Lane line generation method, device, equipment and medium based on pitch angle estimation

Abstract

The present invention discloses a lane line generation method, apparatus, device, and medium based on pitch angle estimation, relating to the field of intelligent vehicle perception technology. The method comprises: detecting a captured road image to obtain a lane line image; performing perspective projection transformation on lane line coordinate points in the lane line image to obtain BEV lane line coordinate points under fixed extrinsic parameters; determining a first BEV lane line, acquiring sampling points on the first BEV lane line, and determining a first pitch angle correction angle based on the lane line parallel assumption; updating the first pitch angle correction angle in real time using a Kalman filter to obtain a second pitch angle correction angle, and obtaining a second pitch angle based on the second pitch angle correction angle and the first pitch angle in a fixed extrinsic parameter matrix; generating a second BEV lane line based on the second pitch angle; and the second BEV lane line being the target lane line. This solution improves lane line accuracy.

Description

Technical Field

[0001] The present invention relates to the field of intelligent vehicle perception technology, and in particular to a lane line generation method, device, equipment and medium based on pitch angle estimation. Background Art

[0002] Lane generation technology plays a key role in intelligent driving, providing crucial support for safe and efficient vehicle operation. Existing technologies typically determine lane lines in an image and then convert them into a bird's-eye view (BEV). However, real-world driving often involves severe road bumps, switching between straights and curves, lane changes, uphill and downhill driving, and other complex road conditions. Under these conditions, the pitch angle of the onboard camera changes in real time and is temporally correlated. Therefore, lane generation methods require estimation of the camera's pitch angle.

[0003] Existing methods for estimating the pitch angle of a vehicle-mounted monocular camera rely on obtaining the vanishing point in the image, calculating the camera pose based on the distance between the vanishing point and the optical axis, and then converting the image lane lines to the BEV coordinate system. This method is difficult to calculate the vanishing point and it is difficult to strike a balance between accuracy and efficiency. Alternatively, the image optical flow method is used to calculate the camera pose and then convert the image lane lines to the BEV coordinate system. However, this method has high computational complexity and is time-consuming. Alternatively, calibrated fixed camera extrinsics are directly used and then the image lane lines are converted to the BEV coordinate system. This method has poor accuracy. In addition, there is a method that uses perspective projection changes and prior information of parallel lane lines to construct a cost function with camera pitch angles and a grid region of interest under BEV. The unknown pitch angles are obtained by minimizing the cost function. This method obtains a local optimal solution, not a global optimal solution. The construction and solution process is complex and time-consuming. Uncertain errors will appear in the solution results during vehicle driving, and the pitch angle curve will appear jagged. The longer the time, the greater the cumulative error and the poorer the robustness. At the same time, in addition to directly using the calibrated fixed camera extrinsic parameter method, the above method is very easy to fail when the lane lines disappear and when missed detections or false detections occur due to poor quality of lane line detection in the image. Summary of the Invention

[0004] The present invention aims to at least solve the technical problems existing in the prior art. To this end, the present invention proposes a lane line generation method, device, equipment and medium based on pitch angle estimation.

[0005] A first aspect of the present invention provides a lane line generation method based on pitch angle estimation, the method comprising:

[0006] The lane line detection model is used to detect the road image collected by the monocular camera in real time to obtain the lane line image;

[0007] Performing perspective projection transformation on the lane line coordinate points in the lane line image according to the intrinsic parameter matrix, distortion coefficient, and fixed extrinsic parameter matrix of the monocular camera to obtain the BEV lane line coordinate points under fixed extrinsic parameters;

[0008] Determining a first BEV lane line based on the BEV lane line coordinate points, obtaining sampling points on the first BEV lane line, and determining a first pitch angle correction angle based on the lane line parallel assumption and the coordinates of the sampling points and the camera installation height; the sampling points are the starting points and end points of two non-parallel lane lines of the same length;

[0009] updating the first pitch angle correction angle in real time through a Kalman filter to obtain a second pitch angle correction angle, and obtaining a second pitch angle according to the second pitch angle correction angle and the first pitch angle in the fixed extrinsic parameter matrix;

[0010] A second BEV lane line is generated based on the second pitch angle; the second BEV lane line is the target lane line.

[0011] Optionally, acquiring a sampling point on the first BEV lane line includes:

[0012] Determining a first horizontal line at a fixed interval based on the BEV lane line coordinate point;

[0013] Determining a projection of the first transverse line in the lane line image based on the projection relationship, i.e., a second transverse line;

[0014] Determine an intersection of the second horizontal line and the lane line in the lane line image as a sampling point in the lane line image;

[0015] The BEV lane line coordinate point corresponding to the sampling point in the lane line image is obtained as the sampling point on the first BEV lane line.

[0016] Optionally, the first pitch angle correction angle is expressed as:

[0017]

[0018] in, is the first pitch angle correction angle, is the distance between the end points of two non-parallel lane lines, is the width of the starting point of the two non-parallel lane lines, h is the camera installation height, and L is the distance between the starting point and the end point of the two non-parallel lane lines.

[0019] Optionally, if there is only one lane line or no lane line in the first BEV lane line at the current moment, the pitch angle correction angle predicted by the Kalman filter is used as the second pitch angle correction angle at the current moment.

[0020] A second aspect of the present invention provides a lane line generation device based on pitch angle estimation, the device comprising:

[0021] An image acquisition module is used to detect the road image captured by the monocular camera in real time based on the lane line detection model to obtain a lane line image;

[0022] A first processing module is configured to perform perspective projection transformation on the lane line coordinate points in the lane line image according to the intrinsic parameter matrix, distortion coefficient, and fixed extrinsic parameter matrix of the monocular camera to obtain the BEV lane line coordinate points under the fixed extrinsic parameter;

[0023] a second processing module, configured to determine a first BEV lane line based on the BEV lane line coordinate points, obtain sampling points on the first BEV lane line, and determine a first pitch angle correction angle based on the lane line parallelism assumption and the coordinates of the sampling points and the camera installation height; the sampling points being the starting point and the end point of two non-parallel lane lines of the same length;

[0024] an updating module, configured to update the first pitch angle correction angle in real time through a Kalman filter to obtain a second pitch angle correction angle, and obtain the second pitch angle according to the second pitch angle correction angle and the first pitch angle in the fixed extrinsic parameter matrix;

[0025] A generating module is used to generate a second BEV lane line based on the second pitch angle; the second BEV lane line is a target lane line.

[0026] A third aspect of the present invention provides an electronic device, comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the lane line generation method as described in the first aspect.

[0027] A fourth aspect of the present invention provides a computer-readable storage medium, which stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement the lane line generation method as described in the first aspect.

[0028] A lane line generation method, device, equipment and medium based on pitch angle estimation have the following beneficial effects: this scheme combines fixed camera calibration extrinsic parameter matrix data and lane line parallel prior knowledge, calculates according to the BEV perspective projection relationship, and estimates the camera pitch angle in real time. Compared with the method based on image vanishing points, this scheme has small computational complexity and high efficiency, while improving the accuracy of lane lines, taking into account both accuracy and efficiency; compared with the image optical flow method which requires a large amount of calculation, this scheme only requires simple algebraic calculations, which improves efficiency; compared with the method using calibrated fixed camera extrinsic parameters, this scheme estimates the camera pitch angle in real time, improving the accuracy of lane lines; in addition, this scheme effectively avoids the complexity brought about by constructing a pitch angle cost function and optimizing the solution, reduces time consumption, and predicts and updates the camera pitch angle based on Kalman filtering, effectively avoiding the problem of estimation failure caused by long-term accumulated errors and poor working conditions, making the entire process more robust and applicable to more common and complex driving scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A flowchart of a lane line generation method based on pitch angle estimation provided by an embodiment of the present invention;

[0030] Figure 2 A schematic diagram of horizontal line sampling based on the camera's fixed extrinsic matrix mapping image;

[0031] Figure 3 Schematic diagram of perspective projection transformation for coordinate points;

[0032] Figure 4 The use of fixed calibration extrinsics results in the determined lane lines tilting inwards;

[0033] Figure 5 The lane lines determined by using fixed calibration external parameters are tilted outwards;

[0034] Figure 6 This is the actual situation map of the lane line;

[0035] Figure 7 A BEV lane line diagram in which one of the two inner lines of the self-lane in the first BEV lane line is missing;

[0036] Figure 8 A BEV lane line diagram with no other outer lines on the missing inner line side of the first BEV lane line;

[0037] Figure 9 A BEV lane line diagram having only one lane line in the first BEV lane line;

[0038] Figure 10 Schematic diagram for calculating the first pitch angle correction angle;

[0039] Figure 11 This is a curve diagram of the camera pitch angle under bumpy road conditions;

[0040] Figure 12 This is a test picture for a straight road scene;

[0041] Figure 13 Testing for cornering scenarios Figure 1 ;

[0042] Figure 14 Testing for cornering scenarios Figure 2 ;

[0043] Figure 15 This is a test image for the ramp scene;

[0044] Figure 16 Testing for lane-changing scenarios Figure 1 ;

[0045] Figure 17 Testing for lane-changing scenarios Figure 2 . DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0047] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, "multiple" means two or more. In addition, the use of "based on" or "according to" means openness and inclusiveness, because the process, steps, calculations or other actions "based on" or "according to" one or more of the conditions or values ​​may be based on additional conditions or values ​​beyond the stated in practice.

[0048] The embodiment of the present invention provides a lane line generation method based on pitch angle estimation, such as Figure 1 As shown, it includes the following steps:

[0049] Step 101: Detect a road image captured in real time by a monocular camera based on a lane line detection model to obtain a lane line image.

[0050] In this embodiment, a CNN-based segmentation model can be used to segment the lane line image mask, and then the lane line pixels are extracted. The pixels are fitted with a cubic curve using the least squares method to obtain the lane line image. Other methods can also be used to obtain the lane line image, and this embodiment does not limit this.

[0051] Step 102: Perform perspective projection transformation on the lane line coordinate points in the lane line image according to the intrinsic parameter matrix, distortion coefficient, and fixed extrinsic parameter matrix of the monocular camera to obtain BEV lane line coordinate points under fixed extrinsic parameters.

[0052] The intrinsic parameter matrix and distortion coefficients of a monocular camera can be determined using the Zhang Zhengyou calibration method. If these are already set when the camera leaves the factory, no further calibration is required. After the monocular camera is installed, extrinsic calibration is performed to obtain a fixed extrinsic parameter matrix. Specifically, extrinsic calibration is performed by measuring the three-dimensional coordinates of ground targets, finding their corresponding pixel coordinates, and solving the transformation matrix using the PnP method.

[0053] Sample each lane line in the lane line image to obtain the coordinate point.

[0054] The coordinate points are transformed into perspective projection according to the intrinsic parameter matrix, distortion coefficient, and fixed extrinsic parameter matrix to obtain all BEV lane line coordinate points, namely the BEV lane line coordinate point set.

[0055] Step 103: Determine a first BEV lane line based on the BEV lane line coordinate points, obtain sampling points on the first BEV lane line, and determine a first pitch angle correction angle based on the lane line parallel assumption and the coordinates of the sampling points and the camera installation height; the sampling points are the starting point and end point of two non-parallel lane lines of the same length.

[0056] In a possible implementation, acquiring a sampling point on the first BEV lane line includes:

[0057] Determining a first horizontal line at a fixed interval based on the BEV lane line coordinate point;

[0058] Determining a projection of the first transverse line in the lane line image based on the projection relationship, i.e., a second transverse line;

[0059] Determine an intersection of the second horizontal line and the lane line in the lane line image as a sampling point in the lane line image;

[0060] The BEV lane line coordinate point corresponding to the sampling point in the lane line image is obtained as the sampling point on the first BEV lane line.

[0061] To ensure accuracy and facilitate subsequent pitch angle calculation, the lane lines in the image are sampled every S meters in this step. The specific value of S is set according to actual needs and is not specifically limited in this embodiment. Specifically, the actual longitudinal distance range is fixed to 5-85 meters. Figure 2 , map the vertical coordinate to the image based on the camera's fixed external parameter matrix, and draw horizontal lines for sampling at fixed distances. That is, in the BEV coordinate, the longitudinal distance in front of the vehicle is between 5 meters and 85 meters, and a horizontal line is drawn every S meters. The value of S can be set to 1 meter. Then, according to the projection relationship, calculate the projection of these horizontal lines in the image, and get the following: Figure 2 The horizontal line in . Figure 2 The intersection of the horizontal line in the image and the lane line in the image is the sampling point in the image, that is, the coordinate point obtained by sampling each lane line in the lane line image mentioned above, and then the coordinate point in the lane line image is transformed into the BEV coordinate to obtain the BEV lane line coordinate point. Figure 3 , Figure 3 The left side is the lane line image, and the coordinate system is the uv coordinate system. Figure 3 The right side is the BEV image, the coordinate system is the xy coordinate system, from Figure 3 It can be seen that the coordinates in the lane line image are ( , ) The BEV point coordinates corresponding to the sampling point are ( , ).

[0062] Since the calibration extrinsic parameter matrix is ​​directly used in step 102, especially the fixed pitch angle, the determined lane lines appear to tilt inward or outward, while in reality the pitch angle changes continuously with the undulation of the road surface. Taking the actual parallel lane lines as an example, Figure 4 In the case where the lane line is tilted inwards due to the use of fixed calibration external parameters, Figure 5 This is the case where the determined lane line tilts outwards due to the use of fixed calibration external parameters. Figure 6 is the actual situation of the lane lines, that is, the lane lines are parallel.

[0063] Considering that the two lanes of the ego vehicle are parallel in most driving scenarios, this embodiment corrects the pitch angle in the fixed extrinsic parameter based on the prior knowledge of parallel lane lines.

[0064] Two non-parallel lane lines are screened out from the first BEV lane line, and a set of sampling points is obtained. The set of sampling points is obtained from the two screened non-parallel lane lines, and the two lane lines are actually parallel.

[0065] Assume that in the BEV coordinate system, the lateral distance relative to the lane centerline is The line within meters is the inner line, and the horizontal distance relative to the center line of the lane The line outside the meter is the outer line. For example, The value is set to 4.5 meters, The value of can also be set by the implementer according to the specific situation. In the first BEV lane line, the two inner lines of the own lane are given priority. If the image detection quality is poor, for example, only the inner line on the own vehicle side is detected, that is, one of the two inner lines of the own lane is missing, then it is necessary to filter the other line among all BEV lane lines. In the case that one of the two inner lines of the own lane is missing, Figure 7 As shown, the other lane lines on the missing side are preferred, that is, the outer line on the missing side. If there is no other outer line on the missing side, such as Figure 8 , and then select the outer line closest to the inner line on the other side for calculation. In addition, there is a situation where there is only one lane line in the figure, such as Figure 9 shown.

[0066] like Figure 10 As shown, based on The two thick lines in the coordinate system represent the actual parallel lane lines, while the two thin lines in the XZ coordinate system represent the non-parallel lane lines caused by the change in the camera's pitch angle. Based on the properties of similar triangles, the tangent function value is used as the pitch correction angle, which is the first pitch correction angle.

[0067] In a possible implementation, the first pitch angle correction angle is expressed as:

[0068]

[0069] in, is the first pitch angle correction angle, is the distance between the end points of two non-parallel lane lines, is the starting point width of the two non-parallel lane lines, which is also the distance between the two actually parallel lane lines. h is the camera installation height. L is the distance between the starting point and the end point of the two non-parallel lane lines.

[0070] By compensating the pitch angle in the fixed extrinsic parameter, a corrected pitch angle can be obtained, which is the sum of the first pitch angle correction angle and the pitch angle in the fixed extrinsic parameter.

[0071] It should be noted that in most scenarios, namely straight-line scenarios, the above pitch correction angle calculation method only requires a set of sampling coordinate points, namely four coordinate points. Using these four coordinates, combined with the above formula, the first pitch correction angle can be determined. However, in curved and sloped scenarios, the calculation results using only four coordinate points are not stable. Furthermore, considering that the detection quality of the self-driving lane within 30 meters ahead is relatively higher, it can be assumed that a strong parallelism prior exists. Therefore, for curved and sloped scenarios, the calculation is performed by taking four sampling points every S meters within a certain distance in front of the vehicle. Here, S is the same as the value of S in step 102. Each group of four sampling points is formed, and the pitch correction angle determined by each group of sampling points is averaged to obtain the final first pitch correction angle. It should be noted that here, it is assumed that there is a strong parallelism prior for the self-driving lane within 30 meters ahead. The value of 30 meters is determined based on experience and actual measurements. For example, assuming that the vehicle is on a straight road, four points need to be selected in front of the vehicle. Two points can be selected 5 meters away from the two lane lines of the own lane, and another two points can be selected 5+S meters away from the two lane lines of the own lane. A total of four sampling points form a group, and the first pitch angle correction angle is determined based on these four sampling points.

[0072] Among them, the curve scene can be achieved through the central lane cubic curve Parameters It is determined that this parameter is related to the curvature of the cubic curve of the center lane. Meters and tail is the front of the vehicle The pitch correction angles calculated at the two locations are compared, and a threshold is set for the pitch correction angles calculated at the two locations. If the difference is greater than the set threshold, it is considered that there is a slope.

[0073] For complex scenarios such as curves and slopes, the length of parallel lines is strategically shortened, and special measurement error processing is performed. That is, the average value of multiple groups of sampling points is taken to make the parallel line hypothesis more consistent with actual application scenarios.

[0074] Step 104: Update the first pitch angle correction angle in real time through a Kalman filter to obtain a second pitch angle correction angle, and obtain a second pitch angle according to the second pitch angle correction angle and the first pitch angle in the fixed extrinsic parameter matrix.

[0075] In order to make the whole process more robust, reduce the long-term cumulative error, and avoid failure under poor working conditions, the first pitch angle correction angle Input the Kalman filter and perform prediction updates on it.

[0076] The established Kalman filter is as follows:

[0077] The state variables include: ,in, yes The derivative of represents the camera pitch compensation angular velocity, and the state transfer matrix F is:

[0078]

[0079] in, Indicates a time interval.

[0080] The observation quantity Z is , the observation error is set according to the quality of the two lane lines screened in step 103. The worse the quality of the lane line, the greater the observation error.

[0081] The second pitch angle correction angle is obtained after real-time updating through Kalman filtering, and is added to the first pitch angle in the fixed external parameter matrix to obtain the second pitch angle.

[0082] In a possible implementation, if at the current moment there is only one lane line or no lane line in the first BEV lane line, the pitch angle correction angle predicted by the Kalman filter is used as the second pitch angle correction angle at the current moment.

[0083] In the embodiment of the present invention, in response to the disappearance of lane lines, the temporal correlation of the camera pitch angle is used to perform prediction, thereby increasing robustness and avoiding functional failure.

[0084] Step 105: Generate a second BEV lane line based on the second pitch angle; the second BEV lane line is the target lane line.

[0085] According to the setting of the inner and outer lines in the BEV coordinate system in step 103, the two BEV lane lines can be marked for lane change judgment during the driving process of the vehicle, thereby enhancing the BEV perception capability of the smart car and providing effective protection for smart driving related functions.

[0086] Reference Figure 11 In this embodiment of the present invention, a bumpy road scene is selected for testing. Three lines are drawn based on the fixed camera pitch angle calibration value, the first pitch angle correction angle calculated in step 103, the pitch angle determined by the first pitch angle in the fixed extrinsic parameter matrix, and the second pitch angle determined by the Kalman filter. Figure 11In the figure, the horizontal axis is the image frame number (Frame ID), and the vertical axis is the pitch angle (pitch), and the unit is Degrees. Among the three curves in the figure, calibration is the index fixed value, that is, the fixed camera pitch angle calibration value, that is, the angle used by the fixed external parameter method; calculation is the calculated value, which is the pitch angle obtained by compensating the calculated value each time, that is, the first pitch angle correction angle is calculated according to the method in this embodiment, but without the effect of Kalman filtering, that is, the pitch angle determined according to the first pitch angle correction angle; kalman filter is the result after Kalman filtering, that is, the final result, the second pitch angle, obtained by the method of the embodiment of the present invention. From Figure 11 It can be seen that the final result curve obtained by the Kalman filter is in line with expectations and is relatively stable, with the overall fluctuation range remaining within 0.5 degrees.

[0087] The embodiment of the present invention also selects typical driving scenes for testing, including straight roads, curves, slopes, and lane changes. Figure 12-17 As shown, Figure 12 For straight road scenes, Figure 13-14 For the bend scene, Figure 15 For the ramp scene, Figure 16-17 For lane-changing scenarios, Figure 12-17 The left side of the center shows a pixel image, with the dark gray line representing the lane detection result. The right side shows a BEV image, with the white line representing the generated lane line result. The lane lines are also numbered, with the two inner lane lines of the ego lane numbered 1 and 2, and the lane lines numbered 0 and 3 as the two outer lane lines of the inner lane. This solution can robustly generate lane lines in different scenarios, meeting the perception requirements of complex driving scenarios.

[0088] In summary, in the embodiment of the present invention, this solution combines fixed camera calibration extrinsic parameter matrix data and lane line parallel prior knowledge, calculates according to the BEV perspective projection relationship, and estimates the camera pitch angle in real time. Compared with the method based on the image vanishing point, this solution has small computational complexity and high efficiency, while improving the accuracy of the lane line, taking into account both accuracy and efficiency; compared with the image optical flow method which requires a large amount of calculation, this solution only requires simple algebraic calculations, which improves efficiency; compared with the method using calibrated fixed camera extrinsics, this solution estimates the camera pitch angle in real time, improving the accuracy of the lane line; in addition, this solution effectively avoids the complexity brought about by constructing the pitch angle cost function and optimizing the solution, reduces time consumption, and predicts and updates the camera pitch angle based on the Kalman filter, effectively avoiding the problem of estimation failure caused by long-term accumulated errors and poor working conditions, making the entire process more robust and applicable to more common and complex driving scenarios.

[0089] An embodiment of the present invention further provides a lane line generation device based on pitch angle estimation, the device comprising:

[0090] An image acquisition module is used to detect the road image captured by the monocular camera in real time based on the lane line detection model to obtain a lane line image;

[0091] A first processing module is configured to perform perspective projection transformation on the lane line coordinate points in the lane line image according to the intrinsic parameter matrix, distortion coefficient, and fixed extrinsic parameter matrix of the monocular camera to obtain the BEV lane line coordinate points under the fixed extrinsic parameter;

[0092] a second processing module, configured to determine a first BEV lane line based on the BEV lane line coordinate points, obtain sampling points on the first BEV lane line, and determine a first pitch angle correction angle based on the lane line parallelism assumption and the coordinates of the sampling points and the camera installation height; the sampling points being the starting point and the end point of two non-parallel lane lines of the same length;

[0093] an updating module, configured to update the first pitch angle correction angle in real time through a Kalman filter to obtain a second pitch angle correction angle, and obtain the second pitch angle according to the second pitch angle correction angle and the first pitch angle in the fixed extrinsic parameter matrix;

[0094] A generating module is used to generate a second BEV lane line based on the second pitch angle; the second BEV lane line is a target lane line.

[0095] In another embodiment provided by the present invention, an electronic device is also provided, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the lane line generation method in the embodiment of the present invention.

[0096] In another embodiment provided by the present invention, a computer-readable storage medium is also provided, which stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement the lane line generation method in the embodiment of the present invention.

[0097] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention shall be covered by the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A lane line generation method based on pitch angle estimation, characterized in that: include: The lane line detection model is used to detect the road image collected by the monocular camera in real time to obtain the lane line image; Performing perspective projection transformation on the lane line coordinate points in the lane line image according to the intrinsic parameter matrix, distortion coefficient, and fixed extrinsic parameter matrix of the monocular camera to obtain the BEV lane line coordinate points under fixed extrinsic parameters; Determine a first BEV lane line based on the BEV lane line coordinate points, obtain a sampling point on the first BEV lane line, and determine a first pitch angle correction angle based on the lane line parallel assumption and the coordinates of the sampling point and the camera installation height; The sampling points are the starting point and the end point of two non-parallel lane lines of the same length; Acquiring a sampling point on the first BEV lane line includes: Determining a first horizontal line at a fixed interval based on the BEV lane line coordinate point; Determining a projection of the first transverse line in the lane line image based on the projection relationship, i.e., a second transverse line; Determine an intersection of the second horizontal line and the lane line in the lane line image as a sampling point in the lane line image; Obtaining a BEV lane line coordinate point corresponding to a sampling point in the lane line image as the sampling point on the first BEV lane line; The expression of the first pitch angle correction angle is: in, is the first pitch angle correction angle, is the distance between the end points of two non-parallel lane lines, is the width of the starting point of the two non-parallel lane lines, h is the camera installation height, and L is the distance between the starting point and the end point of the two non-parallel lane lines; The first pitch angle correction angle is updated in real time using a Kalman filter to obtain a second pitch angle correction angle, and the second pitch angle is obtained based on the second pitch angle correction angle and the first pitch angle in the fixed extrinsic parameter matrix; if there is only one lane line or no lane line in the first BEV lane line at the current moment, the pitch angle correction angle predicted by the Kalman filter is used as the second pitch angle correction angle at the current moment; A second BEV lane line is generated based on the second pitch angle; the second BEV lane line is the target lane line.

2. A lane line generation device based on pitch angle estimation, characterized in that: The device comprises: An image acquisition module is used to detect the road image captured by the monocular camera in real time based on the lane line detection model to obtain a lane line image; A first processing module is configured to perform perspective projection transformation on the lane line coordinate points in the lane line image according to the intrinsic parameter matrix, distortion coefficient, and fixed extrinsic parameter matrix of the monocular camera to obtain the BEV lane line coordinate points under the fixed extrinsic parameter; The second processing module is configured to determine a first BEV lane line based on the BEV lane line coordinate points, obtain sampling points on the first BEV lane line, and determine a first pitch angle correction angle based on the lane line parallel assumption and the coordinates of the sampling points and the camera installation height; the sampling points are the starting points and end points of two non-parallel lane lines of the same length; obtaining the sampling points on the first BEV lane line includes: Determining a first horizontal line at a fixed interval based on the BEV lane line coordinate point; Determining a projection of the first transverse line in the lane line image based on the projection relationship, i.e., a second transverse line; Determine an intersection of the second horizontal line and the lane line in the lane line image as a sampling point in the lane line image; Obtaining a BEV lane line coordinate point corresponding to a sampling point in the lane line image as the sampling point on the first BEV lane line; The expression of the first pitch angle correction angle is: in, is the first pitch angle correction angle, is the distance between the end points of two non-parallel lane lines, is the width of the starting point of the two non-parallel lane lines, h is the camera installation height, and L is the distance between the starting point and the end point of the two non-parallel lane lines; an updating module, configured to update the first pitch angle correction angle in real time using a Kalman filter to obtain a second pitch angle correction angle, and obtain the second pitch angle based on the second pitch angle correction angle and the first pitch angle in the fixed extrinsic parameter matrix; and if there is only one lane line or no lane line in the first BEV lane line at a current moment, use the pitch angle correction angle predicted by the Kalman filter as the second pitch angle correction angle at a current moment; A generating module is used to generate a second BEV lane line based on the second pitch angle; the second BEV lane line is a target lane line.

3. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the lane line generation method as described in claim 1.

4. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the lane line generation method as claimed in claim 1.

Images