speed and yaw rate. In the embodiment of the method, which information 4 used for lane boundary marking, this information may be, for example, a polynomial with which the lane boundary marker is modeled and which is provided by an external lane detection unit. In this case, in one exemplary embodiment, a last-determined polynomial is made available for a period of time if the lane recognition unit fails to recognize the lane boundary marking, for example because the marker is missing.

In this case, the method comprises two image processing which independently accesses these influencing variables 5 . 6 , The first image processing 5 determined from the influencing variables curvature and lateral deposition of the roadway edge. The second image processing 6 determines from the influencing variables a yaw angle difference.

The of the image processing 5 and 6 certain sizes are then modeled by a model 7 used to using the information on the vehicle's own movement 3 , and optionally the lane boundary marking information, to model the lane boundary as a second order polynomial. In this case, a lane boundary marking is to be understood as any marking which delimits the lane, for example from another lane, from an area of the lane not intended for driving or from the other lane edge.

The first image processing 5 calculated in one step 8th a search range co-determined by camera type and camera viewing angle and generated in step 9 a lower resolution view of the search area. The longitudinal size of the search area can be influenced by an object recognition unit such that the search area does not include a recognized object. If no object is detected or the search area is not influenced by an object recognition unit, the search area may for example extend laterally a few meters centered about a longitudinal axis of the vehicle and longitudinally have a multiple extent, wherein a beginning of the search range coincides, for example, with a lowest pixel row or by a lowest of those pixel rows is defined in which the complete lateral extent of the search area is completely covered by the image. This lowest pixel row can vary between images as the camera calibration varies.

In 2 is the search area through the points 100 . 300 . 400 and 600 limited and 3 shows a view of the through the points 100 . 300 . 400 and 600 limited search area. The less resolved supervision is curvature corrected by means of different road curvature hypotheses and each correction is assigned a quality measure. Pixel lines of the image are shifted individually according to the respective roadway curvature hypothesis. 4 shows an example of a corrected according to a roadway curvature hypothesis supervision. The correction, which has been assigned the largest quality measure, is then determined as the amount equal to the roadway curvature to be determined. This is done in step 10 , Experiments have shown that a resolution of the supervision of less than 100 × 100 pixels allows the execution of the steps in real time and at the same time determines good approximations of the actual road curvature.

The determined quality vector assigns a quality measure to each pair of adjacent image columns, which depends on the gray values and number of pixels in each of the adjacent columns.

The largest component of the quality vector is then the quality measure, which is assigned to the respective correction.

Subsequently, a higher-resolution view of the search area is curvature-corrected by means of the determined roadway curvature. This is done in step 11 , The lateral deposit is thus determined using a curvature-corrected according to the specific road curvature supervision of the search area. 5 shows by way of example one according to the on the basis of the image in 2 certain road curvature corrected higher resolution supervision. In one step 12 another quality vector is generated using the higher resolution supervision.

Within the further quality vector is in step 13 set a search interval for a roadside search. Optionally, this is done after determining a lane boundary marker in step 16 in 9 and using a lane boundary marker. The lane boundary marking can be made available externally or determined, for example, using the further quality vector. Alternatively or additionally, another quality vector can be used for the search interval determination. This is in 10 shown.

This still another quality vector, if it finds use, in step 14 be determined using a coherence metric, where the coherence metric maps the pixels to coherence measures that are higher than eigenvalues of the curved-corrected, transformed with a texture tensor resolved supervision in step 14 can be calculated. 6 shows by way of example the coherence measures assigned to the pixels as a gray scale image. Isotropic areas are greyish, homogeneous areas are white and areas with ideal orientation are shown in black. In step 15 are then determined in the search area, the edge of the road and a lateral storage of the vehicle to the particular edge of the vehicle. The coherence metric may be subjected to a distance transformation in the vertical direction to remove or attenuate artifactual isotropic gradient distributions.

The accuracy with which the lateral support can be determined from the edge of the road depends directly on the resolution of the higher resolution supervision. Experiments have shown that a resolution that allows the execution of the steps in real time, at the same time determines good approximations of the actual lateral storage.

The second image processing 6 calculated in one step 17 another, smaller search area and created in step 18 a lower resolution view of the other, smaller search area. In 2 becomes the further, smaller search area through the points 100 . 200 . 500 and 600 limited and 7 shows a view of the through the points 100 . 200 . 500 and 600 limited other, smaller search area. The lower resolution view of the smaller search area is corrected by means of different yaw angle difference hypotheses. In this case, pixel lines of the lower-resolution view of the smaller search area are shifted individually according to the respective yaw angle difference hypothesis.

For each correction, a quality vector is created and a largest component of the vector, which is assigned to the correction as a quality measure, is determined. The correction, which has been assigned the largest quality measure, is then determined as amount equal to the yaw angle difference to be determined. This is done in step 19 , The further, smaller search area may, for example, extend laterally a few meters centered around a longitudinal axis of the vehicle and have a multiple extension longitudinally, wherein a beginning of the search area coincides, for example, with a lowest pixel row or is defined by a lowest of those pixel rows in which the complete lateral extent of the search area is completely captured by the image. This lowest pixel row can vary between images as the camera calibration varies.

The in step 15 certain lateral deposit of the lane edge, in step 19 certain yaw angle difference and the in step 10 certain roadway curvature will then step in 20 in the context of the model generation 7 using the calibration data 2 the camera converted into a vehicle coordinate system. The particular lateral tray becomes an optional step 22 smoothed with a previously determined lateral tray, such as by low-pass filtering. The yaw angle difference in vehicle centered coordinates will be in step 22 smoothed with a previously determined yaw angle difference, such as low pass filtering. And the lane curvature in vehicle-centered coordinates gets in step 22 smoothed with a previously determined roadway curvature, such as low-pass filtering. For example, a weighted sum of the current determined roadway curvature and a smoothed, immediately previously determined roadway curvature is determined. The weights used are greater than or equal to zero and add to 1. Analogously, for example, a weighted sum of the current determined yaw angle difference and a smoothed, previously determined yaw angle difference and a weighted sum of the current determined track width and a smoothed, immediately previously determined track width determined.

Subsequently, in step 23 a smoothed lateral deposit of the roadway edge in the vehicle coordinate system determined.

It can be taken into account here that the current smoothed lateral deposit can deviate from a previously determined, smoothed lateral deposit only within an interval determined by the driving physics. In the case of a maximum lateral acceleration a and a time duration t between the previously determined and the currently determined lateral deposit, the deviation can amount to a maximum of s = 0.5 · a · t ² . If, for example, at a frame rate of 15 frames per second, reference is made to a lateral offset determined on the basis of the immediately preceding image, then t = 0.067 seconds.

Finally, an absolute term of a polynomial is determined using the smoothed lateral table, a linear term of the polynomial using the smoothed yaw angle difference, and a quadratic term of the polynomial using the smoothed curve. In the context of the described embodiment, the change in curvature is set to zero regardless of the image. Since the lateral deviation α used for the polynomial determination and the yaw angle difference β used are determined by the vehicle for a longitudinal distance x ₀ , which corresponds to the longitudinal beginning of the search regions, the lateral deposition of the roadway edge from the vehicle longitudinal axis as a function of the distance results x from the origin of the vehicle-borne and vehicle-centered coordinate system and the smoothed curvature γ as: f (x) = α + β · (x - x ₀ ) + 0.5 · γ · (x - x ₀ ) ² (5)

The yaw angle difference corresponds to the change of the lateral tray and then results as: f '(x) = β + γ * (x - x ₀ ) (6)

And the road curvature results as: f '' (x) = γ (7)

However, it is also possible to also model the change in curvature, for example on the basis of the polynomial modeling the lane boundary marking.

The polynomial with these links then represents the modeling according to this embodiment of the roadway edge. 8th shows an example image of an area ahead of a vehicle, wherein a calculated based on the information present in the example image information polynomial 999 second order for modeling the right lane edge is superimposed example of the example image.

Thus, in one embodiment of the method of modeling a lane boundary using an image, at least two column sum vectors are generated by summing columns of at least two low-resolution, curvature-corrected planets of a search area in the image. The low-resolution curvature-corrected views are corrected according to at least two different road curvatures. Then, a lane curvature is estimated using the column sum vectors, and another column sum vector is generated by summing the columns of a high-resolution, curvature-corrected plan view of the search area. The high-resolution curvature-corrected top view is corrected according to the estimated road curvature. Now, a lateral deviation lane marker is determined using the lane boundary marker information and a search window within the further column header vector using the lane boundary marker information. Then, another column sum vector is generated by assigning, to each pixel, the high resolution, curvature corrected plan, a value corresponding to its gradient distribution in a neighborhood of the respective pixel, and adding the assigned values column by column. A beginning vector component of the search window within the further column sum vector is corrected using threshold values and from still another column sum vector. Within the corrected search window a maximum is determined. A lateral deposit to the lane edge is then determined using the determined maximum. Furthermore, at least two other column sum vectors are generated by summing columns of at least two low resolution, yaw angle difference corrected, planets of another, smaller search area in the image. In this case, the low-resolution, yaw angle-difference-corrected planets are corrected according to at least two different yaw angle differences. Using the other column sum vectors, a yaw rate difference is determined, and a lane width is determined using the lateral lane mark and lane edge shelves. The determined lane width, the determined yaw angle difference, and the determined curvature are smoothed, and using the smoothed track width, a frame rate and a lateral acceleration maximum value, a corrected lateral offset to the roadway edge is determined. Finally, the absolute term, linear and quadratic term of the polynomial are determined, the absolute term of the polynomial being determined using the corrected lateral offset, the linear term of the polynomial using the smoothed yaw rate difference, and the quadratic term of the polynomial using the smoothed curve.

The different curvature hypotheses used in one embodiment may be selected from a set of curvatures that correspond to radii that deviate from a reference radius at most by a maximum radius difference. The reference radius is based on one in the information 4 determined for lane boundary marking curvature of the lane boundary marker and / or a last determined curvature determined.

The different yaw angle difference hypotheses used may, in one embodiment, be selected from a set of yaw rate differences that differ from a reference yaw angle difference by at most one maximum yaw angle difference. In this case, the reference yaw angle difference is based on one in the lane boundary marking information 4 contained yaw angle difference of the lane boundary marker and / or a last determined yaw angle difference determined.

Other embodiments are either without determination of the yaw angle difference or without determination of the curvature and take the respective undetermined size equal to zero. It is also possible to do several image processing 5 with not or only slightly overlapping search areas in the image to determine the edge of the lane. These are the independently of each other determined lateral shelves from step 15 used as bases for a model estimation. In the simplest case, this may be a polynomial regression, but other, more complex techniques are also applicable. The longitudinal extent and / or position of the search ranges may be adaptively changed depending on the yaw rate of the vehicle. In other embodiments, the search ranges are determined dynamically. Thus, a central axis of the search areas, for example as a function of the strength of a cornering, can be shifted relative to a vehicle longitudinal axis.

The described invention is suitable for modeling a right-hand roadside edge as well as for modeling a left-hand roadway edge. The described invention is also suitable for assisting the detection of a lane edge marking. The invention described is also suitable for the selective determination of a single or only two components of the edge polynomial. The results of the described invention can advantageously be further improved if information is used on a lane boundary marker and / or on a lane edge marker. In this case, for example, the modeling of a lane boundary on the basis of a lane marking on this one lane edge and / or a lane marking on the other lane edge done.

LIST OF REFERENCE NUMBERS

1

camera image

2

calibration data

3

Information about a vehicle's own movement

4

optional information about a lane boundary marker

5

Image processing for determining a curvature and a lateral deposit

6

Image processing for determining a yaw angle difference

7

model generation

8th

Set a search scope

9

Generate a low-resolution view of the search area

10

Determining a roadway curvature

11

Create a higher-resolution, curvature-corrected view of the search area

12

Generating a quality vector

13

Determining the lane boundary marker within the quality vector

14

optionally generating another quality vector

15

Determining a search area in the quality vector

16

Determine the lateral shelf

17

Set a different, smaller search range

18

Generating a low-resolution view of the further search area

19

Determining the yaw angle difference

20

Convert to a vehicle coordinate system

21

Calculation of a track width

22

Low pass filtering of track width, yaw angle difference and curvature

23

Calculation of a smoothed lateral deposit

24

Modeling the roadside as a polynomial

100

Limitation of search areas

200

Limitation of search areas

300

Limitation of search areas

400

Limitation of search areas

500

Limitation of search areas

600

Limitation of search areas

999

polynomial

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 non-patent literature

• DIN 70000 [0005]
• Dean Pomerleau describes in "RALPH: Rapidly Adapting Lateral Position Handlers" [0009]

Claims (11)

A method for modeling a lane boundary, wherein the lane boundary using a grayscale image ( 1 ), which is an area ahead of a vehicle and detected by a geometrically calibrated camera, is modeled as a polynomial of at least second order and wherein the method is characterized by comprising the steps of: - determining ( 5 ) of a roadway curvature using at least two curvature-corrected elevations of a search area in the image, wherein the curvature-corrected elevations are corrected according to at least two different roadway curvature hypotheses, - determining ( 5 ) of a lateral storage using a curvature-corrected supervision of the search area according to the determined road curvature, - determining ( 6 ) a yaw angle difference using at least two yaw angle difference corrected planets of another, smaller search area in the image, the yaw angle difference corrected corrections being corrected according to at least two different yaw angle difference hypotheses, and - determining ( 7 . 24 ) of an absolute term of the polynomial using the lateral shelf, a linear term of the polynomial using the yaw rate difference and a quadratic term of the polynomial using the curve. The method of claim 1, wherein the camera captures images at a frame rate, and wherein the method is characterized by further comprising the steps of: - smoothing ( 22 ) of the lateral deposit, the yaw angle difference and the curvature using a lane width, a yaw angle difference and a curvature determined using an image acquired immediately before, and - determining ( 23 ) of a corrected lateral shelf using the smoothed lateral shelf, the frame rate and a lateral acceleration maximum value (FIG. 3 ), wherein - the absolute member is determined using the corrected lateral shelf, the linear member using the smoothed yaw angle difference, and the square member using the smoothed curvature. Method according to claim 2, wherein smoothing ( 22 ) of the lateral tray comprises the steps of: determining a track width using the lateral tray and information for later depositing a lane boundary marker, smoothing the track width, and determining a smoothed lateral tray using the smoothed track width. Method according to one of the preceding claims, characterized in that the determining ( 10 ) of the roadway curvature comprises the following steps: - generating ( 10 ) of at least two column sum vectors by summing columns of the at least two curvature-corrected planets, and - determining ( 10 ) of the roadway curvature using the column sum vectors. Method according to one of the preceding claims, characterized in that the determining ( 19 ) of the yaw angle difference comprises the following steps: - generating ( 19 ) of at least two other column sum vectors by summing up columns of the at least two yaw angle difference corrected planets, and - determining ( 19 ) of a yaw angle difference using the column sum vectors. Method according to one of the preceding claims, characterized in that the determining ( 16 ) of the lateral tray comprises the following steps: - generating ( 11 ) another, according to the specific road curvature curvature-corrected supervision of the search area, - generating ( 12 ) of another column sum vector by adding up the columns of the further, curvature-corrected plan view, - generating ( 14 ) of yet another column sum vector by assigning, to each pixel of the further, curvature-corrected plan view, a value corresponding to a gradient distribution in a neighborhood of the respective pixel and adding the assigned values column by column, - determining ( 15 ) of a search interval within the further column sum vector using the still further column sum vector, - determining ( 16 ) of a maximum within the search range and - determining ( 16 ) of the lateral tray using the determined maximum. A method according to claim 5, characterized in that the further, curvature-corrected supervision has a higher resolution than the at least two curvature-corrected views. Method according to one of the preceding claims, characterized in that the at least two different lane curvature hypotheses are selected from an interval using a variable step size, the interval using Information about a lane boundary marking curvature is determined. Method according to one of the preceding claims, characterized in that the at least two different yaw angle difference hypotheses are selected from a further interval, wherein the further interval is determined using information on a lane boundary marking yaw angle difference. A device for modeling a lane edge, the device comprising a geometrically calibrated camera that can be arranged in a vehicle so that it can capture gray scale images of a lying in front of the vehicle in the direction of the vehicle area, and wherein the device for modeling the lane edge as a polynomial at least second order, characterized in that the device comprises means configured to: - determine a roadway curvature using at least two curvature-corrected elevations of a search area in the image, wherein the curvature-corrected elevations are corrected according to at least two different roadway curvature hypotheses, - a lateral shelf determine a yaw angle difference using at least two yaw angle difference corrected corrections of another, smaller search bin in the image, the yaw angle difference corrected corrections being corrected according to at least two different yaw angle difference hypotheses, and - an absolute term of the polynomial using the lateral shelf, a linear term of the polynomial using the yaw rate difference and a quadratic term of the polynomial using the curve determine. Vehicle with a device according to claim 10.

Images