Method for generating pixel-level segmentation label of lane line in intelligent vehicle lamp

Abstract

The invention discloses a method for generating a pixel-level segmentation label of a lane line in an intelligent vehicle lamp, and the method comprises the following steps: 1, obtaining a 2D image containing the lane line; a second step of manually marking key points; a third step of fitting a center curve; 4, determining a y coordinate minimum value and a y coordinate maximum value; 5, calculating the width of a lane line of any point on the center curve; and step 6, generating a lane line segmentation label. According to the pixel-level segmentation label generation method for the lane line in the intelligent vehicle lamp, a mode of combining point labeling with line fitting and width calculation is used for replacing direct manual segmentation connected domain labeling, and the problems of high cost and low efficiency in lane line labeling are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of lane line detection, and in particular to a method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights. Background Art

[0002] Lane segmentation is a computer vision task that belongs to the application field of image segmentation. It is mainly used in autonomous driving, assisted driving systems and intelligent transportation systems. Its core goal is to detect and segment the lane area from the road image, which is crucial for the light blanket projection function in smart headlights. The lane line is identified by the algorithm and accurately segmented from the image or video, which makes it easier for the vehicle to determine its relative position to the lane and ensure the accuracy of the headlight projection area.

[0003] In order to train a lane segmentation model, a large number of well-labeled lane datasets are usually required. The labeling process involves annotating the lane lines in an image or video with pixel-level accurate labels. Each pixel is either labeled as part of the lane line, or as background or other objects.

[0004] In addition to using pixel-level precise labels to distinguish lane lines from the background, other attributes of the marked lane lines need to be added, such as lane line type, color, occlusion and other attributes. Lane line marking faces many challenges. Different weather and lighting conditions will affect the visual performance of lane lines. Lane lines vary in style in different road scenes, and may have different numbers, colors and shapes. Lane lines on old roads are worn, faded or irregular. These challenges make lane line marking more difficult. Currently, the segmentation and annotation methods are divided into pure manual annotation and interactive semi-automatic annotation:

[0005] 1. Pure manual labeling is that the labeler describes the outline of the lane line in the image in the form of "points" to form a connected domain of the outer edge of the lane line. The advantage of this labeling method is that the label is accurate. By describing the outer contour of the lane line by "points", the lane line width can be obtained; the disadvantage is that the labeling method that describes the outer contour requires a large number of points to be found, which is costly and inefficient.

[0006] 2. Interactive semi-automatic labeling refers to the use of a general segmentation model (such as SAM) to manually provide key position prompts and generate the segmentation connected domain of the target. The advantage of this labeling method is that it requires fewer points to find, saving the process of manually finding the outer edge of the target; the disadvantage is that it cannot completely segment the dotted lane lines and the lane lines that are partially blocked by obstacles, and the segmentation contour is not as accurate as manual.

[0007] For lane segmentation, it is not possible to use a semi-automatic interactive method to perform segmentation and annotation, such as Figure 2As shown in the figure, for a lane line composed of dotted lines, the road part between the two dotted lines also needs to be marked as a lane line. However, the general segmentation model cannot treat these two clearly separated areas as the same target, which results in the lane line segmentation and marking can only be done manually, resulting in high labeling costs and a long labeling cycle. In other words, if the segmentation method is used for lane line labeling, it will cause problems of high labeling costs and long cycles.

[0008] The existing method for obtaining the real labels of lane lines is to manually mark the outer edge contours of the lane lines to form a closed connected domain, and use the connected domain as the segmentation label of the lane lines.

[0009] A lane line marking method, device, electronic device and storage medium are disclosed in a Chinese patent (application number CN202310835344.3, announcement date 2023.10.13), including: obtaining multiple first target images collected at a first moment; fitting the collected trajectory points in each of the first target images respectively to obtain the fitted sub-lane lines corresponding to the first target image; connecting the fitted sub-lane lines in each of the first target images to obtain the fitted lane lines; querying the high-precision map database according to the fitted lane lines to obtain the second target image saved at a second moment; wherein the second moment is earlier than the first moment; comparing the lane line deviation between the fitted lane line and the target lane line in the second target image to obtain the lane line to be updated in the second target image; marking the lane line to be updated.

[0010] A Chinese patent (application number CN202011046531.6, announcement date 2020.09.29) discloses an image segmentation and annotation method, a model training method, a device and a storage medium, including: obtaining a touch trajectory; according to the touch trajectory, obtaining a touch point and the number of touches corresponding to the touch point; calculating the attention weight of the touch point according to the number of touches; and annotating the image according to the touch point and the attention weight.

[0011] Obviously, the marking methods involved in the above two patent documents have the problems of high cost and low efficiency. Summary of the invention

[0012] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0013] To this end, the present invention proposes a method for generating pixel-level segmentation labels for lane lines in intelligent headlights. By using "point" labeling combined with line fitting and width calculation to replace direct manual segmentation of connected domain labeling, the problems of high cost and low efficiency in lane line labeling are solved.

[0014] According to an embodiment of the present invention, a method for generating pixel-level segmentation labels for lane lines in an intelligent vehicle light comprises the following steps:

[0015] Step 1: Obtain a 2D image containing lane lines: Use the camera in the smart headlight to collect a 2D image of the actual lane lines;

[0016] Step 2: Manually mark key points: In the acquired 2D image, manually mark the plane rectangular coordinate system XOY, and for one of the lane lines, manually mark a series of key points describing the x-axis center of the lane line;

[0017] Step 3: Fitting the center curve: Based on the manually marked key points, use the least squares method to fit the center curve of the lane line to obtain a center curve that describes the x-axis center of the lane line;

[0018] Step 4, determine the minimum and maximum y coordinates: Set the projection coordinates of the farthest point where the road intersects the sky in the 2D image as the vanishing point, and determine the y coordinate of the road "vanishing point" in the 2D image through camera calibration. At this time, it is the minimum y coordinate of the lane line in the 2D image; and the width w_max of the nearest lane line captured by the camera in the 2D image, w_max is the maximum width of the lane line on the x-axis in the 2D image and the y-axis coordinate of the lane line is the largest at this time;

[0019] Step 5: Calculate the lane line width at any point on the center curve: According to the actual lane line width and the camera imaging principle, it can be known that the lane line width in the 2D image changes linearly. The lane line width of each point (x, y) on the fitted center curve is calculated using formula (1). Formula (1) is:

[0020]

[0021] The meanings of the symbols in formula (1) are as follows:

[0022] w represents the lane width corresponding to any point on the center curve;

[0023] y represents the y-axis coordinate value corresponding to any point on the central curve;

[0024] y_min represents the minimum y coordinate of the lane line in the 2D image;

[0025] y_max represents the maximum y coordinate of the lane line in the 2D image;

[0026] w_max represents the maximum width of the lane line on the x-axis in the 2D image;

[0027] Step 6: Generate lane line segmentation labels: Calculate the width of the lane line at different y coordinates, and then calculate the x coordinates of the left and right sides of the lane line at each y coordinate through the x-axis center curve of the lane line. Connect each generated point to form a closed connected domain, that is, generate the lane line segmentation label.

[0028] The beneficial effects of the present invention are as follows:

[0029] 1. The lane marking method described in the present invention changes from manually finding the contour points of the lane lines to finding the key points describing the x-axis center of the lane lines, which reduces the number of marking steps and the time required to mark a lane line, and can effectively save the marking cost in actual marking.

[0030] Second, according to the camera imaging principle and the actual attributes of the lane line width, the present invention proposes to assign a value to the actual lane line width by using the lane line width obtained by calibration, thereby solving the lane line width problem;

[0031] 3. The segmentation result in the present invention is not obtained by manually marking contour points, but is obtained by fitting and adding a preset width, thus avoiding the manual process of finding "points".

[0032] According to one embodiment of the present invention, in step 2, the process of manually marking the plane rectangular coordinate system XOY is: taking the upper left corner of the 2D image as the origin, horizontally to the right as the positive direction of the x direction, and vertically downward as the positive direction of the y direction.

[0033] According to one embodiment of the present invention, in step 6, a point (x, y) on the fitted center curve is known, and the x coordinate of the lane line on the left side of the point is calculated as shown in formula (2):

[0034]

[0035] The meanings of the symbols in formula (2) are as follows:

[0036] x represents the x-axis coordinate value corresponding to any point on the central curve;

[0037] w represents the lane width corresponding to any point on the center curve;

[0038] x 左 Indicates the x-axis coordinate value of the left lane line corresponding to any point on the center curve.

[0039] According to one embodiment of the present invention, in step 6, a point (x, y) on the fitted center curve is known, and the x-coordinate of the lane line to the right of the point is calculated as shown in formula (3):

[0040]

[0041] The meanings of the symbols in formula (3) are as follows:

[0042] x represents the x-axis coordinate value corresponding to any point on the central curve;

[0043] w represents the lane width corresponding to any point on the center curve;

[0044] x 右 Indicates the x-axis coordinate value of the lane line on the right side corresponding to any point on the center curve.

[0045] According to one embodiment of the present invention, in step 1, the lane line 2D image is acquired by a camera installed in a vehicle headlight.

[0046] According to an embodiment of the present invention, the installation position and angle of the camera are fixed after installation calibration.

[0047] According to one embodiment of the present invention, the camera is installed in the front left high-beam headlight of the vehicle.

[0048] According to one embodiment of the present invention, the camera uses a wide-field camera that can capture four lane lines on the road ahead, where the four lane lines are two lane lines on the left side of the vehicle and two lane lines on the right side of the vehicle.

[0049] According to an embodiment of the present invention, in step 2, after the selected key points are connected into a broken line, a lane line is evenly divided in the x direction.

[0050] According to one embodiment of the present invention, when the curvature of the lane line is greater, the number of key points is greater; when the curvature of the lane line is smaller, the number of key points is smaller.

[0051] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 It is the overall flow chart of the present invention;

[0055] Figure 2 This is a schematic diagram of segmenting a lane line composed of dotted lines;

[0056] Figure 3 It is a schematic diagram of the camera imaging the lane line;

[0057] Figure 4 This is a diagram of selecting key points and fitting central curves in a 2D image. Figure 1 ;

[0058] Figure 5 This is a diagram of selecting key points and fitting central curves in a 2D image. Figure 2 . DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] See Figure 1 , Figure 3 , Figure 4 and Figure 5 , the following specifically describes a method for generating pixel-level segmentation labels for lane lines in an intelligent vehicle lamp according to an embodiment of the present invention with reference to the accompanying drawings.

[0061] The method for generating pixel-level segmentation labels for lane lines in an intelligent vehicle lamp of the present invention comprises the following steps:

[0062] Step 1: Obtain a 2D image containing lane lines: Use the camera in the smart headlight to collect a 2D image of the actual lane lines; the image collected by the camera is a projection of the scene in front of the vehicle when the vehicle is driving on the 2D image, and the projection contains lane line information.

[0063] Among them, the 2D image of the lane line is collected by a camera installed in the vehicle headlight. Specifically, the installation position and angle of the camera are fixed after installation and calibration. The camera is installed in the front high-beam headlight on the left front of the vehicle. The camera complies with vehicle regulations and can be a forward-looking camera for assisted driving currently available on the market. The camera can capture all lane lines on the road ahead, which is related to the camera's field of view. For usage scenarios, four lane lines are currently concerned. Taking the vehicle as the reference point, the four lane lines are two lane lines on the left side of the vehicle and two lane lines on the right side of the vehicle. That is, the camera uses a wide-field camera that can capture four lane lines on the road ahead. The four lane lines are two lane lines on the left side of the vehicle and two lane lines on the right side of the vehicle.

[0064] Step 2: Manually mark key points: In the acquired 2D image, manually mark the plane rectangular coordinate system XOY, and for one of the lane lines, manually mark a series of key points describing the x-axis center (center in the x direction) of the lane line.

[0065] The process of manually marking the plane rectangular coordinate system XOY is: take the upper left corner of the 2D image as the origin, horizontally to the right as the positive direction of the x direction, and vertically downward as the positive direction of the y direction. The x-axis center is the average of the maximum and minimum coordinates of a lane line in the x direction when a fixed y value is selected.

[0066] It should be noted that there is no requirement for the number of manually annotated key points, but the polyline formed by connecting the manually annotated key points from small to large y values ​​needs to divide the lane line into two parts in the x direction, that is, after the selected key points are connected into a polyline, the lane line is evenly divided in the x direction. The greater the curvature of the lane line, the more key points there are; the smaller the curvature of the lane line, the fewer key points there are.

[0067] Step 3: Fitting the center curve: Based on the manually marked key points, use the least squares method to fit the center curve of the lane line to obtain a center curve that describes the x-axis center of the lane line;

[0068] Step 4, determine the minimum and maximum y coordinates: Set the projection coordinates of the farthest point where the road intersects the sky in the 2D image as the vanishing point, and determine the y coordinate of the "vanishing point" of the road (the road where the car is traveling) in the 2D image through camera calibration. At this time, it is the minimum y coordinate of the lane line in the 2D image; and the width w_max of the nearest lane line captured by the camera in the 2D image, w_max is the maximum width of the lane line on the x-axis in the 2D image and the y-axis coordinate of the lane line is the largest at this time;

[0069] Step 5: Calculate the lane line width at any point on the center curve: According to the actual lane line width and the camera imaging principle, it can be known that the lane line width in the 2D image changes linearly. The lane line width of each point (x, y) on the fitted center curve can be approximately calculated by formula (1). Formula (1) is:

[0070]

[0071] The meanings of the symbols in formula (1) are as follows:

[0072] w represents the lane width corresponding to any point on the center curve;

[0073] y represents the y-axis coordinate value corresponding to any point on the central curve;

[0074] y_min represents the minimum y coordinate of the lane line in the 2D image;

[0075] y_max represents the maximum y coordinate of the lane line in the 2D image;

[0076] w_max represents the maximum width of the lane line on the x-axis in the 2D image;

[0077] Step 6: Generate lane line segmentation labels: Calculate the width of the lane line at different y coordinates, and then calculate the x coordinates of the left and right sides of the lane line at each y coordinate through the x-axis center curve of the lane line. Connect each generated point to form a closed connected domain, that is, generate the lane line segmentation label.

[0078] In step 6, a point (x, y) on the fitted center curve is known, and the x-coordinate of the lane line to the left of the point is calculated as shown in formula (2):

[0079]

[0080] The meanings of the symbols in formula (2) are as follows:

[0081] x represents the x-axis coordinate value corresponding to any point on the central curve;

[0082] w represents the lane width corresponding to any point on the center curve;

[0083] x 左 Indicates the x-axis coordinate value of the left lane line corresponding to any point on the center curve.

[0084] Given a point (x, y) on the fitted center curve, the x-coordinate of the lane line to the right of the point is calculated as shown in formula (3):

[0085]

[0086] The meanings of the symbols in formula (3) are as follows:

[0087] x represents the x-axis coordinate value corresponding to any point on the central curve;

[0088] w represents the lane width corresponding to any point on the center curve;

[0089] x 右 Indicates the x-axis coordinate value of the lane line on the right side corresponding to any point on the center curve.

[0090] In the application of smart headlights, the perceived image is collected by the camera installed in the vehicle headlight. The installation position and angle of the camera are fixed after installation and calibration. When the vehicle is driving, the camera does not have a large shaking problem. This ensures that the area occupied by the road (the entire road) in the 2D image collected by the camera is almost fixed, and the area where the lane line is located in the 2D image is fixed. According to the description of road markings in the regulations, the lane line has a fixed width, and then according to the imaging principle of the camera, such as Figure 3 , Figure 4 and Figure 5 As shown, it can be concluded that the width of the lane line in the 2D image is negatively correlated with the distance between the lane line and the camera. That is, when the width of the lane line in the 2D image is larger, it indicates that the distance between the lane line and the camera is smaller; or, when the width of the lane line in the 2D image is smaller, it indicates that the distance between the lane line and the camera is larger. The farthest point of the road (the road on which the car is traveling) in the 2D image is described as the "vanishing point". At the vanishing point, the width of the road (the road on which the car is traveling) becomes 0, and the width of the lane line is also 0. Because the camera is fixed, the y coordinate of the vanishing point in the 2D image is also fixed, which makes the area where the lane line appears in the y axis of the 2D image coordinate system fixed. It should be noted that the y coordinate of the vanishing point will be determined by calibration according to the installation angle of the camera.

[0091] The width of the lane line in the 2D image at the nearest point is obtained through camera calibration, and it is known that the width of the lane line is 0 when it reaches the vanishing point in the 2D image. The width change rate of the lane line in the 2D image can be calculated, as shown in formula (1). Then, if the center point of a lane line in the 2D image is known, the width of the lane line can be described by the calculation formula to form a complete segmented area. According to the above ideas, the lane line segmentation annotation is converted to the annotation of the center point of the lane line. The annotation of the point is easier than the segmentation of the connected domain, so it can effectively reduce the annotation cost and reduce the annotation time.

[0092] The method of manually drawing a line in the center of the lane line to mark the lane line has the problem that the manually drawn line is not smooth enough and the efficiency is not high enough. You can select key points on a lane line and then obtain a line describing the center of the lane line by central curve fitting. Depending on the curvature of the lane line, the intervals for selecting key points are also different. It is necessary to ensure that the lane line is evenly divided in the x direction after the selected key points are connected into a broken line. Therefore, the greater the curvature of the lane line, the more key points need to be selected. However, the key points are all at the center of the x-axis of the lane line, which can effectively eliminate the influence of small errors on the lane line detection results.

[0093] The method for generating pixel-level segmentation labels for lane lines in intelligent headlights of the present invention obtains a center line describing the lane line in the 2D image by fitting the key points describing the center of the x-axis of the lane line in the 2D image, and then restores the actual area of ​​the lane line according to the known preset lane line width to obtain the segmentation result of the lane line, which can effectively solve the technical problems in the above-mentioned lane line segmentation and labeling; specifically, in view of the high labeling cost and long cycle caused by using the segmentation method to label the lane line, the lane line labeling method described in the present invention changes from manually finding the contour points of the lane line to finding the key points describing the center of the x-axis of the lane line, the labeling steps are reduced, and the time for labeling a lane line is reduced, which can effectively save the labeling cost in actual labeling; in view of the problem that the lane line width cannot be obtained by using the point method to label the lane line, the present invention proposes to assign the lane line width obtained by calibration to the actual lane line width according to the camera imaging principle and the actual attributes of the lane line width, so as to solve the lane line width problem; in view of the problem that the connected domain of the lane line segmentation and labeling is irregular and it is difficult to find the "point", the segmentation result in the present invention is not obtained by manually marking the contour points, but by fitting and adding the preset width, avoiding the manual "point" finding process.

[0094] The method for generating pixel-level segmentation labels for lane lines in smart headlights of the present invention uses "point" annotation combined with line fitting and width calculation to replace direct manual segmentation and connected domain annotation, thereby solving the problems of high cost and low efficiency in lane line annotation. Compared with directly annotating connected domains, the present invention only needs to annotate a small number of central key points of a lane line, and then obtains the complete lane line segmentation connected domain through calculation, which simplifies the workload of manual annotation, improves annotation efficiency, and can provide a large amount of training data for the perception model of smart headlights more quickly, thereby accelerating the development of smart headlights.

[0095] In the lane line perception task of smart headlights, it is necessary to identify the position of the lane line in front of the vehicle to locate the light blanket projection function. The segmentation model can provide lane line information of pixel sets to provide guarantee for the light blanket projection function. This drives the lane line to provide pixel-level labels. The present invention provides lane line pixel-level labels for perception by marking "points" and then converting "points" into segmentation. The following are the specific steps for generating lane line pixel-level segmentation labels in the present invention:

[0096] Aiming at the intelligent headlight perception scenario, the present invention proposes a lane line segmentation label generation method with lower cost and higher efficiency. It adopts the key point method for manual labeling, and then generates the connected domain describing the outer contour of the lane line as the lane line segmentation label through calculation, which can reduce the cost of obtaining lane line segmentation labels and improve the acquisition efficiency.

[0097] There are three key parts in this invention to ensure that lane segmentation labels can be successfully generated from "point" labels. The three key parts are as follows:

[0098] First, the first key part is to fix the acquisition angle after camera calibration, to ensure that the y coordinate of the "vanishing point" of the road in the same batch of 2D images remains unchanged, and the width of the nearest lane line collected in the image is fixed. Ensuring that the above information in a batch of data remains unchanged can ensure that the generated lane line width conforms to the actual situation and meets the perception requirements of smart headlights.

[0099] The second key part is to fit the x-axis center curve of the lane line by the least squares method. Using the manually annotated key points, the least squares method is used to perform polynomial fitting to generate a center curve that describes the x-axis center of the lane line. The purpose of generating this center curve is to sample the center curve. The x-coordinates of the sampling points are added with half of the calculated lane line width and subtracted with half of the calculated lane line width, as shown in formulas (2) and (3). This will form a connected domain for lane line segmentation, that is, the label for lane line segmentation.

[0100] The third key part is to calculate the actual width of the lane line at each position of the center curve by fixing the road "vanishing point" in the 2D image, the width of the nearest lane line is fixed, and the linear transformation of the lane line width in the 2D image, so as to obtain the segmentation label of the outer contour of the lane line. The width of the lane line at the "vanishing point" is 0 and the y coordinate of the lane line "vanishing point" is known; the maximum width and maximum y coordinate of the lane line in the 2D image are known; the width of the lane line in the 2D image changes linearly. According to the above conditions, the formula is used to calculate the width of the lane line at different y coordinates, and then the x coordinates of the left and right sides of the lane line at each y coordinate are calculated through the x-axis center curve of the lane line. Each generated point is connected to form a closed connected domain, which is the segmentation label of the lane line, that is, the closed center curve describing the outer edge of the lane line.

[0101] It should be noted that the maximum width of the lane line in the 2D image is the width of the lane line when it appears at the bottom edge of the 2D image. This width can be obtained based on the horizontal distance of the bottom edge of the image from the camera in the actual scene and the actual width of the lane line. The maximum y coordinate of the lane line in the 2D image is the height of the 2D image.

[0102] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights, characterized in that: The following steps are involved: Step 1: Obtain a 2D image containing lane lines: Use the camera in the smart headlight to collect a 2D image of the actual lane lines; Step 2: Manually mark key points: In the acquired 2D image, manually mark the plane rectangular coordinate system XOY, and for one of the lane lines, manually mark a series of key points describing the x-axis center of the lane line; Step 3: Fitting the center curve: Based on the manually marked key points, use the least squares method to fit the center curve of the lane line to obtain a center curve that describes the x-axis center of the lane line; Step 4, determine the minimum and maximum y coordinates: Set the projection coordinates of the farthest point where the road intersects the sky in the 2D image as the vanishing point, and determine the y coordinate of the road "vanishing point" in the 2D image through camera calibration. At this time, it is the minimum y coordinate of the lane line in the 2D image; and the width w_max of the nearest lane line captured by the camera in the 2D image, w_max is the maximum width of the lane line on the x-axis in the 2D image and the y-axis coordinate of the lane line is the largest at this time; Step 5: Calculate the lane line width at any point on the center curve: According to the actual lane line width and the camera imaging principle, it can be known that the lane line width in the 2D image changes linearly. The lane line width of each point (x, y) on the fitted center curve is calculated using formula (1). Formula (1) is: The meanings of the symbols in formula (1) are as follows: w represents the lane width corresponding to any point on the center curve; y represents the y-axis coordinate value corresponding to any point on the central curve; y_min represents the minimum y coordinate of the lane line in the 2D image; y_max represents the maximum y coordinate of the lane line in the 2D image; w_max represents the maximum width of the lane line on the x-axis in the 2D image; Step 6: Generate lane line segmentation labels: Calculate the width of the lane line at different y coordinates, and then calculate the x coordinates of the left and right sides of the lane line at each y coordinate through the x-axis center curve of the lane line. Connect each generated point to form a closed connected domain, that is, generate the lane line segmentation label.

2. The method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights according to claim 1, characterized in that: In step 2, the process of manually labeling the plane rectangular coordinate system XOY is: take the upper left corner of the 2D image as the origin, horizontally to the right as the positive direction of the x direction, and vertically downward as the positive direction of the y direction.

3. The method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights according to claim 2, characterized in that: In step 6, a point (x, y) on the fitted center curve is known, and the x-coordinate of the lane line to the left of the point is calculated as shown in formula (2): The meanings of the symbols in formula (2) are as follows: x represents the x-axis coordinate value corresponding to any point on the central curve; w represents the lane width corresponding to any point on the center curve; x 左 Indicates the x-axis coordinate value of the left lane line corresponding to any point on the center curve.

4. The method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights according to claim 2, characterized in that: In step 6, a point (x, y) on the fitted center curve is known, and the x-coordinate of the lane line to the right of the point is calculated as shown in formula (3): The meanings of the symbols in formula (3) are as follows: x represents the x-axis coordinate value corresponding to any point on the central curve; w represents the lane width corresponding to any point on the center curve; x 右 Indicates the x-axis coordinate value of the lane line on the right side corresponding to any point on the center curve.

5. The method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights according to claim 1, characterized in that: In step 1, the lane line 2D image is acquired by a camera installed in the vehicle headlight.

6. The method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights according to claim 5, characterized in that: The installation position and angle of the camera are fixed after installation and calibration.

7. The method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights according to claim 1, characterized in that: The camera is installed in the front high-beam headlight on the left front of the vehicle.

8. The method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights according to claim 1, characterized in that: The camera uses a wide-view camera that can capture four lane lines on the road ahead, where the four lane lines are two lane lines on the left side of the vehicle and two lane lines on the right side of the vehicle.

9. The method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights according to claim 1, characterized in that: In step 2, the selected key points are connected into polylines and then a lane line is evenly divided in the x direction.

10. The method for generating pixel-level segmentation labels for lane lines in intelligent vehicle lights according to claim 9, characterized in that: When the curvature of the lane line is larger, the number of key points is larger; when the curvature of the lane line is smaller, the number of key points is smaller.

Images