Method for detecting lane markings by validating based on line width; control unit; parking space detection system; and driver assistance system

Abstract

Method for detecting a road marking (2) for a vehicle (1), by - Providing an image of the vehicle's surroundings (U) using a camera (3), - Identifying a potential road marking (4) in the image of the surroundings (U), - Determining the extent of the potential road marking (4) perpendicular to a main extension direction (22) of the potential road marking (4) as the line width (20) of the potential road marking (4), and - Recognition of the road marking (2) by validating the potential road marking (4) as plausible based on its line width (20), only if the line width (20) changes along the main extension direction (22) by no more than a predetermined amount, wherein - a plurality of values ​​(21) for the line width (20) along the main extension direction (22) is determined, and - the potential road marking (4) is validated as plausible at most if individual values ​​from the majority of values ​​(21) for the line width (20) differ from each other by no more than a predetermined amount and - a regression line (27) is fitted to the majority of values ​​(21) for the line width (20) and a slope of the regression line (27) is determined as a measure of the deviation of the majority of values ​​(21) for the line width (20) from each other.

Description

[0001] The invention relates to a method for recognizing a lane marking from an image of a vehicle's surroundings. The invention also relates to a control device. Further aspects of the invention relate to a parking space recognition system, a driver assistance system, and a computer program.

[0002] The detection of lane markings is crucial for a wide variety of driver assistance systems. Examples include autonomous driving systems, lane keeping assist, and parking space detection systems. Lane markings are typically detected using an image of the vehicle's surroundings. This image is either captured by a single camera on the vehicle or compiled from raw images from multiple cameras. Known image analysis algorithms are then used to identify the lane markings within the image. For instance, image gradients are created across different areas of the image, and a lane marking is detected if the gradient matches a predefined pattern.

[0003] For the correct functioning of the driver assistance system or the parking space detection system, it is necessary that lane markings are reliably and safely detected, see JP 2011 243 161 A or KASAI, T., Onoguchi, K.: Lane detection system for vehicle platooning using multi-information map. In: 13 thInternational IEEE Conference on Intelligent Transportation Systems. IEEE, 2010. pp. 1350-1356. doi: 10.1109 / ITSC.2010.5625215. Incorrectly detected lane markings can severely disrupt the functionality of a driver assistance system or parking space detection system. Examples of incorrectly detected lane markings include so-called false positives. False positives refer to situations where a lane marking is detected in the image of the surroundings, even though no such marking actually exists. For example, shadows of objects can be mistakenly identified as lane markings.

[0004] Furthermore, a parking space recognition system for detecting a parking space or a parking space within a parking space is known from the prior art (see EP 2788245 A1). In particular, the parking space recognition system is designed to detect the dimensions, position, and orientation of parking spaces in a parking lot with multiple spaces. Alternatively or additionally, the parking space recognition system can be designed to actively (for example, by intervening in a driving function of the vehicle) or passively (for example, by providing driving instructions to a driver of the vehicle) support a parking maneuver of the vehicle in the parking space.

[0005] The present application concerns enabling reliable and safe detection of road markings, particularly for driver assistance systems and / or parking space detection systems. Specifically, the present invention aims to make the detection of road markings more robust against so-called false positive detection errors.

[0006] This problem is solved according to the invention by the subject matter of the independent claims. Advantageous embodiments with expedient further developments are the subject matter of the dependent claims.

[0007] A first aspect of the invention relates to a method for detecting a road marking for a vehicle. In other words, the method is intended for use in a vehicle. In the course of the method, an image of the vehicle's surroundings is first provided. A potential road marking is identified in this image. In particular, the potential road marking is identified according to any method known from the prior art. An example of this was explained at the outset. Subsequently, the extent of the potential road marking perpendicular to a principal direction of extension of the potential road marking is determined as the line width of the potential road marking. The principal direction of extension of the potential road marking is, in particular, the direction in which the road marking has its greatest extent.Another criterion is that the extent of the potential road marking must be greater by a predetermined factor than its extent in other spatial directions. Therefore, before determining the extent of the potential road marking perpendicular to its main direction of extension, the main direction of extension can first be determined. For example, the main direction of extension can be determined by checking in which direction the road marking has its greatest extent. The main direction of extension can represent the line of the road marking.

[0008] In a further step, the road marking is detected. This is done by validating the potential road marking based on its line width. The validation is performed only if the line width changes by no more than a predetermined amount along the main direction of extension. A plurality of values ​​for the line width along the main direction of extension are determined, and the potential road marking is validated as plausible only if individual values ​​from this plurality of line width values ​​differ from each other by no more than a predetermined amount. Thus, the invention is based on the idea that the potential road marking, which is initially determined from the image of the surroundings using a conventional method, is validated before it is recognized or considered as the actual road marking. According to the invention, this validation of the potential road marking is performed based on its line width.For example, the line width of the road marking must meet a predetermined criterion for the potential road marking to be validated as plausible. If the line width of the potential road marking does not meet the predetermined criterion, the potential road marking may be considered a false positive. Generally, it may be stipulated that the potential road marking is only recognized as a road marking if it is validated as plausible, with plausibility being checked based on the line width. If the potential road marking cannot be validated as plausible, or if it is identified as implausible, it may be rejected as a false positive.In other words, potential road markings identified using a conventional method are subjected to a plausibility check in order to filter out falsely positive potential road markings.

[0009] This validation of the potential road marking based on its line width stipulates that the potential road marking is validated as plausible only if the line width changes by no more than a predetermined amount along its main direction of extension. In other words, it is checked whether the line width of the potential road marking remains constant, at least within the limits defined by this predetermined amount. This takes into account the fact that road markings typically have a constant line width along their length or main direction of extension. Therefore, the potential road marking can be rejected as implausible if its line width changes by more than the predetermined amount along its main direction of extension.In contrast, the potential lane marking can be validated as plausible if the change in its line width along the main direction of extension does not exceed the predetermined limit. This allows for further improved detection of false positives for potential lane markings.

[0010] This validation of potential road markings based on their line width involves determining multiple line width values ​​along the main direction of travel. The potential road marking is validated as plausible only if individual line width values ​​within this multiple range differ from each other by no more than a predetermined amount. For example, a line width value can be determined at periodic intervals along the main direction of travel as part of this multiple range. Subsequently, it is checked whether the multiple line width values ​​differ from each other by no more than the predetermined amount. In this way, potential road markings whose line width is determined along the main direction of travel can be particularly advantageously identified as plausible or implausible.

[0011] For the purposes of this application, the word "image" also refers in particular to an image signal, for example a digital or analog image signal or a video signal. The image may be a single image, a sequence of images, or a moving image, in particular a video.

[0012] The following briefly explains two exemplary methods for determining the line width of the potential lane marking: For example, an image gradient is created in the surroundings perpendicular to the main direction of extension across the lane marking. Two maxima of the image gradient along a line perpendicular to the main direction of extension across the potential lane marking can be considered the boundary points of the potential lane marking. The distance between the two maxima along such a line perpendicular to the main direction of extension can be determined as the line width, or a value for the line width. Alternatively, the line width can be determined using a binary image generated from the surroundings image. The generation of the binary image is explained in more detail below.In the binary image, two points that both lie on a line perpendicular to the principal direction of extension, and at which a first color transitions into a second color or vice versa, can be considered boundary points. The distance between these two boundary points can be determined as the line width or as the value for the line width.

[0013] It is self-evident that the present method does not require the processing of an entire road marking or potential road marking at once. For example, the image of the surroundings may only depict a portion of the road marking. Alternatively or additionally, only sections of the road within a specific image area may be of interest. Therefore, it is possible that the method is not performed on the entire road marking but only on fragments or parts thereof. For the sake of simplicity, however, the present invention always refers to "the road marking" or "the potential road marking," regardless of whether this refers to the entire road marking or only a part of it.

[0014] A further development provides that the image of the surroundings is generated from at least one raw image from at least one camera of the vehicle, wherein a perspective of the at least one raw image is transformed into a bird's-eye view. In particular, the generation of the image of the surroundings based on the raw images of the multiple cameras can be modeled on a so-called surround-view system. The transformation into a bird's-eye view is particularly advantageous in this case, since no perspective distortion of the line width occurs. Advantageously, within the scope of the present invention, the bird's-eye view is thus oriented perpendicular to a ground plane, for example, a roadway on which the vehicle is positioned.

[0015] One advanced method involves generating the image of the surroundings as a binary image. Alternatively or additionally, a binary image can be generated from the image of the surroundings at any step of the process. The binary image is specifically an image that contains only two different colors. The term "colors" is to be understood broadly here; in particular, black and white are also considered colors in the context of the binary image. When generating the binary image, each pixel of an initial image is assigned either a first color or a second color according to a predetermined rule. By appropriately selecting the parameters for generating the binary image, a potential lane marking can be particularly well highlighted in the image of the surroundings or in the binary image itself. This facilitates the determination of the main direction of extension and / or the line width.For example, the first color is white and the second color is black.

[0016] Further training stipulates that the potential road marking is validated as plausible only if the line width exceeds a first predetermined width value and / or falls below a second predetermined width value. In other words, the potential road marking can be validated, at least partially, based on a simple width filter. The first predetermined width value can, for example, specify a lower limit for the line width, whereby the potential road marking is considered implausible if its line width falls below the first predetermined width value. Alternatively or additionally, the second predetermined width value can represent an upper limit for the line width, whereby the potential road marking is considered implausible if its line width exceeds the second predetermined width value.In other words, it can be stipulated that only potential road markings whose line width lies within the interval between the first and second predetermined width values ​​are considered plausible. Accordingly, it can be stipulated that only potential road markings whose line width lies within this interval are recognized as (actual) road markings. The first predetermined width value and / or the second predetermined width value can, for example, be specified country-specifically and adapted to common line widths of road markings in the respective country. This type of width filter provides a particularly simple way to filter out implausible potential road markings.

[0017] Whether the individual values ​​for the line width differ from each other by no more than the predetermined amount can be checked, for example, by examining the variance of the majority of line width values. For instance, a variance of the majority of line width values ​​is determined. The predetermined amount by which the majority of line width values ​​may differ from each other to be considered plausible can, in this case, be given by a predetermined limit value for the variance. In other words, a variance of the majority of line width values ​​is determined, where the predetermined amount is given by a predetermined limit value for the variance. In this case, the potential road marking can only be validated as plausible if the variance of the majority of line width values ​​does not exceed the predetermined limit value for the variance.The variance provides a particularly simple measure for determining the deviation of the majority of values ​​for line width from each other.

[0018] If the image of the surroundings, or the edges of the potential lane markings within the image, are highly noisy, then even with a constant line width for the potential lane markings, a high variance can occur for the majority of line width values. Therefore, especially in the case of highly noisy images, it is advantageous to perform a regression analysis for the majority of line width values. Accordingly, a further development proposes fitting a regression line to the majority of line width values. The slope of the regression line can then be determined as a measure of the deviation between the majority of line width values.For example, a mathematical graph is virtually generated as the basis for the regression line, in which an x-axis represents the longitudinal path of the potential lane marking along its principal direction of extension, and a y-axis represents the respective values ​​for the line width. Of course, actually generating such a mathematical graph is not necessary; a similar evaluation is possible without creating the graph and is therefore also part of the present invention. The slope of the regression line represents a change in the line width along the principal direction of extension of the potential lane marking. For example, a slope of zero on the regression line means that the line width is constant along the principal direction of extension. The slope of the regression line can, in a sense, be considered the "divergence" of the lane marking.

[0019] The regression line can be calculated, for example, using the least squares method. This method provides a particularly accurate representation of the individual line width values ​​from the majority of values. To reduce the susceptibility of the least squares method to pixel defects and the resulting large outliers in the majority of line width values, the RANSAC algorithm can be additionally applied. RANSAC stands for random sample consensus. It is a procedure for filtering outliers and gross errors from a set of measurements. Therefore, the RANSAC algorithm can be used first, followed by the least squares method, to establish the regression line.

[0020] For example, a predetermined interval can be specified for the slope of the regression line. For instance, it may be stipulated that the potential lane marking is validated as plausible only if the slope of the regression line lies within the predetermined interval. Accordingly, it may be stipulated that the potential lane marking is recognized as an (actual) lane marking only if the slope of the regression line constructed for it lies within the predetermined interval. Conversely, the potential lane marking can be rejected as implausible if the slope of the regression line constructed for it lies outside the predetermined interval.

[0021] Alternatively or additionally, according to another method, a plurality of boundary points for the potential road marking can be determined on both sides of the potential road marking along its main direction of extension. Subsequently, it can be determined from these two pluralitys of boundary points whether the line width changes along the main direction of extension of the potential road marking. The boundary points can be determined analogously to the values ​​for the line width by identifying the point with the greatest image gradient perpendicular to the main direction of extension. Alternatively, the boundary points can be defined, again analogously to the values ​​for the line width, by the boundary between the first and second color values ​​in the binary image. In particular, the distance between two boundary points at the same height along the main direction of extension corresponds to a value for the line width.

[0022] In a further refinement, it can be provided that a regression line is fitted to each flank of the potential lane marking, passing through the respective plurality of boundary points, and is then considered the respective boundary of the lane marking. The two regression lines on the two flanks of the potential lane marking can be adjusted to the respective boundary points of each flank, analogous to the regression line that is adjusted to the plurality of values ​​for the line width. Therefore, the adjustment of the regression lines will not be explained again here. In other words, the respective boundary of the lane marking is determined by the two regression lines on the flanks of the potential lane marking. In this way, the boundary of the lane marking can be reliably determined even in a noisy image or a low-resolution image.In particular, it can be stipulated that the potential road marking is validated as plausible only if the two edges of the road marking deviate from a parallel alignment by no more than a predetermined amount. In other words, it is checked whether the edges of the road marking are at least approximately parallel to each other. If this is not the case, the potential road marking can be rejected as implausible. This provides an alternative method for verifying whether the line width of the potential road marking is constant along its main direction of extension.

[0023] The main direction of extension of the potential road marking can be determined using a skeletonization method. For example, the potential road marking, even in part, can be subjected to a skeletonization algorithm. Examples of skeletonization algorithms include fuzzy skeletonization and Zhang Suen skeletonization. A skeleton of the potential road marking obtained in this way can be defined as the main direction of extension.

[0024] Further training involves validating potential road markings using a support vector machine. A support vector machine is also frequently referred to as a "support vector machine" in German. Such a machine can combine several of the aforementioned methods. For example, the width filter, the calculation of the variance across multiple line width values, and the evaluation of regression lines—either for the line width values ​​and / or for the edges of the road marking—can be combined. Dependencies between the different evaluation methods can be automatically determined by the support vector machine based on training data. Thus, the reliable detection of road markings can be further improved using a support vector machine.

[0025] Once the lane markings have been validated as plausible, the position of a parking space can be identified within the surrounding image based on these validated lane markings. In other words, the validated lane markings are used to determine the position of a parking space. Specifically, the respective positions of parking spaces within a parking lot can be identified based on the validated lane markings, or multiple validated lane markings. By first validating the potential lane markings, a high degree of reliability is ensured for parking space identification.

[0026] A second aspect of the invention relates to a control device for detecting a lane marking for a vehicle, comprising a receiver unit for receiving an image of the vehicle's surroundings and a processing unit for determining a potential lane marking in the image of the surroundings, as well as for determining the extent of the potential lane marking transverse to a principal direction of extension of the potential lane marking as the line width of the potential lane marking, and for detecting the lane marking by validating the potential lane marking as plausible based on its line width. In particular, the control device is configured to execute a method for detecting a lane marking according to one or more of the embodiments described above. The control device can be configured to execute program code of a computer program product described below.

[0027] A third aspect of the invention relates to a parking space detection system for recognizing a parking space, comprising at least one camera for capturing an image of the surroundings and a control unit of the type mentioned above. The parking space detection system can further be configured to recognize one or more parking spaces in the parking lot based on the lane markings. Preferably, the parking space detection system is further configured to recognize whether the detected parking spaces are free or occupied. In a further embodiment, the parking space detection system can be configured to assist with the parking maneuver of the vehicle. To assist with the parking maneuver, the parking space detection system can, for example, control a driving function (steering, accelerator, or brake) of the vehicle and / or issue driver instructions to the driver of the vehicle.

[0028] Furthermore, the invention includes a driver assistance system for detecting lane markings, which is configured to perform a method according to one or more of the embodiments described above. The driver assistance system can, for example, be configured as a lane keeping assistant or an assistant for autonomous vehicle control. The driver assistance system according to the invention can include the control unit according to the invention.

[0029] Furthermore, the invention comprises a vehicle, in particular a motor vehicle, with the aforementioned parking space recognition system and / or the aforementioned driver assistance system. Preferably, the vehicle is designed as a motor vehicle, for example as a passenger car or truck.

[0030] Another aspect of the invention relates to a computer program product with program code means stored in a computer-readable medium for carrying out the method for detecting a road marking of the type described above when the computer program product is executed on a processor of an electronic control unit. Accordingly, the invention also includes a computer-readable medium on which such a computer program product is stored.

[0031] Advantageous embodiments and further developments of the above-mentioned method also apply analogously to the control unit, the computer program product, the driver assistance system, and the parking space detection system according to the invention, and vice versa. For this reason, further developments and features, as well as their advantages, which have already been disclosed within the scope of the method according to the invention, are not disclosed again here for the control unit and the camera device according to the invention.

[0032] Further features of the invention are evident from the claims, the figures, and the description of the figures. The features and combinations of features mentioned above in the description, as well as those subsequently mentioned in the description of the figures and / or shown in the figures alone, are not only usable in the combinations specified but also in other combinations without departing from the scope of the invention. Thus, embodiments that are not explicitly shown and explained in the figures but can be derived and generated from the explained embodiments by separate combinations of features are also to be considered as encompassed and disclosed by the invention. Embodiments and combinations of features that do not exhibit all the features of an originally formulated independent claim are also to be considered disclosed.Furthermore, embodiments and combinations of features, in particular those set out above, are to be considered disclosed which go beyond or deviate from the combinations of features set out in the cross-references of the claims.

[0033] This shows: Fig. 1 a vehicle with a parking space recognition system and a road marking in a schematic top view; Fig. 2 the road marking Fig. 1 in a larger view; Fig. 3 Fig. 2 as a schematic binary image; Fig. 4 an exemplary gradient course perpendicular to a main extension direction of the road marking; and Fig. 5 an exemplary regression line over a plurality of values ​​for the line width.

[0034] Fig. Figure 1 shows a schematic top view of a vehicle 1 equipped with a parking space detection system 8 and a driver assistance system 9. The parking space detection system 8 comprises at least one camera 3, in this case four cameras 3. A first camera 3 is located at the front of the vehicle 1. A second camera 3 is located at the rear of the vehicle 1. For example, the first camera 3 can be located in the area of ​​a radiator grille or behind a windshield of the vehicle 1. For example, the second camera 3 can be located at the rear of the vehicle 1, for example, in the area of ​​a license plate holder or a trunk lid. A third and a fourth camera 3 are located on the respective side mirrors 10 of the vehicle 1.In this case, the third camera 3 is mounted on the left-hand exterior mirror of vehicle 1 (relative to vehicle 1), and the fourth camera 3 is mounted on the right-hand exterior mirror 10 of vehicle 1 (relative to vehicle 1). A detection area 16 of the first camera 3 can be oriented forward (relative to vehicle 1). A detection area 17 of the second camera 3 can be oriented rearward (relative to vehicle 1). Detection areas 18 and 19 of the third and fourth cameras 3 can be oriented laterally (relative to vehicle 1), specifically to the left and right. The detection areas 16, 17, 18, and 19 each represent sub-areas of the surrounding area U of vehicle 1.

[0035] In Fig. Figure 1 also shows a lane marking 2 in the vicinity U of vehicle 1. Lane marking 2 and further lane markings 29 designate a parking space 28. Lane marking 2 and the further lane markings 29 are, in particular, lines painted on a road surface. The lines forming lane marking 2 and the further lane markings 29 can be, for example, white or yellow. The parking space recognition system 8 is designed to recognize parking space 28 based on lane markings 2 and 29. To ensure safe and reliable recognition of parking space 28, reliable recognition of lane markings 2 and 29 is also necessary. The recognition of lane marking 2 is explained in more detail below.

[0036] Fig. Figure 2 shows the lane marking 2 enlarged in certain areas. First, the lane marking 2 is captured by the cameras 3. From the raw images of the cameras 3, an image of the vehicle 1's surroundings U is generated. In this image of the surroundings U, the vehicle 1's surroundings U is shown from a bird's-eye view. Thus, the lane marking 2 is shown vertically from above in the image of the surroundings U. Fig. In section 2, a coordinate system x, y is also drawn, which is related to the Fig. 4 and the Fig. 5 is displayed.

[0037] In the image of the environment U, lane marking 2 is identified as a potential lane marking 4. For example, the representation of lane marking 2 in the image of the environment U contains a pattern that corresponds to a predetermined pattern for a potential lane marking 4. In other words, the potential lane marking 4 in the image of the environment U can be recognized by pattern recognition, which identifies potential lane markings 4 based on a pattern specific to lane markings 2, 19. For example, to recognize the potential lane marking 4, an image gradient G in the image of the environment is evaluated. By recognizing the predetermined pattern in the image gradient G, the potential lane marking 4 can be identified.

[0038] This potential road marking 4 must be validated as plausible in a further step. Here, the plausibility of the potential road marking 4 is validated based on its line width 20. The line width 20 represents an extent of the potential road marking 4 perpendicular to its main direction of extension 22. Therefore, it may be necessary to first determine the main direction of extension 22 of the potential road marking 4 in a further step. For example, the potential road marking 4 is determined using a skeletonization method. In this case, a skeleton of the potential road marking 4 can be defined as the main direction of extension 22.

[0039] The extent of the potential lane marking 4 perpendicular to the main direction of extension 22, i.e., the line width 20, is then determined. For this purpose, the image gradient 6 can be evaluated, for example, along straight lines 23 oriented perpendicular to the main direction of extension 22. Fig. Figure 4 shows this image gradient G as an example along one of the lines 23 in Fig. 2. Here, two extreme values ​​of the image gradient are recorded as boundary points 24 of the potential lane marking 4. The distance between these two boundary points 24 on the same straight line 23 can be recorded as a value 21 for the line width. In this case, a plurality of values ​​21 for the line width 20 are recorded. As can be seen from Fig. As can be seen in Figure 2, this is done at specific intervals 34 along the main extension direction 22. In other words, a value 21 for the line width 20 is determined at each interval 34. Values ​​at positions x1, x2, and x3 along the main extension direction and along the x-direction, respectively, are shown as examples. Fig. 2. The intervals 34 can be chosen randomly, fixed, or predetermined within the procedure. The intervals 34 can be the same or different from each other.

[0040] Alternatively, the values ​​21 for the line width or the boundary points 24 can be determined in a binary image 3 of the environment U. Fig. 3 shows the same image as Fig. 2 is converted into a binary image 32. In the binary image 32, each pixel is assigned one of two color values. For example, the color values ​​of the binary image 32 can be black and white. For example, a first color value, in particular black, can be assigned to areas 30. Furthermore, a second color value, in particular white, can be assigned to an area 31. By cleverly choosing the assignment rule for assigning the color values, the binary image 32 can be generated such that area 31 represents the potential lane marking 4. In other words, area 31 can represent the dimensions and position of the potential lane marking 4. The boundary points 24 are then defined by the boundaries between areas 30 and 31.Subsequently, values ​​21 for the line width 20 can also be determined in the binary image by determining the distance between two boundary points 24, which both lie on a common straight line 23, which is oriented perpendicular to the principal extension direction 22.

[0041] The edges 26 of the potential road marking 4 can be determined by fitting a regression line through the respective edge points 24 of a flank 34 of the potential road marking 4. In other words, the edges 26 can be determined by fitting a respective regression line to the respective edge points 24 of the flanks 34. In this case, the potential road marking 4 can be rejected as implausible if the two edges 26 of the potential road marking 4 deviate from being parallel to each other by more than a predetermined amount. In other words, it is determined whether the two edges 26 are at least approximately parallel to each other in order to validate the potential road marking 4 as plausible.

[0042] Another method to validate the potential road marking 4 as plausible shows Fig. 5. This contains several values ​​21 for the line width 20 along the x-axis. Fig. 2 plotted. The x-axis runs parallel to the main extension direction 22 of the potential road marking 4. According to Fig. A regression line 27 is fitted to the values ​​21 for the line width 20. A slope can then be determined for the regression line 27. This slope of the regression line 27 is then a measure of a change in the line width 20 along the principal extension direction 22. In other words, if the slope of the regression line 27 is zero, the potential lane marking 4 has a constant line width 20. For example, it is intended that the potential lane marking 4 is validated as plausible and recognized as the actual lane marking 2 if the slope of the regression line 27 deviates from zero by at most a predetermined amount. For example, the predetermined amount can be given by an interval within which the slope of the regression line 27 may lie.If the slope of the regression line 27 is outside the interval or if the slope of the regression line 27 deviates from zero by more than the predetermined amount, the potential road marking 4 can be discarded as a false positive road marking.

[0043] Advantageously, the regression lines can be fitted by the boundary points 24, which define the edges 26 of the potential lane marking 4, and / or the regression line 27 can be fitted using a combination of the RANSAC algorithm and the least squares method. This allows the RANSAC algorithm to first identify large outliers that deviate significantly from the other values. These outliers are then excluded from the fit of the respective regression lines using the least squares method. Such outliers can arise, for example, from inaccurately measured boundary points 24 due to a noisy or low-resolution image of the surroundings U. Therefore, by constructing the regression lines, particularly in the manner described, the susceptibility to errors caused by low-resolution images and noise is reduced.

[0044] Alternatively or additionally, a variance can be determined for the values ​​21 for the line width 20. The variance can then be considered a measure indicating the deviation of the majority of values ​​21 for the line width 20 from one another. For example, it can be stipulated that the potential lane marking 4 is rejected as implausible if the variance is greater than a predetermined limit for the variance. Accordingly, it can be stipulated that the potential lane marking 4 is recognized as the (actual) lane marking 2 if the variance of the values ​​21 is less than the predetermined limit for the variance.

[0045] Alternatively or additionally, the plausibility of the potential lane marking 4 can be validated using a simple width filter. For example, one or more values ​​21 for the line width 20 are compared with a first predetermined width value and / or a second predetermined width value. In this case, the potential lane marking 4 can be rejected as implausible if the respective value 21 for the line width 20 is smaller than or larger than the first predetermined width value. In other words, the potential lane marking 4 can only be considered plausible if the respective value 21 for the line width 20 lies within an interval between the first and second predetermined width values.

[0046] Now back to… Fig.In the vehicle 1, a control unit 5 is provided for detecting the lane marking 2. The control unit 5 includes a receiver 50 for receiving the image of the vehicle's surroundings U. Furthermore, the control unit 5 includes a processing unit 51 for determining the potential lane marking 4 in the image of the surroundings U. The processing unit 51 is also configured to determine the extent of the potential lane marking 4 perpendicular to its main direction of extension 22 as the line width 20 of the potential lane marking 4. Additionally, the processing unit 51 is configured to recognize the lane marking 2 as plausible by validating the potential lane marking 4 based on its line width 20. In particular, the control unit 5, or the processing unit 51, is configured to perform some or more of the previously described procedural steps.

[0047] The control unit 5 can include a so-called support vector machine as part of the computing unit 51. The support vector machine can be configured to perform the detection of the lane marking 2 by validating the potential lane marking 4 using one or more of the steps or evaluation methods described above: evaluating the variance of the majority of values ​​21 for the line width 20, evaluating the boundary points 24, in particular by fitting a respective regression line to define the boundaries 26, fitting and evaluating the regression line 27 using the values ​​21 for the line width 20, and simple width filtering with the first specified width value and / or the second specified width value.This support vector machine can be trained using training data, which might include, for example, a large number of sample images of the environment U with sample road markings. In this way, the individual evaluation methods can be refined and / or the relationships between several evaluation methods can be optimally combined. In other words, the relationships between several evaluation methods can be determined and / or optimized by the support vector machine using the training data.

Claims

[1] Method for detecting a road marking (2) for a vehicle (1), by - Providing an image of the vehicle's surroundings (U) using a camera (3), - Identifying a potential road marking (4) in the image of the surroundings (U), - Determining the extent of the potential road marking (4) perpendicular to a main extension direction (22) of the potential road marking (4) as the line width (20) of the potential road marking (4), and - Recognition of the road marking (2) by validating the potential road marking (4) as plausible based on its line width (20), only if the line width (20) changes along the main extension direction (22) by no more than a predetermined amount, wherein - a plurality of values ​​(21) for the line width (20) along the main extension direction (22) is determined, and - the potential road marking (4) is validated as plausible at most if individual values ​​from the majority of values ​​(21) for the line width (20) differ from each other by no more than a predetermined amount and - a regression line (27) is fitted to the majority of values ​​(21) for the line width (20) and a slope of the regression line (27) is determined as a measure of the deviation of the majority of values ​​(21) for the line width (20) from each other. [2] Method according to claim 1, characterized by , that the image of the environment (U) is generated from at least one raw image from at least one camera (3) of the vehicle (1), wherein a perspective of the at least one raw image is transformed into a bird's-eye view. [3] Method according to any one of the preceding claims, characterized by , that the image of the environment is converted into a binary image (32). [4] Method according to any one of the preceding claims, characterized by, that the potential road marking (4) is validated as plausible at most if the line width (20) exceeds a first predetermined width value and / or the line width (20) falls below a second predetermined width value. [5] Method according to any one of the preceding claims, characterized by , that a variance of the majority of values ​​(21) for the line width (20) is determined and the predetermined measure is given by a predetermined limit for the variance. [6] Method according to any one of the preceding claims, characterized by , that - on both flanks (34) of the potential road marking (4) along the main extension direction (22) a plurality of edge points (24) for the potential road marking (4) is determined, and - based on the two plurals at boundary points (24), it is determined whether the line width (20) changes along the main extension direction (22) of the potential road marking (4). [7] Method according to claim 6, characterized by , that on both flanks (34) of the potential road marking (4) a respective compensating line is laid through the respective plurality of edge points (24) for the potential road marking (4) and is assumed to be the respective edge (26) of the road marking (2). [8] Method according to any one of the preceding claims, characterized by , that the main extension direction (22) of the potential road marking (4) is determined by means of a skeletonization procedure. [9] Method according to any one of the preceding claims, characterized by , that the validation of the potential road marking (4) is carried out using a support vector machine. [10] Method according to any one of the preceding claims, characterized by , that based on the validated road marking (2) a position of a parking space (28) is recognized in the image of the surroundings (U). [11] Control device (5) for detecting a road marking (2) for a vehicle (1), with - a receiving unit (50) for receiving an image of an environment (U) of the vehicle (1), and - a computing unit (51) for determining a potential road marking (4) in the image of the environment (U), as well as for determining an extent of the potential road marking (4) transverse to a principal extension direction (22) of the potential road marking (4) as line width (20) of the potential road marking (4) and for recognizing the road marking (2) by validating the potential road marking (4) as plausible based on its line width (20), if the line width (20) changes along the principal extension direction (22) by no more than a predetermined amount. [12] Parking space recognition system (8) for a vehicle (1) for recognizing a parking space, with - at least one camera (3) for capturing an image of the surroundings (U), and - a control device (5) according to claim 12. [13] Driver assistance system (9) for detecting a road marking (2) which is configured to perform a method according to any one of claims 1 to 10. [14] Computer program product comprising program code means stored in a computer-readable medium for carrying out the method for detecting a road marking (2) according to any one of claims 1 to 10 when the computer program product is executed on a processor of an electronic control device (5).

Images