Verification method for lane marking, related generation method, and related equipment

Abstract

The present invention discloses a verification method for lane line marking, a related generation method, and related equipment. The verification method includes: obtaining lane line detection data for lane line marking, and performing a first evaluation on the lane line detection data to obtain a first evaluation result; obtaining a lane line map generated based on the lane line detection data, and performing a second evaluation on the lane line map to obtain a second evaluation result; and obtaining verification information for lane line marking based on the first evaluation result and the second evaluation result, wherein the verification information is used to characterize the accuracy of the lane line marking. Lane line marking is adjusted based on the verification information, thereby improving the accuracy of lane line marking, enhancing the safety of autonomous driving, and saving manpower and material resources.

Description

Technical Field

[0001] The present invention relates to the field of high-precision map technology, and in particular to a verification method for lane marking, a related generation method, and related equipment. Background Art

[0002] In areas such as autonomous driving and high-precision mapping, map data collection typically involves both professional and crowdsourcing methods. Professional data collection typically requires a large number of specialized data collectors, surveying equipment, and collection vehicles. While this method offers high data accuracy, it requires extensive professional support, is costly, and makes subsequent data updates difficult. Consequently, crowdsourcing, with its high update frequency, minimal computational effort, and fast data transmission, is gaining increasing popularity.

[0003] The main on-board sensors used in crowdsourcing mapping include cameras, LiDAR (Light Detection and Ranging), and RTK GPS (Real-Time Kinematic Global Positioning System). Both LiDAR and cameras can sense the vehicle's surroundings and provide positioning capabilities. RTK GPS can even provide millimeter-level positioning accuracy when the signal is good. By processing the data collected by these sensors, three-dimensional semantic maps can be created and updated. The information on traffic elements such as roads, traffic signs, lane lines, obstacles, and pedestrians in the semantic map will be used to control the vehicle's steering, speed, path planning, lane changes, etc. Among them, the precise marking of lane lines directly affects many safety behaviors of autonomous vehicles.

[0004] For L4 / L5 autonomous driving vehicles, high-precision Figure 1 Lane marking information and coordinates generally have precise three-dimensional physical and semantic information, and lane markings need to be accurately annotated. Currently, most lane markings are generated by visually generating two-dimensional lane markings, which only contain information on the x- and y-axes, but no z-axis height information. Two-dimensional lane marking information is generally sufficient for autonomous driving purposes such as path planning under simple road conditions. However, in complex road conditions such as urban roads, such as elevated bridges, the lack of z-axis height information will directly lead to vehicle path control failures. Manual annotation is also labor-intensive and time-consuming.

[0005] Therefore, the automated marking of three-dimensional lane lines is particularly important. At the same time, the verification of the automated marking of three-dimensional lane lines will directly affect the accuracy and use of lane line marking. Summary of the Invention

[0006] The main technical problem solved by the present invention is to provide a verification method and related generation method and related equipment for lane line marking, which can improve the accuracy of lane line marking, enhance the safety of autonomous driving, and save manpower and material resources.

[0007] To solve the above technical problems, the present invention adopts a technical solution: providing a verification method for lane marking, comprising:

[0008] Acquiring lane line detection data for lane line marking, and performing a first evaluation on the lane line detection data to obtain a first evaluation result;

[0009] Obtaining a lane map generated based on the lane detection data, and performing a second evaluation on the lane map to obtain a second evaluation result;

[0010] Verification information for lane marking is obtained based on the first evaluation result and the second evaluation result, wherein the verification information is used to characterize the accuracy of the lane marking.

[0011] The lane line detection data includes a first portion of data and a second portion of data, the first portion of data corresponds to a main image, and the second portion of data corresponds to the main image and at least one auxiliary image, and both the main image and the auxiliary image are about lane lines;

[0012] The first evaluation of the lane line detection data to obtain a first evaluation result includes:

[0013] Obtaining a first error of the first portion of data relative to lane line point cloud data, wherein the lane line point cloud data is used to represent the lane line;

[0014] Obtaining a second error and a third error of the second portion of data relative to the lane line point cloud data;

[0015] The first evaluation result is obtained according to the first error, the second error, and the third error.

[0016] The first error is an average accuracy rate of lane line detection on the main image in the first part of data;

[0017] The second error is the distance between a point in the lane line point cloud data and a line connecting two points of the lane line in the second portion of data;

[0018] The third error is the chamfer distance between the lane line point in the second portion of data and the lane line point cloud data;

[0019] The first evaluation result is a weighted sum of the first error, the second error, and the third error.

[0020] The obtaining of a second error of the second portion of data relative to the lane line point cloud data includes:

[0021] Obtain each point in the lane line point cloud data, and obtain two points in the second part of data that are closest to each point;

[0022] Obtaining a first distance from each point to a line connecting the corresponding two points, thereby obtaining a plurality of first distances;

[0023] obtaining an average of a plurality of the first distances to define the average as the second error;

[0024] The obtaining of a third error of the second portion of data relative to the lane line point cloud data includes:

[0025] Obtain each point in the lane line point cloud data, and obtain the point in the second part of data that is closest to each point;

[0026] Obtaining a second distance from each point to a corresponding point, thereby obtaining a plurality of second distances;

[0027] A sum of a plurality of the second distances is obtained to be defined as the third error.

[0028] The performing a second evaluation on the lane map to obtain a second evaluation result includes:

[0029] Obtaining a fourth error of the lane map relative to a ground-truth lane map, wherein the ground-truth lane map is annotated based on lane point cloud data;

[0030] Obtaining a fifth error of the lane line map relative to the lane line ground truth map;

[0031] The second evaluation result is obtained according to the fourth error and the fifth error.

[0032] The fourth error is the distance between a point in the lane line true value map and a line connecting two points of the lane line in the lane line map;

[0033] The fifth error is a chamfer distance between a point on the lane line in the lane line map and a point on the lane line ground truth map;

[0034] The second evaluation result is a weighted sum of the fourth error and the fifth error.

[0035] The obtaining of a fourth error of the lane line map relative to the lane line true value map includes:

[0036] Obtain each point in the lane line true value map, and obtain the two points in the lane line map closest to each point;

[0037] Obtaining a third distance from each point to a line connecting the corresponding two points, thereby obtaining a plurality of third distances;

[0038] obtaining an average value of a plurality of the third distances to define the average value as the fourth error;

[0039] The obtaining of a fifth error of the lane line map relative to the lane line true value map includes:

[0040] Obtain each point in the lane line true value map, and obtain the point in the lane line map closest to each point;

[0041] Obtaining a fourth distance from each point to a corresponding point, thereby obtaining a plurality of fourth distances;

[0042] A sum of a plurality of the fourth distances is obtained to be defined as the fifth error.

[0043] The second technical solution adopted by the present invention is to provide a method for generating a driving map, comprising:

[0044] Execute the lane marking process on the original map to obtain the lane marking result;

[0045] Performing a verification process on the lane marking result to generate the driving map;

[0046] Wherein, the verification process includes the verification method for lane line marking.

[0047] The third technical solution adopted by the present invention is: providing a vehicle-mounted device, including a processor and a memory coupled to the processor, the memory is used to store a computer program, and the processor is used to execute the computer program to implement the verification method for lane line marking or the driving map generation method.

[0048] The fourth technical solution adopted by the present invention is: providing a non-volatile computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, it is used to implement the verification method for lane line marking or the generation method of the driving map.

[0049] The above scheme performs a first evaluation on the acquired lane line detection data to obtain a first evaluation result and generates a lane line map. According to the first evaluation result, the accuracy of the lane line detection data can be verified; a second evaluation is performed on the lane line map to obtain a second evaluation result. According to the second evaluation result, the accuracy of the lane line map can be verified; according to the first and second evaluation results, verification information for lane line marking is obtained. The verification information can represent the accuracy of the lane line marking. The lane line marking is adjusted according to the verification information, thereby improving the accuracy of the lane line marking, enhancing the safety of autonomous driving, and saving manpower and material resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of an embodiment of a verification method for lane marking according to the present invention;

[0051] Figure 2 It is a structural diagram of an embodiment of the vehicle-mounted device of the present invention;

[0052] Figure 3 It is a structural diagram of an embodiment of a non-volatile computer-readable storage medium of the present invention. DETAILED DESCRIPTION

[0053] The present application will be further described in detail below in conjunction with the accompanying drawings and examples. It is particularly noted that the following examples are only intended to illustrate the present application and are not intended to limit the scope of the present application. Similarly, the following examples are only some examples of the present application and not all examples. All other examples obtained by those of ordinary skill in the art without creative work are intended to fall within the scope of protection of this application.

[0054] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0055] It should be noted that the terms "first", "second" and "third" in this application are only used for descriptive purposes and should not be understood as indicating or suggesting relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first", "second" and "third" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units that are inherent to these processes, methods, products or devices.

[0056] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of the present invention's lane marking verification method. It should be noted that the method of the present invention is not limited to the lane marking method if the results are substantially the same. Figure 1 The process sequence shown is limited. This method can be applied to vehicle-mounted equipment with computing and other functions. The vehicle-mounted equipment can execute this method by receiving lane line information collected by sensor equipment. The sensor equipment can be a laser radar or camera installed on the vehicle. The sensor equipment captures road information in real time while the vehicle is driving. The road information contains multiple frames of road images. Figure 1 As shown, the method includes the following steps:

[0057] S1. Acquire lane line detection data for lane line marking, and perform a first evaluation on the lane line detection data to obtain a first evaluation result.

[0058] To obtain lane detection data for lane marking, terminal sensor devices perceive the vehicle's surrounding environment and lane information, and output lane detection data using corresponding perception models and algorithms. For example, a crowdsourced collection vehicle uses a LiDAR to collect LiDAR point cloud data of the lane lines on the road the vehicle is traveling on. A camera on the collection vehicle collects image data of the lane lines. The image data includes the road scene surrounding the vehicle and can be down-converted to maintain the same frequency as the LiDAR. The LiDAR point cloud data and image data are input into the perception model for training to generate lane detection data.

[0059] The lane line detection data is evaluated for the first time to obtain a first evaluation result. The accuracy of the lane line detection data output by the perception model can be verified based on the first evaluation result.

[0060] S2. Obtain a lane map generated based on the lane detection data, and perform a second evaluation on the lane map to obtain a second evaluation result.

[0061] Based on the lane detection data and the results of the first evaluation, a lane map is generated through multi-frame accumulation, coordinate system conversion, curve fitting, and curve point resampling. A second evaluation is performed on the generated lane map, and the accuracy of the generated lane map can be verified based on the results of the second evaluation.

[0062] S3. Acquire verification information for lane line marking based on the first evaluation result and the second evaluation result, wherein the verification information is used to represent the accuracy of the lane line marking.

[0063] Based on the first and second evaluation results, verification information for lane marking is obtained. The verification information can represent the accuracy of the lane marking. It is understood that adjusting the lane marking based on the verification information can improve the accuracy of the lane marking. That is, by adjusting the training process of the perception model, the impact on the output lane detection data is reduced, thereby improving the accuracy of the output lane detection data. Alternatively, by adjusting the curve fitting process, the impact on the generated lane map is reduced, thereby improving the accuracy of the output lane map. This can achieve updated lane markings for known road sections and automatic lane marking for unknown road sections.

[0064] In this embodiment, a first evaluation is performed on the acquired lane line detection data to obtain a first evaluation result, and a lane line map is generated. The accuracy of the lane line detection data can be verified based on the first evaluation result; a second evaluation is performed on the lane line map to obtain a second evaluation result. The accuracy of the lane line map can be verified based on the second evaluation result; based on the first evaluation result and the second evaluation result, verification information for lane line marking is obtained. The verification information can characterize the accuracy of the lane line marking. The lane line marking is adjusted based on the verification information, thereby improving the accuracy of the lane line marking, enhancing the safety of autonomous driving, and saving manpower and material resources.

[0065] In one embodiment of the present invention, the lane line detection data includes a first part of data and a second part of data, the first part of data corresponds to a main picture, the second part of data corresponds to the main picture and at least one auxiliary picture, and the main picture and the auxiliary picture are both about the lane line.

[0066] In this embodiment, multiple cameras, for example, six cameras, can be positioned around the vehicle to capture image data of the road surrounding the vehicle. One of the cameras is defined as the primary camera, for example, the camera positioned directly in front of and in the center of the vehicle, and the remaining cameras are defined as auxiliary cameras. The image data captured by the primary camera is defined as the primary image, and the image data captured by the auxiliary cameras is defined as the auxiliary images. During lane line detection, the primary and auxiliary images are both captured at the same time and are images of the road, from which lane line data can be detected, i.e., lane line detection data. The primary image is processed, for example, through recognition and detection, to obtain a first portion of data. The first portion of data includes data corresponding to lane lines in the primary image, i.e., the first portion of data corresponds to the primary image. The primary and auxiliary images are input into a perception model for training, and the perception model outputs a second portion of data. The second portion of data includes data corresponding to lane lines in both the primary and auxiliary images, i.e., the second portion of data corresponds to both the primary and auxiliary images.

[0067] It is understandable that the total number of cameras can also be set to seven, eight, or other achievable numbers without specific limitation.

[0068] At this time, the first evaluation of the lane line detection data to obtain a first evaluation result includes:

[0069] Obtaining a first error of the first portion of data relative to lane line point cloud data, wherein the lane line point cloud data is used to represent the lane line;

[0070] Obtaining a second error and a third error of the second portion of data relative to the lane line point cloud data;

[0071] The first evaluation result is obtained according to the first error, the second error, and the third error.

[0072] Road information is collected by a professional map collection vehicle equipped with high-precision data acquisition equipment. The collected road information is then positioned and mapped to generate lane line point cloud data. This lane line point cloud data serves as ground truth data and is used to verify the accuracy of the first and second data parts. A first error is obtained for the first data part relative to the lane line point cloud data, while second and third errors are obtained for the second data part relative to the lane line point cloud data. Based on these first, second, and third errors, a first evaluation result is generated, which verifies the accuracy of the first and second data parts, as well as the training accuracy of the perception model.

[0073] In one embodiment of the present invention, the first error is the average accuracy of lane line detection on the main image in the first part of the data; the second error is the distance between a point in the lane line point cloud data and a line connecting two points of the lane line in the second part of the data; the third error is the chamfer distance between a point of the lane line in the second part of the data and a point in the lane line point cloud data; the first evaluation result is the weighted sum of the first error, the second error and the third error.

[0074] In this embodiment, the first part of the data and the lane line point cloud data are projected into the same coordinate system, and the intersection over union (IoU) of the first part of the data relative to the lane line point cloud data is calculated, that is, the intersection over union (IoU) of the lane line image data and the lane line point cloud data in the main image is calculated to obtain the IoU calculation result. The IoU calculation result is compared with the preset IoU overlap threshold (IoU threshold). If the IoU calculation result is greater than the IoUthreshold, it indicates that the prediction is correct (True Positive, TP). If the IoU calculation result is less than or equal to the IoUthreshold, it indicates that the prediction is wrong (False Positive, FP). The calculation formula for accuracy (Precision) is:

[0075] P=TP / (TP+FP)

[0076] The calculation formula for recall is:

[0077] R=TP / (TP+FN)

[0078] Among them, FN represents the wrong prediction as other classes.

[0079] Based on the calculation results of accuracy and recall, a smooth PR (Precision & Recall) curve is drawn. The average precision (AP) is the area between the PR curve and the horizontal axis Recall, and the average of 40 recall points is used. After obtaining the smooth PR curve, the area under the PR curve is calculated by integration as the final average accuracy value, and the average accuracy value is defined as the fraction S AP , that is, the score S AP is the first error. Score S AP The higher the value, the better the detection accuracy. AP The value range is 0 to 1.

[0080] In this embodiment, the lane line point cloud data includes multiple lane line truth lines, each of which includes multiple truth points. The second part of the data includes multiple lane line detection lines, each of which includes multiple detection points. The lane line truth lines and lane line detection lines are projected into the same coordinate system, and the distance between the truth point in the lane line truth line and the line connecting the two detection points in the lane line detection line is calculated. The calculated result d P2L Defined as the second error.

[0081] Calculate the chamfer distance d between the true value point in the lane line truth line and the detection point in the lane line detection line CD , defined as the third error.

[0082] For the first error S AP , the second error d P2L and the third error d CD Perform weighted summation to obtain the weighted summation result, which is defined as the first evaluation result d. The calculation formula for the first evaluation result d is:

[0083] d=w1*(1-S AP )*(d P2L +d CD )+w2*d P2L +w3*d CD

[0084] Among them, w1, w2, and w3 are weights, w1+w2+w3=1.

[0085] In this embodiment, w1 = 0.25, w2 = 0.4, and w3 = 0.35. The average accuracy is the accuracy of the detection result of the main image corresponding to the first portion of data, that is, the accuracy of the single image. Due to perspective projection, the accuracy will be poor when the perception results in the image are projected into three-dimensional space. Therefore, the value of w1 is minimized. In other embodiments, the values ​​of w1, w2, and w3 can also be other values ​​that are achievable and are not specifically limited.

[0086] In one embodiment of the present invention, obtaining a second error of the second portion of data relative to the lane line point cloud data includes:

[0087] Obtain each point in the lane line point cloud data, and obtain two points in the second part of data that are closest to each point;

[0088] Obtaining a first distance from each point to a line connecting the corresponding two points, thereby obtaining a plurality of first distances;

[0089] An average value of a plurality of first distances is obtained to be defined as the second error.

[0090] In this embodiment, the lane line truth line and the lane line detection line are projected into the same coordinate system, and each truth point in the lane line truth line is obtained. The two detection points in the lane line detection line closest to the truth point in the lane line truth line are searched for. The two searched detection points are connected, and the vertical distance from the truth point to the line connecting the two detection points is calculated. By calculating the vertical distance from each truth point in the lane line truth line to the line connecting the two nearest detection points, multiple first distances can be obtained. The average value d of the multiple first distances is calculated. P2L , define the mean value d P2L is the second error.

[0091] In other embodiments, each detection point in the lane detection line is obtained, and a search is performed near the detection point in the lane detection line for two ground-truth points in the lane detection line that are closest to the detection point. A line is then connected between the two found ground-truth points, and the vertical distance between the detection point and the line connecting the two ground-truth points is calculated. By calculating the vertical distance between each detection point in the lane detection line and the line connecting the two closest ground-truth points in this manner, multiple first distances can also be obtained.

[0092] The obtaining of a third error of the second portion of data relative to the lane line point cloud data includes:

[0093] Obtain each point in the lane line point cloud data, and obtain the point in the second part of data that is closest to each point;

[0094] Obtaining a second distance from each point to a corresponding point, thereby obtaining a plurality of second distances;

[0095] A sum of a plurality of the second distances is obtained to be defined as the third error.

[0096] In this embodiment, each truth point in the lane line truth line is obtained, and the detection point with the minimum distance in the lane line detection line corresponding to each truth point is obtained. The minimum distance between each truth point and the corresponding detection point is obtained, which is the second distance. Multiple second distances are obtained, and then the second distances corresponding to each truth point are summed to obtain the sum result d CD , defined as the third error. Among them, the third error d CD The calculation formula is:

[0097]

[0098] Among them, S1 represents a point cloud composed of multiple true value points, x represents any point in the true value point cloud, S2 represents a point cloud composed of multiple detection points, and y represents any point in the detection point cloud.

[0099] In one embodiment of the present invention, performing a second evaluation on the lane map to obtain a second evaluation result includes:

[0100] Obtaining a fourth error of the lane map relative to a ground-truth lane map, wherein the ground-truth lane map is annotated based on lane point cloud data;

[0101] Obtaining a fifth error of the lane line map relative to the lane line ground truth map;

[0102] The second evaluation result is obtained according to the fourth error and the fifth error.

[0103] The system uses road information collected by a professional map collection vehicle to locate and map lane point cloud data. This highly accurate lane point cloud data is based on the location and mapping of each point, which has varying intensity information. Based on this information, points are manually selected and annotated to create a true lane map. The fourth and fifth errors of the lane map relative to the true lane map are then calculated. A second evaluation result is generated based on these fourth and fifth errors, which can be used to verify the lane map's accuracy.

[0104] In one embodiment of the present invention, the fourth error is the distance between a point in the lane line true value map and a line connecting two points of the lane line in the lane line map; the fifth error is the chamfer distance between a point on the lane line in the lane line map and a point in the lane line true value map; the second evaluation result is the weighted sum of the fourth error and the fifth error.

[0105] In this embodiment, the lane line truth map includes multiple lane line truths, each of which includes multiple truth points. The lane line map also includes multiple lane line detection lines, each of which includes multiple detection points. The lane line truth lines and lane line detection lines are projected into the same coordinate system, and the distance between each truth point on the lane line truth line and the line connecting two detection points on the lane line detection line is calculated. The resulting calculation result, D1, is defined as the fourth error.

[0106] Calculate the chamfer distance D2 between the true value point in the lane line true value map and the detection point in the lane line map, which is defined as the fifth error.

[0107] Perform weighted summation on the fourth error D1 and the fifth error D2 to obtain a weighted summation result, which is defined as the second evaluation result D. The calculation formula of the second evaluation result D is:

[0108] D=W1*D1+W2*D2

[0109] Wherein, W1 and W2 are weights, W1+W2=1, in this embodiment, W1=0.6, W2=0.4. In other embodiments, the values ​​of W1 and W2 can also be other values ​​that can be realized and are not specifically limited.

[0110] After completing the first and second evaluations of the lane line detection data, further weighted summation is performed on the first evaluation result d and the second evaluation result D to obtain the weighted summation result and the verification information score. The calculation formula of the verification information score is:

[0111] score=w_1*d+w_2*D

[0112] Where w_1 and w_2 are weights, w_1 + w_2 = 1. In this embodiment, w_1 = 0.35 and w_2 = 0.65. The greater weight of w_2 is because errors in the generated lane map affect lane markings. A smaller final verification score indicates more accurate lane markings, meaning more accurate lane detection data. Adjusting lane markings based on verification information can improve lane marking accuracy, enabling updated lane markings on known sections and automatic lane marking on unknown sections.

[0113] In one embodiment of the present invention, obtaining a fourth error of the lane line map relative to the lane line ground truth map includes:

[0114] Obtain each point in the lane line true value map, and obtain the two points in the lane line map closest to each point;

[0115] Obtaining a third distance from each point to a line connecting the corresponding two points, thereby obtaining a plurality of third distances;

[0116] An average value of a plurality of the third distances is obtained to be defined as the fourth error.

[0117] In this embodiment, the lane line truth line in the lane line truth map and the lane line detection line in the lane line map are projected into the same coordinate system. For each truth point in the lane line truth map, a search is performed near the truth point in the lane line truth line for two detection points on the lane line detection line closest to the truth point. A line is then connected between the two detected points, and the perpendicular distance from the truth point to the line connecting the two detection points is calculated. By calculating the perpendicular distance from each truth point in the lane line truth line to the line connecting the two closest detection points, multiple third distances are obtained. The average value D1 of these multiple third distances is calculated, and this average value D1 is defined as the fourth error.

[0118] In other embodiments, each detection point in the lane line map is obtained, and two truth points in the lane line truth line closest to the detection point in the lane line detection line are searched near the detection point. The two truth points obtained by the search are connected, and the vertical distance from the detection point to the line connecting the two nearest truth points is calculated. Multiple third distances can also be obtained.

[0119] The obtaining of a fifth error of the lane line map relative to the lane line true value map includes:

[0120] Obtain each point in the lane line true value map, and obtain the point in the lane line map closest to each point;

[0121] Obtaining a fourth distance from each point to a corresponding point, thereby obtaining a plurality of fourth distances;

[0122] A sum of a plurality of the fourth distances is obtained to be defined as the fifth error.

[0123] In this embodiment, each true value point in the lane line true value map is obtained, and the detection point with the smallest distance in the lane line map corresponding to the true value point is obtained. The minimum distance between the true value point and the detection point is obtained, which is the fourth distance. Multiple fourth distances are obtained, and then the fourth distance corresponding to each true value point is summed to obtain the summation result D2, which is defined as the fifth error. The calculation formula of the fifth error D2 is:

[0124]

[0125] Among them, S1 represents a point cloud composed of multiple true value points, x represents any point in the true value point cloud, S2 represents a point cloud composed of multiple detection points, and y represents any point in the detection point cloud.

[0126] An embodiment of the present invention further provides a method for generating a driving map, comprising the following steps:

[0127] Execute the lane marking process on the original map to obtain the lane marking result;

[0128] A verification process is performed on the lane marking result to generate the driving map; wherein the verification process includes the verification method for lane marking in the above embodiment.

[0129] In this embodiment, a crowdsourcing collection vehicle collects lidar point cloud data and camera image data. This data is then trained using a perception model to generate lane detection data. This lane detection data is then used to generate a lane map, implementing a lane labeling process and obtaining lane labeling results. It will be appreciated that the lane labeling results are then subjected to a verification process, including the lane labeling verification method described in the aforementioned embodiment, to obtain a verification result. The lane labeling process is then modified based on the verification result to generate a driving map. The lane lines in the driving map are in a three-dimensional semantic form.

[0130] See also Figure 2 , Figure 2 1 is a schematic structural diagram of an embodiment of an on-vehicle device of the present invention. The on-vehicle device 100 includes a processor 101 and a memory 102 coupled to the processor. The memory 102 is used to store a computer program, and the processor 101 is used to execute the computer program to implement the lane marking verification method or the driving map generation method in the above-mentioned embodiment.

[0131] See also Figure 3 , Figure 3 1 is a schematic structural diagram of an embodiment of a non-volatile computer-readable storage medium of the present invention. The non-volatile computer-readable storage medium 200 is used to store a computer program 201. When the computer program 201 is executed by the processor 101, it is used to implement the lane marking verification method or the driving map generation method in the above-mentioned embodiment.

[0132] The non-volatile computer-readable storage medium 200 can be a server, a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., which can store program codes.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules or units is merely a logical functional division. In actual implementation, other division methods may be used. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not implemented.

[0134] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0135] In addition, each functional unit in each embodiment of the present application may be integrated into a processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The above-mentioned integrated units may be implemented in the form of hardware or software functional units.

[0136] The above description is only an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A verification method for lane marking, characterized in that: include: Acquiring lane line detection data for lane line marking, and performing a first evaluation on the lane line detection data to obtain a first evaluation result; Obtaining a lane map generated based on the lane detection data, and performing a second evaluation on the lane map to obtain a second evaluation result; Obtaining verification information for lane marking based on the first evaluation result and the second evaluation result, wherein the verification information is used to represent the accuracy of the lane marking; The lane line detection data includes a first portion of data and a second portion of data, the first portion of data corresponds to a main image, and the second portion of data corresponds to the main image and at least one auxiliary image, and both the main image and the auxiliary image are about lane lines; The first evaluation of the lane line detection data to obtain a first evaluation result includes: Obtaining a first error of the first portion of data relative to lane line point cloud data, wherein the lane line point cloud data is used to represent the lane line; Obtaining a second error and a third error of the second portion of data relative to the lane line point cloud data; The first evaluation result is obtained according to the first error, the second error, and the third error.

2. The verification method according to claim 1, wherein: The first error is an average accuracy rate of lane line detection on the main image in the first part of data; The second error is the distance between a point in the lane line point cloud data and a line connecting two points of the lane line in the second portion of data; The third error is the chamfer distance between the lane line point in the second portion of data and the lane line point cloud data; The first evaluation result is a weighted sum of the first error, the second error, and the third error.

3. The verification method according to claim 2, wherein: The obtaining a second error of the second portion of data relative to the lane line point cloud data includes: Obtain each point in the lane line point cloud data, and obtain two points in the second part of data that are closest to each point; Obtaining a first distance from each point to a line connecting the corresponding two points, thereby obtaining a plurality of first distances; obtaining an average of a plurality of the first distances to define the average as the second error; The obtaining of a third error of the second portion of data relative to the lane line point cloud data includes: Obtain each point in the lane line point cloud data, and obtain the point in the second part of data that is closest to each point; Obtaining a second distance from each point to a corresponding point, thereby obtaining a plurality of second distances; A sum of a plurality of the second distances is obtained to be defined as the third error.

4. The verification method according to claim 1, wherein: The performing a second evaluation on the lane map to obtain a second evaluation result includes: Obtaining a fourth error of the lane map relative to a ground-truth lane map, wherein the ground-truth lane map is annotated based on lane point cloud data; Obtaining a fifth error of the lane line map relative to the lane line ground truth map; The second evaluation result is obtained according to the fourth error and the fifth error.

5. The verification method according to claim 4, characterized in that: The fourth error is the distance between a point in the lane line true value map and a line connecting two points of the lane line in the lane line map; The fifth error is a chamfer distance between a point on the lane line in the lane line map and a point on the lane line ground truth map; The second evaluation result is a weighted sum of the fourth error and the fifth error.

6. The verification method according to claim 5, characterized in that: The obtaining of a fourth error of the lane line map relative to the lane line true value map includes: Obtain each point in the lane line true value map, and obtain the two points in the lane line map closest to each point; Obtaining a third distance from each point to a line connecting the corresponding two points, thereby obtaining a plurality of third distances; obtaining an average value of a plurality of the third distances to define the average value as the fourth error; The obtaining of a fifth error of the lane line map relative to the lane line true value map includes: Obtain each point in the lane line true value map, and obtain the point in the lane line map closest to each point; Obtaining a fourth distance from each point to a corresponding point, thereby obtaining a plurality of fourth distances; A sum of a plurality of the fourth distances is obtained to be defined as the fifth error.

7. A method for generating a driving map, characterized in that: include: Execute the lane marking process on the original map to obtain the lane marking result; Performing a verification process on the lane marking result to generate the driving map; The verification process includes the verification method for lane marking according to any one of claims 1 to 6.

8. A vehicle-mounted device, characterized in that: The system comprises a processor and a memory coupled to the processor, the memory being used to store a computer program, and the processor being used to execute the computer program to implement the lane marking verification method according to any one of claims 1 to 6 or the driving map generation method according to claim 7.

9. A non-volatile computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, which, when executed by a processor, is used to implement the lane marking verification method according to any one of claims 1 to 6 or the driving map generation method according to claim 7.

Images