Consistency evaluation method and device of 3D lane line and 2D lane line and vehicle

Abstract

This application relates to a method, apparatus, and vehicle for consistency assessment of 3D lane lines and 2D lane lines. The method includes: transforming 3D lane lines into 3D projected 2D lane lines based on vehicle pose, camera extrinsic parameters, and camera intrinsic parameters; dilating the 2D lane lines and the 3D projected 2D lane lines; calculating the intersection-union ratio (IUGR) of the dilated 2D lane lines and the 3D projected 2D lane lines; determining the associated line pairs of the 2D lane lines and the 3D projected 2D lane lines based on the IUGR; obtaining the assessment result for each associated line pair according to a preset assessment strategy; and statistically analyzing the assessment results of all associated line pairs to obtain the consistency assessment result of the 3D lane lines and 2D lane lines. This solves the problems of inconsistencies between the true values ​​of 3D and 2D lane lines and limitations in the assessment methods for the true values ​​of 3D and 2D lane lines, leading to inaccurate and incomplete assessment results, thus improving the efficiency of the assessment process and the reliability of the assessment results.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and in particular to a method, device and vehicle for evaluating the consistency of 3D lane lines and 2D lane lines. Background Technology

[0002] With the rapid development of automotive intelligence technologies, the adoption rate of intelligent assisted driving (AWD) is increasing. One of the essential conditions for realizing AWD is the recognition of lane lines around the vehicle, especially BEV (Bird's-Eye View) lane line recognition based on deep learning and panoramic cameras. A crucial type of data required for deep learning-based BEV lane line recognition is ground truth 3D lane line data that matches the 2D (Two-Dimensional) lane lines in the image. This data directly affects the accuracy of the BEV lane lines generated by the on-board deep learning model, which in turn directly impacts downstream localization mapping, path planning, and vehicle control, making it vital for the normal execution of AWD functions and driving safety.

[0003] Lane line ground truth data is categorized into 2D and 3D types. Lane line data directly labeled on an image is typically referred to as 2D lane line ground truth, while lane line data labeled in a point cloud is typically referred to as 3D lane line ground truth. These two types of ground truth lane lines are usually labeled using data collected from different sensors, and since the images are projected, there are inherent differences in pose and scale between them. Given fixed sensor intrinsic and extrinsic parameters, quantifying and evaluating the consistency between the two types of ground truth lane line data is crucial.

[0004] In related technologies, the consistency assessment between 3D and 2D lane line ground truth values ​​is mostly based on qualitative evaluation using visualization, or a simplified evaluation based on a single indicator. These technologies do not provide applicable indicators for different types of lane lines, nor can they automatically provide a comprehensive evaluation result for the consistency of all 2D and 3D lane lines within a data package. Therefore, they cannot provide timely and efficient quantitative indicators for the consistency of batches of 2D and 3D lane lines. Summary of the Invention

[0005] This application provides a method, apparatus, and vehicle for evaluating the consistency between 3D lane lines and 2D lane lines, in order to solve the problems of inconsistency between the true value data of 3D and 2D lane lines and the limitations of the evaluation methods for the true values ​​of 3D and 2D lane lines, which lead to inaccurate and incomplete evaluation results. The method automatically and comprehensively evaluates the consistency between 3D and 2D lane lines, thereby improving the efficiency of the evaluation process and the reliability of the evaluation results.

[0006] The first aspect of this application provides a method for evaluating the consistency between 3D lane lines and 2D lane lines, including the following steps:

[0007] The system acquires the 2D lane lines labeled in the image, the first 3D lane lines in the point cloud map coordinate system, the pose of the current vehicle in the point cloud map coordinate system, the camera extrinsic parameters, and the camera intrinsic parameters, and transforms the first 3D lane lines into 3D projected 2D lane lines based on the pose, the camera extrinsic parameters, and the camera intrinsic parameters.

[0008] The 2D lane lines labeled in the image and the 2D lane lines projected in the 3D projection are dilated respectively. The intersection-union ratio of each dilated 2D lane line labeled in the image and each dilated 2D lane line projected in the 3D projection is calculated. Multiple pairs of associated lines between the 2D lane lines labeled in the image and the 2D lane lines projected in the 3D projection are determined based on the intersection-union ratio.

[0009] Based on each pair of associated lines and the type of the corresponding pair, the evaluation result of each pair of associated lines is obtained according to the preset evaluation strategy. The evaluation results of all pairs of associated lines are statistically analyzed to obtain the consistency evaluation result of 3D lane lines and 2D lane lines.

[0010] Optionally, in some embodiments, obtaining the evaluation result of each pair of associated lines based on each pair of lines and the type of the corresponding pair of lines according to a preset evaluation strategy includes:

[0011] When the current associated line pair is a continuous lane line type, determine multiple data points on the current 3D projected 2D lane line, and determine the specified line segment of each data point on the 2D lane line labeled in the current image;

[0012] Calculate the vertical distance between each data point and the corresponding specified line segment, and use the average of all vertical distances as the evaluation result of the current associated line pair.

[0013] Optionally, in some embodiments, obtaining the evaluation result of each pair of associated lines based on each pair of lines and the type of the corresponding pair of lines according to a preset evaluation strategy includes:

[0014] When the current associated line pair is a dashed lane line type, the first starting point and the first ending point on the current 3D projected 2D lane line, and the second starting point and the second ending point on the current image-annotated 2D lane line are determined respectively;

[0015] The first Euclidean distance is obtained based on the first starting point and the second starting point, and the second Euclidean distance is obtained based on the first ending point and the second ending point;

[0016] The average of the first Euclidean distance and the second Euclidean distance is used as the evaluation result of the current associated line pair.

[0017] Optionally, in some embodiments, transforming the first 3D lane line into a 3D projected 2D lane line based on the pose, the camera extrinsic parameters, and the camera intrinsic parameters includes:

[0018] The first 3D lane line is transformed into a second 3D lane line in the vehicle coordinate system according to the pose, and the second 3D lane line is transformed into a third 3D lane line in the camera coordinate system according to the camera extrinsic parameters.

[0019] Based on the camera intrinsic parameters, the third 3D lane line is projected onto the image pixel coordinate system according to a preset cropping and processing strategy to obtain the 3D projected 2D lane line.

[0020] Optionally, in some embodiments, the step of projecting the third 3D lane line onto the image pixel coordinate system based on the camera intrinsic parameters and a preset cropping and processing strategy further includes:

[0021] Based on a preset filtering strategy, multiple 3D points are selected on the third 3D lane line.

[0022] Linear interpolation is performed on the plurality of 3D points, and based on the camera intrinsic parameters, each linearly interpolated 3D point is projected onto the image pixel coordinate system to obtain multiple projection points and a projection image;

[0023] Determine whether each projection point meets the preset filtering conditions;

[0024] If the current projection point meets the preset filtering conditions, then the current projection point is deleted, and the lane line formed by the remaining projection points is used as the 3D projection 2D lane line.

[0025] A second aspect of this application provides a device for evaluating the consistency between 3D lane lines and 2D lane lines, comprising:

[0026] The transformation module is used to acquire the 2D lane line labeled in the image, the first 3D lane line in the point cloud map coordinate system, the pose of the current vehicle in the point cloud map coordinate system, the camera extrinsic parameters, and the camera intrinsic parameters, and to transform the first 3D lane line into a 3D projected 2D lane line according to the pose, the camera extrinsic parameters, and the camera intrinsic parameters.

[0027] The association module is used to perform dilation operations on the 2D lane lines labeled in the image and the 2D lane lines projected in the 3D image respectively, calculate the intersection-union ratio of each dilated 2D lane line labeled in the image and each dilated 2D lane line projected in the 3D image, and determine multiple sets of association line pairs between the 2D lane lines labeled in the image and the 2D lane lines projected in the 3D image based on the intersection-union ratio.

[0028] The evaluation module is used to obtain the evaluation result of each set of associated line pairs based on each set of associated line pairs and the type of the corresponding associated line pairs, according to a preset evaluation strategy, and to statistically analyze the evaluation results of all sets of associated line pairs to obtain the consistency evaluation result of 3D lane lines and 2D lane lines.

[0029] Optionally, in some embodiments, the evaluation module is specifically used for:

[0030] When the current associated line pair is a continuous lane line type, determine multiple data points on the current 3D projected 2D lane line, and determine the specified line segment of each data point on the 2D lane line labeled in the current image;

[0031] Calculate the vertical distance between each data point and the corresponding specified line segment, and use the average of all vertical distances as the evaluation result of the current associated line pair.

[0032] Optionally, in some embodiments, the evaluation module is specifically used for:

[0033] When the current associated line pair is a dashed lane line type, the first starting point and the first ending point on the current 3D projected 2D lane line, and the second starting point and the second ending point on the current image-annotated 2D lane line are determined respectively;

[0034] The first Euclidean distance is obtained based on the first starting point and the second starting point, and the second Euclidean distance is obtained based on the first ending point and the second ending point;

[0035] The average of the first Euclidean distance and the second Euclidean distance is used as the evaluation result of the current associated line pair.

[0036] Optionally, in some embodiments, the transformation module is specifically used for:

[0037] The first 3D lane line is transformed into a second 3D lane line in the vehicle coordinate system according to the pose, and the second 3D lane line is transformed into a third 3D lane line in the camera coordinate system according to the camera extrinsic parameters.

[0038] Based on the camera intrinsic parameters, the third 3D lane line is projected onto the image pixel coordinate system according to a preset cropping and processing strategy to obtain the 3D projected 2D lane line.

[0039] Optionally, in some embodiments, the transformation module is further configured to:

[0040] Based on a preset filtering strategy, multiple 3D points are selected on the third 3D lane line.

[0041] Linear interpolation is performed on the plurality of 3D points, and based on the camera intrinsic parameters, each linearly interpolated 3D point is projected onto the image pixel coordinate system to obtain multiple projection points and a projection image;

[0042] Determine whether each projection point meets the preset filtering conditions;

[0043] If the current projection point meets the preset filtering conditions, delete the current projection point and use the lane line formed by the remaining projection points as the 3D projection 2D lane line.

[0044] A third aspect of this application provides a vehicle comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the consistency evaluation method for 3D lane lines and 2D lane lines as described in the above embodiments.

[0045] A fourth aspect of this application provides a computer program product having a computer program stored thereon that is executed to implement the consistency evaluation method for 3D lane lines and 2D lane lines as described in the above embodiments.

[0046] Therefore, this application transforms the 3D lane lines in the point cloud map coordinate system to the vehicle coordinate system based on the vehicle's pose in the point cloud map. Based on the extrinsic parameters from the vehicle coordinate system to the camera coordinate system, it transforms the 3D lane lines in the vehicle coordinate system to the camera coordinate system. Based on the camera intrinsic parameters and the vehicle mask, it projects the 3D lane lines in the camera coordinate system to the image pixel coordinate system. This process requires interpolation of discrete 3D lane line points to handle view frustum clipping and target occlusion. Based on the cross-union ratio (CUNR) cross-matching index between line pairs, it automatically obtains the correspondence between the 2D lane lines projected from the 3D lane lines and the 2D lane lines labeled in the image. It evaluates the consistency between associated solid lines based on the point-line distance and the consistency between associated dashed lines based on the average endpoint distance. It processes multiple frames of data within a data packet frame by frame and statistically analyzes the results of the multi-frame processing to obtain a quantitative evaluation index of the consistency of all 2D and 3D lane lines within a data packet. This solves the problems of inconsistency between 3D and 2D lane line ground truth data and limitations in the evaluation methods for 3D and 2D lane line ground truth, which lead to inaccurate and incomplete evaluation results. It automatically and comprehensively evaluates the consistency of 3D and 2D lane lines, improving the efficiency of the evaluation process and the reliability of the evaluation results.

[0047] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0048] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0049] Figure 1 This is a flowchart of a consistency evaluation method for 3D lane lines and 2D lane lines provided according to an embodiment of this application;

[0050] Figure 2 This is a schematic diagram illustrating the processing of view frustum clipping and vehicle occlusion problems according to a specific embodiment of this application;

[0051] Figure 3 This is a schematic diagram illustrating the consistency assessment of dashed lane lines and continuous lane lines according to a specific embodiment of this application;

[0052] Figure 4 This is a flowchart of a consistency evaluation method for 3D lane lines and 2D lane lines according to a specific embodiment of this application;

[0053] Figure 5 This is a block diagram of a consistency evaluation device for 3D lane lines and 2D lane lines provided according to an embodiment of this application;

[0054] Figure 6 This is a block diagram of a vehicle provided according to an embodiment of this application. Detailed Implementation

[0055] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0056] The following description, with reference to the accompanying drawings, outlines a method, apparatus, and vehicle for evaluating the consistency of 3D and 2D lane lines according to embodiments of this application. Addressing the issues mentioned in the background art, such as the inconsistency between the ground truth data of 3D and 2D lane lines and the limitations of the evaluation methods for the ground truth of 3D and 2D lane lines, which lead to inaccurate and incomplete evaluation results, this application provides a method for evaluating the consistency of 3D and 2D lane lines. In this method, the following are obtained: 2D lane lines annotated in an image; a first 3D lane line in a point cloud map coordinate system; the pose of the current vehicle in the point cloud map coordinate system; camera extrinsic parameters; and camera intrinsic parameters. The first 3D lane line is then transformed into a 3D projection based on the pose, camera extrinsic parameters, and camera intrinsic parameters. The method involves mapping 2D lane lines; dilating both the image-annotated 2D lane lines and the 3D projected 2D lane lines; calculating the intersection-union ratio (IUR) of each dilated image-annotated 2D lane line and each dilated 3D projected 2D lane line; and determining multiple pairs of associated line pairs based on the IUR. Based on each pair and its type, an evaluation result is obtained for each pair according to a preset evaluation strategy. The evaluation results of all pairs are then statistically analyzed to obtain a consistency evaluation result for the 3D and 2D lane lines. This method solves the problems of inconsistencies between the ground truth data of 3D and 2D lane lines and limitations in the evaluation methods for 3D and 2D lane lines, leading to inaccurate and incomplete evaluation results. It automatically and comprehensively evaluates the consistency of 3D and 2D lane lines, improving the efficiency of the evaluation process and the reliability of the evaluation results.

[0057] Specifically, Figure 1 This is a flowchart illustrating a method for evaluating the consistency of 3D lane lines and 2D lane lines provided in an embodiment of this application.

[0058] like Figure 1 As shown, the consistency assessment method for 3D lane lines and 2D lane lines includes the following steps:

[0059] In step S101, the 2D lane line labeled in the image, the first 3D lane line in the point cloud map coordinate system, the pose of the current vehicle in the point cloud map coordinate system, the camera extrinsic parameters, and the camera intrinsic parameters are obtained, and the first 3D lane line is transformed into a 3D projected 2D lane line based on the pose, camera extrinsic parameters, and camera intrinsic parameters.

[0060] In this application embodiment, the 2D lane lines in the image annotation refer to the lane lines annotated in the panoramic camera image. Specifically, they are represented as a series of 2D points in the image pixel coordinate system, grouped according to their respective lines, and accompanied by attributes such as line type, color, and occlusion degree. The point cloud map in this application embodiment refers to the coordinate-point representation of a 3D scene constructed based on LiDAR (Light Detection and Ranging) multi-sensor fusion SLAM (Simultaneous Localization and Mapping) technology. The 3D lane lines are the lane line annotations on the constructed point cloud map. The first 3D lane line in this application embodiment refers to the representation of actual lane lines in the 3D point cloud map coordinate system. Specifically, it is represented as a series of 3D points, grouped according to their respective lines, and accompanied by line type attributes.

[0061] It is understood that, in order to obtain the consistency evaluation results of 2D lane lines and 3D lane lines, this embodiment of the application first needs to transform the 3D lane lines in the obtained point cloud map coordinate system into 3D projected 2D lane lines, and then associate them with the 2D lane lines annotated in the image, thereby quantifying the consistency between the two. Therefore, this embodiment of the application performs a series of transformations on the 3D lane lines in the point cloud map coordinate system based on the vehicle's pose data, camera extrinsic parameters, and camera intrinsic parameters to obtain the 2D lane lines projected from the 3D lane lines, which are the 3D projected 2D lane lines of this embodiment of the application.

[0062] Specifically, in this application embodiment, the vehicle's pose in the point cloud map coordinate system refers to the relative position and orientation of the camera calibration in the vehicle coordinate system, specifically represented as a combination of position coordinates and orientation quaternions or a pose matrix. The camera extrinsic parameters in this application embodiment are the extrinsic parameters from the vehicle coordinate system to the camera coordinate system, i.e., the relative position and orientation of the camera calibration in the vehicle coordinate system, specifically represented as a combination of position coordinates and orientation quaternions or a pose matrix. The camera intrinsic parameters in this application embodiment refer to the projection parameters and distortion parameters from the camera coordinate system to the image coordinate system at the time of camera imaging.

[0063] Optionally, in some embodiments, transforming the first 3D lane line into a 3D projected 2D lane line based on the pose, camera extrinsic parameters, and camera intrinsic parameters includes: transforming the first 3D lane line into a second 3D lane line in the vehicle coordinate system based on the pose, and transforming the second 3D lane line into a third 3D lane line in the camera coordinate system based on the camera extrinsic parameters; and projecting the third 3D lane line onto the image pixel coordinate system based on the camera intrinsic parameters and a preset cropping and processing strategy to obtain a 3D projected 2D lane line.

[0064] First, based on the vehicle's pose in the point cloud map, this embodiment transforms the 3D lane line (first 3D lane line) in the point cloud map coordinate system to the vehicle coordinate system. The 3D lane line is represented as a series of 3D position points in both the point cloud map coordinate system and the vehicle coordinate system. The specific transformation process is as follows:

[0065]

[0066] in, Let P be the inverse of the vehicle's pose matrix in the point cloud map coordinate system. _map_3Dline P represents the coordinates of 3D lane line data points in the point cloud map coordinate system. _veh_3Dline This represents the coordinates of 3D lane line data points in the vehicle coordinate system.

[0067] Furthermore, this embodiment transforms the 3D lane lines from the vehicle coordinate system to the camera coordinate system based on the camera extrinsic parameters from the vehicle coordinate system to the camera coordinate system. The 3D lane lines are represented as a series of 3D position points in both the vehicle and camera coordinate systems. The specific transformation process is as follows:

[0068]

[0069] in, Let P be the inverse of the camera's extrinsic parameter matrix in the vehicle coordinate system. _veh_3Dline P represents the coordinates of 3D lane line data points in the vehicle coordinate system. _cam_3Dline This represents the coordinates of 3D lane line data points in the camera coordinate system.

[0070] Optionally, in some embodiments, projecting the third 3D lane line onto the image pixel coordinate system based on camera intrinsic parameters and a preset cropping and processing strategy further includes: selecting multiple 3D points on the third 3D lane line based on a preset filtering strategy; performing linear interpolation on the multiple 3D points, and projecting each linearly interpolated 3D point onto the image pixel coordinate system based on camera intrinsic parameters to obtain multiple projection points and a projection image; determining whether each projection point meets preset filtering conditions; if the current projection point meets the preset filtering conditions, deleting the current projection point, and using the lane line formed by the remaining projection points as the 3D projected 2D lane line.

[0071] Understandably, after obtaining the 3D lane lines in the camera coordinate system, this embodiment projects the 3D lane lines from the camera coordinate system to the camera image pixel coordinate system based on camera intrinsic parameters and the vehicle mask, thus obtaining the 3D projected 2D lane lines of this embodiment. The 3D lane lines are represented as a series of 3D position points in the camera coordinate system and as a series of 2D pixels in the camera image pixel coordinate system. Furthermore, since the 3D lane lines are represented as a series of 3D point coordinates, and the dashed lines represent the start and end point coordinates, the projection process from the 3D lane line data points in the camera coordinate system to the imaging pixel plane requires handling the issues of view frustum clipping and vehicle body occlusion.

[0072] The vehicle mask in this embodiment refers to the portion obscured by the vehicle body or camera lens during wide-angle camera imaging, specifically represented as a polygonal region composed of a series of pixels in a 2D image. The view frustum clipping and target occlusion problems in this embodiment refer to the image edge distortion and occlusion caused by the vehicle body or camera lens, which prevent the 2D lane lines obtained after transformation and projection from the 3D lane lines from correctly corresponding in length to the 2D lane lines directly marked on the image. Therefore, to solve these problems, this embodiment performs linear interpolation on the 3D points of this portion of the lane lines before projection to avoid view frustum clipping and target occlusion problems while maintaining the overall shape of the lane lines.

[0073] Specifically, in this embodiment, based on a preset filtering strategy, multiple 3D points are selected on the third 3D lane line. The preset filtering strategy involves selecting 3D points on the lane line closer to the camera, performing linear interpolation on the multiple 3D points, and then using camera intrinsic parameters to project each linearly interpolated 3D point onto the image pixel coordinate system. The specific projection process is as follows:

[0074]

[0075] r 2 =x′ 2 +y′ 2 ;

[0076]

[0077] Among them, X c Y c Z c Let x be the three coordinates of the 3D lane line data points in the camera coordinate system. ′ y ′ Let x″ and y″ be the two coordinate values ​​of the 3D lane line data point in the camera coordinate system on the unit plane of the Z-axis in the camera coordinate system, respectively, and f be the two coordinate values ​​of the 3D lane line data point in the camera coordinate system after distortion removal on the unit plane of the Z-axis in the camera coordinate system. x fy c x c y Here, k1, k2, and k3 are the camera intrinsic parameters, p1 and p2 are the camera tangential distortion parameters, and u and v are the pixel coordinates of the 3D lane line data points projected onto the pixel coordinate system.

[0078] Therefore, this embodiment of the application obtains multiple projection points and projection images. Since the camera's field of view is limited, only 3D points located within the camera's frustum can be projected onto the image. Therefore, this embodiment of the application needs to filter out points that, after projection, lie outside the image boundary to ensure the validity of the projection results.

[0079] Furthermore, to address the issue of vehicle body occlusion, this embodiment of the application needs to identify potentially occluded lane line portions. Optionally, this embodiment can achieve this by checking the distance from the projection point to the image edge or the edge of the vehicle mask (i.e., the representation of the vehicle outline in the image). For each potentially occluded area, this embodiment takes the two nearest projection points to form a ray for subsequent cropping processing. In other words, this embodiment takes the two 3D lane line projection points on the projected image that are closest to the image edge or the edge of the vehicle mask to form a ray.

[0080] Furthermore, based on the rays obtained in the previous step, this embodiment calculates their intersection points with the image edge or the edge of the vehicle mask. Since these points represent occluded or invisible lane line portions, this embodiment needs to filter out the points between these intersection points and the image edge.

[0081] Therefore, the preset filtering conditions in this application embodiment are: 3D points outside the camera's view cone space, and the intersection of the ray formed by the two 3D lane line projection points closest to the image edge or the edge of the vehicle mask with the image edge or the edge of the vehicle mask.

[0082] Furthermore, in this embodiment, the remaining projection points are processed according to the lane line type (continuous line or dashed line). Specifically, if the current lane line is a continuous lane line represented by a series of continuous points, then the continuous lane line formed by the remaining points is the lane line that correctly corresponds to the 2D continuous lane line labeled in the image; if the current lane line is a dashed lane line represented by start and end points, then the dashed lane line formed by the intersection point as the start point and the original end point of the current dashed line is the lane line that correctly corresponds to the 2D dashed lane line labeled in the image. The specific effect is as follows: Figure 2 As shown, the dashed lines are the lane lines directly projected, and the solid lines are the effect after adaptive clipping.

[0083] Therefore, by utilizing the vehicle's pose data in the point cloud map, as well as the camera's extrinsic data, intrinsic data, distortion data, and vehicle mask data, this application embodiment can adaptively handle the issues of view cone clipping and vehicle occlusion during the transformation and projection process from 3D lane lines to images, thereby obtaining a projection of suitable length and greatly improving the accuracy of lane line consistency assessment.

[0084] In step S102, the 2D lane lines labeled in the image and the 2D lane lines projected in the 3D are dilated respectively. The intersection-union ratio of each dilated 2D lane line labeled in the image and each dilated 2D lane line projected in the 3D is calculated. Based on the intersection-union ratio, multiple pairs of associated lines between the 2D lane lines labeled in the image and the 2D lane lines projected in the 3D are determined.

[0085] It is understandable that 3D projected 2D lane lines, which are projected onto a pixel coordinate system and have undergone processing for view frustum clipping and vehicle occlusion, need to be automatically associated with the image-annotated 2D lane lines to achieve further consistency index calculation. This application embodiment uses the intersection-union ratio (IUGR) index between line pairs of the expanded lane lines to perform corresponding association and erroneous association filtering between the 2D lane lines obtained from the 3D lane line transformation projection and the image-annotated 2D lane lines.

[0086] In this embodiment, the intersection-union ratio (IURR) between line pairs refers to the ratio of the intersection to the union of two rectangular regions after two line segments are expanded. In this embodiment, lane line correspondence and erroneous association filtering refer to selecting the pair with the highest IURR as the best association pair based on the IURR of the 2D lane lines obtained after transformation and projection of the 3D lane lines and the expanded 2D lane lines directly labeled on the image, and filtering out association pairs below a certain threshold.

[0087] Specifically, in this embodiment of the application, a dilation operation is performed on the 3D projected 2D lane lines represented in the image coordinate system and the 2D lane lines labeled in the image. The intersection-union ratio of each 3D projected 2D lane line and each 2D lane line labeled in the image after dilation is calculated, and then all intersection-union ratios are cross-validated.

[0088] It should be noted that the cross-validation process in this embodiment refers to selecting the line pairs with the highest intersection-union ratio (IU) as candidate associated line pairs, i.e., the associated line pairs of 2D lane lines annotated in multiple sets of images and 3D projected 2D lane lines in this embodiment. Furthermore, this embodiment can set a preset IU threshold to filter out associated line pairs with IU ratios lower than the preset threshold.

[0089] Therefore, this application embodiment utilizes the intersection-union ratio index of the cross lines between the expanded 3D projected 2D lane lines and the image-annotated 2D lane lines to sort all line pairs by relevance. At the same time, a certain intersection-union ratio threshold can be set to filter out errors and match groups.

[0090] In step S103, based on each group of associated line pairs and the type of the corresponding associated line pairs, the evaluation result of each group of associated line pairs is obtained according to the preset evaluation strategy, and the evaluation results of all groups of associated line pairs are statistically analyzed to obtain the consistency evaluation result of 3D lane lines and 2D lane lines.

[0091] It is understandable that the 3D projection of 2D lane lines and the image annotation of 2D lane lines are associated in the same frame, respectively, for different types of lane lines, namely continuous lane lines and dashed lane lines, and the two lane lines are evaluated using different consistency average indicators.

[0092] Specifically, in this embodiment, based on the type of associated line pairs (continuous or dashed), an evaluation result for each pair of associated line pairs is obtained according to a preset evaluation strategy. This preset evaluation strategy refers to evaluating the consistency between associated continuous lane lines based on the average vertical Euclidean distance between point lines, and evaluating the consistency between associated dashed lane lines based on the average Euclidean distance between endpoints. Furthermore, based on the above steps, all frames within a data packet are processed, and the results are statistically analyzed to obtain a quantitative evaluation index of the consistency between all 2D and 3D lane lines within a data packet.

[0093] In this embodiment, the average Euclidean distance between points and lines refers to the average perpendicular distance from each discrete point on the 2D solid lane line (obtained after transformation and projection of the 3D solid lane line) to the point directly marked on the 2D solid lane line in the image, within an already associated pair of solid lane lines. The average Euclidean distance between endpoints in this embodiment refers to the average point-to-point distance from the start and end points of the 2D dashed lane line (obtained after transformation and projection of the 3D dashed lane line) to the start and end points of the 2D dashed lane line directly marked on the image, within an already associated pair of dashed lane lines.

[0094] Optionally, in some embodiments, based on each pair of associated lines and the type of the corresponding associated line pair, the evaluation result of each pair of associated lines is obtained according to a preset evaluation strategy, including: when the current associated line pair is a continuous lane line type, determining multiple data points on the current 3D projected 2D lane line, and determining the specified line segment of each data point on the 2D lane line labeled in the current image; calculating the vertical distance between each data point and the corresponding specified line segment, and taking the average of all vertical distances as the evaluation result of the current associated line pair.

[0095] For example, when the current associated lane pair is a continuous lane line type, the specific evaluation process is as follows:

[0096] (1) The embodiments of this application are based on 3D projection 2D continuous lane line L 3d2d_cl Each data point in the dataset is denoted as P. i In the corresponding associated image annotation of the 2D continuous lane line L 2d_cl Searching for distance P i The two closest points, denoted as P. i a P i b Thus, the line segment is obtained. That is, the designated line segment on the 2D lane line annotated in the current image of this application embodiment;

[0097] (2) Calculate the 3D projection of the 2D continuous lane line L 3d2d_cl Each data point P i The corresponding 2D continuous lane line L in the image annotation 2d_cl Median segment P i a P i b The vertical distance between them is denoted as d. i ,like Figure 3 As shown in the dashed circle;

[0098] (3) Projecting 2D continuous lane lines L in 3D 3d2d_cl All data points {P0,…,P} n The calculated corresponding distances {d0,…,d} n The average value is calculated and denoted as} That is, 3D projection 2D continuous lane line L 3d2d_cl 2D continuous lane lines L with image annotation 2d_cl Consistency assessment results between them;

[0099] (4) For each 3D projection of the 2D continuous lane line L in the current frame 3d2d_cl Perform the above steps (1)-(3) and average all the obtained consistency evaluation results to obtain the consistency evaluation results of all continuous 3D lane lines and 2D lane lines in the current frame.

[0100] Optionally, in some embodiments, based on each pair of associated lines and the type of the corresponding associated line pair, an evaluation result for each pair of associated lines is obtained according to a preset evaluation strategy, including: when the current associated line pair is a dashed lane line type, determining the first starting point and the first ending point on the current 3D projected 2D lane line, and the second starting point and the second ending point on the current image-annotated 2D lane line; obtaining the first Euclidean distance based on the first starting point and the second starting point, and obtaining the second Euclidean distance based on the first ending point and the second ending point; and using the average of the first Euclidean distance and the second Euclidean distance as the evaluation result of the current associated line pair.

[0101] For example, when the current associated lane pair is a dashed lane line type, the specific evaluation process is as follows:

[0102] (1) Calculate the 3D projection and 2D dashed lane line L respectively. 3d2d_dl The start and end points P0 and P1, and the corresponding 2D dashed lane lines L in the image annotation. 2d_dl Start and end points P a P b The Euclidean distance between two points is denoted as d0 and d1, as follows: Figure 3 As shown in the solid circle;

[0103] (2) Calculate the average value of the Euclidean distances d0 and d1 between the start and end points of the dashed lane line associated with the calculated dashed lane line in (1), and record it as . That is, 3D projection 2D dashed lane line L 3d2d_dl The 2D dashed lane line L labeled in the image 2d_dl Consistency assessment results between them;

[0104] (3) For each 3D projected 2D dashed lane line L in the current frame 3d2d_dl Perform steps (1) and (2) and average all the obtained consistency evaluation results to obtain the consistency evaluation results of all dashed 3D lane lines and 2D lane lines in the current frame.

[0105] Therefore, the embodiments of this application can use the average point-to-point distance between the two ends of a line segment to perform a quantitative evaluation of the consistency between dashed lines; and can use the average perpendicular distance from a point to a line to perform a quantitative evaluation of the consistency between solid lines.

[0106] Furthermore, based on the steps of the above embodiments, this application embodiment can process all frames within a data packet, obtain the consistency evaluation results between all types of 3D lane lines and 2D lane lines in all frames, and calculate the average of the consistency evaluation results between each type of 3D lane line and 2D lane line in all frames, that is, obtain the consistency quantitative evaluation index between each type of 3D lane line and 2D lane line in a data packet.

[0107] This can greatly improve the accuracy of BEV lane lines generated by the deep learning model on the vehicle, providing an important basis for downstream positioning and mapping, path planning and driving control, as well as the normal execution of intelligent assisted driving functions and driving safety.

[0108] To enable those skilled in the art to further understand the consistency evaluation method of 3D lane lines and 2D lane lines in the embodiments of this application, the following examples illustrate the implementation steps of the method.

[0109] Specifically, Figure 4 A flowchart illustrating a specific embodiment of the consistency evaluation method for 3D lane lines and 2D lane lines in this application is shown below. Figure 4 As shown, the consistency assessment method for 3D lane lines and 2D lane lines includes the following steps:

[0110] Step S401: Based on the vehicle's pose in the point cloud map, transform the 3D lane lines in the point cloud map coordinate system to the vehicle coordinate system.

[0111] Step S402: Based on the extrinsic parameters from the vehicle coordinate system to the camera coordinate system, transform the 3D lane lines in the vehicle coordinate system to the camera coordinate system.

[0112] Step S403: Based on the camera intrinsic parameters and the vehicle mask, project the 3D lane line in the camera coordinate system to the camera image pixel coordinate system;

[0113] Step S404: Check and address camera view frustum clipping and vehicle occlusion issues;

[0114] Step S405: Based on the intersection-union ratio index between lane pairs after expansion, perform lane line correspondence association and erroneous association filtering between the 2D lane lines obtained by 3D lane line transformation projection and the 2D lane lines labeled in the image.

[0115] Step S406: Evaluate the consistency between associated solid lines based on the average vertical Euclidean distance between the points and lines, and evaluate the consistency between associated dashed lines based on the average Euclidean distance between the endpoints.

[0116] Step S407: Based on the above steps, process all frames in a data packet and perform statistics on the results to obtain a quantitative evaluation index of consistency between all 2D lane lines and 3D lane lines in a data packet.

[0117] This enables the automatic and comprehensive evaluation of the consistency between 3D and 2D lane lines, while ensuring the accuracy and comprehensiveness of the evaluation results. This improves the efficiency of the evaluation process and ensures the reliability of the evaluation results.

[0118] According to the consistency evaluation method for 3D lane lines and 2D lane lines proposed in the embodiments of this application, the method acquires the image-annotated 2D lane lines, the first 3D lane lines in the point cloud map coordinate system, the pose of the current vehicle in the point cloud map coordinate system, camera extrinsic parameters, and camera intrinsic parameters. Based on the pose, camera extrinsic parameters, and camera intrinsic parameters, the first 3D lane lines are transformed into 3D projected 2D lane lines. The image-annotated 2D lane lines and the 3D projected 2D lane lines are dilated respectively. The intersection-union ratio (IUGR) of each dilated image-annotated 2D lane line and each dilated 3D projected 2D lane line is calculated. Based on the IUGR, multiple sets of associated line pairs of image-annotated 2D lane lines and 3D projected 2D lane lines are determined. Based on each set of associated line pairs and the type of the corresponding associated line pairs, the evaluation result of each set of associated line pairs is obtained according to a preset evaluation strategy. The evaluation results of all sets of associated line pairs are statistically analyzed to obtain the consistency evaluation result of 3D lane lines and 2D lane lines. This solves the problems of inconsistency between 3D and 2D lane line ground truth data and limitations in the evaluation methods for 3D and 2D lane line ground truth, which lead to inaccurate and incomplete evaluation results. It automatically and comprehensively evaluates the consistency of 3D and 2D lane lines, improving the efficiency of the evaluation process and the reliability of the evaluation results.

[0119] Next, referring to the accompanying drawings, a consistency evaluation device for 3D lane lines and 2D lane lines according to an embodiment of this application is described.

[0120] Figure 5 This is a block diagram of a consistency evaluation device for 3D lane lines and 2D lane lines according to an embodiment of this application.

[0121] like Figure 5 As shown, the consistency evaluation device 10 for 3D lane lines and 2D lane lines includes: a transformation module 100, an association module 200, and an evaluation module 300.

[0122] Specifically, the transformation module 100 is used to acquire the 2D lane lines labeled in the image, the first 3D lane lines in the point cloud map coordinate system, the pose of the current vehicle in the point cloud map coordinate system, camera extrinsic parameters, and camera intrinsic parameters, and transform the first 3D lane lines into 3D projected 2D lane lines based on the pose, camera extrinsic parameters, and camera intrinsic parameters; the association module 200 is used to perform dilation operations on the 2D lane lines labeled in the image and the 3D projected 2D lane lines respectively, calculate the intersection-union ratio of each dilated 2D lane line labeled in the image and each dilated 2D projected 2D lane line, and determine multiple sets of associated line pairs of 2D lane lines labeled in the image and 2D projected 2D lane lines based on the intersection-union ratio; the evaluation module 300 is used to obtain the evaluation result of each set of associated line pairs based on each set of associated line pairs and the type of the corresponding associated line pairs according to a preset evaluation strategy, and statistically analyze the evaluation results of all sets of associated line pairs to obtain the consistency evaluation result of the 3D lane lines and the 2D lane lines.

[0123] Optionally, in some embodiments, the evaluation module 300 is specifically used to: when the current associated line pair is a continuous lane line type, determine multiple data points on the current 3D projected 2D lane line, and determine the specified line segment of each data point on the 2D lane line labeled in the current image; calculate the vertical distance between each data point and the corresponding specified line segment, and take the average of all vertical distances as the evaluation result of the current associated line pair.

[0124] Optionally, in some embodiments, the evaluation module 300 is specifically used to: when the current associated line pair is a dashed lane line type, determine the first starting point and the first ending point on the current 3D projected 2D lane line, and the second starting point and the second ending point on the current image-annotated 2D lane line; obtain a first Euclidean distance based on the first starting point and the second starting point, and obtain a second Euclidean distance based on the first ending point and the second ending point; and use the average of the first Euclidean distance and the second Euclidean distance as the evaluation result of the current associated line pair.

[0125] Optionally, in some embodiments, the transformation module 100 is specifically used to: transform the first 3D lane line into a second 3D lane line in the vehicle coordinate system according to the pose, and transform the second 3D lane line into a third 3D lane line in the camera coordinate system according to the camera extrinsic parameters; and project the third 3D lane line onto the image pixel coordinate system according to the camera intrinsic parameters based on a preset cropping and processing strategy to obtain a 3D projected 2D lane line.

[0126] Optionally, in some embodiments, the transformation module 100 is further configured to: select multiple 3D points on the third 3D lane line based on a preset filtering strategy; perform linear interpolation on the multiple 3D points, and project each linearly interpolated 3D point onto the image pixel coordinate system based on camera intrinsic parameters to obtain multiple projection points and a projection image; determine whether each projection point meets the preset filtering conditions; if the current projection point meets the preset filtering conditions, delete the current projection point, and use the lane line composed of the remaining projection points as the 3D projected 2D lane line.

[0127] It should be noted that the explanation of the aforementioned embodiment of the consistency evaluation method for 3D lane lines and 2D lane lines also applies to the consistency evaluation device for 3D lane lines and 2D lane lines in this embodiment, and will not be repeated here.

[0128] According to the consistency evaluation device for 3D lane lines and 2D lane lines proposed in the embodiments of this application, the device acquires the image-annotated 2D lane lines, the first 3D lane lines in the point cloud map coordinate system, the pose of the current vehicle in the point cloud map coordinate system, camera extrinsic parameters, and camera intrinsic parameters. Based on the pose, camera extrinsic parameters, and camera intrinsic parameters, the first 3D lane lines are transformed into 3D projected 2D lane lines. The image-annotated 2D lane lines and the 3D projected 2D lane lines are dilated respectively. The intersection-union ratio (IUGR) of each dilated image-annotated 2D lane line and each dilated 3D projected 2D lane line is calculated. Based on the IUGR, multiple sets of associated line pairs of image-annotated 2D lane lines and 3D projected 2D lane lines are determined. Based on each set of associated line pairs and the type of the corresponding associated line pairs, the evaluation result of each set of associated line pairs is obtained according to a preset evaluation strategy. The evaluation results of all sets of associated line pairs are statistically analyzed to obtain the consistency evaluation result of 3D lane lines and 2D lane lines. This solves the problems of inconsistency between 3D and 2D lane line ground truth data and limitations in the evaluation methods for 3D and 2D lane line ground truth, which lead to inaccurate and incomplete evaluation results. It automatically and comprehensively evaluates the consistency of 3D and 2D lane lines, improving the efficiency of the evaluation process and the reliability of the evaluation results.

[0129] Figure 6 A schematic diagram of the structure of a vehicle provided in an embodiment of this application. The vehicle may include:

[0130] The memory 601, the processor 602, and the computer program stored on the memory 601 and capable of running on the processor 602.

[0131] When the processor 602 executes the program, it implements the consistency evaluation method for 3D lane lines and 2D lane lines provided in the above embodiments.

[0132] Furthermore, the vehicle also includes:

[0133] Communication interface 603 is used for communication between memory 601 and processor 602.

[0134] The memory 601 is used to store computer programs that can run on the processor 602.

[0135] The memory 601 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0136] If the memory 601, processor 602, and communication interface 603 are implemented independently, then the communication interface 603, memory 601, and processor 602 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0137] Optionally, in a specific implementation, if the memory 601, processor 602, and communication interface 603 are integrated on a single chip, then the memory 601, processor 602, and communication interface 603 can communicate with each other through an internal interface.

[0138] The processor 602 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0139] This application also provides a computer program product on which a computer program is stored, which, when executed, implements the above-described method for evaluating the consistency of 3D lane lines and 2D lane lines.

[0140] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0141] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0142] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0143] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0144] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0145] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for consistency evaluation of 3D lane lines and 2D lane lines, characterized in that, The method comprises the following steps: obtaining image-labeled 2D lane lines, first 3D lane lines in a point cloud map coordinate system, a current vehicle pose in the point cloud map coordinate system, camera extrinsic parameters, and camera intrinsic parameters, and transforming the first 3D lane lines into 3D projection 2D lane lines according to the pose, the camera extrinsic parameters, and the camera intrinsic parameters; respectively performing dilation operations on the image-labeled 2D lane lines and the 3D projection 2D lane lines, calculating intersection-over-union ratios of each dilated image-labeled 2D lane line and each dilated 3D projection 2D lane line, and determining a plurality of groups of associated line pairs of the image-labeled 2D lane lines and the 3D projection 2D lane lines according to the intersection-over-union ratios; based on each group of associated line pairs and a type of the corresponding associated line pair, obtaining an evaluation result of each group of associated line pairs according to a preset evaluation strategy, and statistically processing the evaluation results of all groups of associated line pairs to obtain a consistency evaluation result of the 3D lane lines and the image-labeled 2D lane lines; wherein, the evaluation result of each group of associated line pairs is obtained according to the preset evaluation strategy based on each group of associated line pairs and the type of the corresponding associated line pair, comprising: when the current associated line pair is a continuous lane line type, a plurality of data points on the current 3D projection 2D lane line are determined, and a specified line segment of each data point on the current image-labeled 2D lane line is determined; the perpendicular distance between each data point and the corresponding specified line segment is calculated, and the average of all perpendicular distances is taken as the evaluation result of the current associated line pair; the evaluation result of each group of associated line pairs is obtained according to the preset evaluation strategy based on each group of associated line pairs and the type of the corresponding associated line pair, comprising: when the current associated line pair is a dashed lane line type, a first starting point and a first ending point on the current 3D projection 2D lane line and a second starting point and a second ending point on the current image-labeled 2D lane line are respectively determined; a first Euclidean distance is obtained according to the first starting point and the second starting point, and a second Euclidean distance is obtained according to the first ending point and the second ending point; the average of the first Euclidean distance and the second Euclidean distance is taken as the evaluation result of the current associated line pair.

2. The method of claim 1, wherein, the first 3D lane lines are transformed into 3D projection 2D lane lines according to the pose, the camera extrinsic parameters, and the camera intrinsic parameters, comprising: the first 3D lane lines are transformed into second 3D lane lines in a vehicle coordinate system according to the pose, and the second 3D lane lines are transformed into third 3D lane lines in a camera coordinate system according to the camera extrinsic parameters; the third 3D lane lines are projected to an image pixel coordinate system based on a preset cutting and processing strategy according to the camera intrinsic parameters, to obtain the 3D projection 2D lane lines.

3. The method of claim 2, wherein, the third 3D lane lines are projected to an image pixel coordinate system based on a preset cutting and processing strategy according to the camera intrinsic parameters, further comprising: a plurality of 3D points are selected on the third 3D lane lines based on a preset screening strategy; Linear interpolation is performed on the plurality of 3D points, and each linearly interpolated 3D point is projected to the image pixel coordinate system based on the camera intrinsic parameter to obtain a plurality of projection points and a projection image; It is judged whether each projection point satisfies a preset filtering condition; If the current projection point satisfies the preset filtering condition, the current projection point is deleted, and a lane line composed of the remaining projection points is taken as the 3D projection 2D lane line.

4. An apparatus for consistency evaluation of 3D lane lines and 2D lane lines, characterized by, Comprise: The transformation module is used for obtaining image-labeled 2D lane lines, first 3D lane lines in a point cloud map coordinate system, a current vehicle pose in the point cloud map coordinate system, a camera extrinsic parameter, and a camera intrinsic parameter, and transforming the first 3D lane lines into 3D projection 2D lane lines according to the pose, the camera extrinsic parameter and the camera intrinsic parameter; The association module is used for respectively performing inflation operations on the image-labeled 2D lane lines and the 3D projection 2D lane lines, calculating intersection-over-union ratios of each inflated image-labeled 2D lane line and each inflated 3D projection 2D lane line, and determining a plurality of groups of image-labeled 2D lane lines and 3D projection 2D lane line association line pairs according to the intersection-over-union ratios; The evaluation module is used for obtaining evaluation results of each group of association line pairs according to a preset evaluation strategy based on each group of association line pairs and a type of the corresponding association line pair, and statistically obtaining a consistency evaluation result of the 3D lane lines and the image-labeled 2D lane lines by evaluating all groups of association line pairs. The evaluation module is specifically configured to: when a current association line pair is a continuous lane line type, a plurality of data points on a current 3D projection 2D lane line are determined, and a specified line segment of each data point on a current image-labeled 2D lane line is determined; a perpendicular distance between each data point and the corresponding specified line segment is calculated, and a mean value of all perpendicular distances is taken as an evaluation result of the current association line pair. The evaluation module is specifically configured to: when the current association line pair is a dashed lane line type, a first starting point and a first ending point on the current 3D projection 2D lane line are respectively determined, and a second starting point and a second ending point on the current image-labeled 2D lane line are respectively determined; A first Euclidean distance is obtained according to the first starting point and the second starting point, and a second Euclidean distance is obtained according to the first ending point and the second ending point; and a mean value of the first Euclidean distance and the second Euclidean distance is taken as an evaluation result of the current association line pair.

5. A vehicle characterized by comprising: Comprise: A memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the consistency evaluation method of the 3D lane line and the 2D lane line according to any one of claims 1-3.

6. A computer program product comprising a computer program, characterized in that, The computer program is executed to implement the consistency evaluation method of the 3D lane line and the 2D lane line according to any one of claims 1-3.

Images