Lane line detection method and device

Abstract

The application provides a lane line detection method and device, the method comprising: obtaining a lane line image to be detected; inputting the lane line to be detected into a lane line detection model to obtain a lane line detection result output by the lane line detection model; wherein the lane line detection model is used for respectively performing point position probability prediction, point position detection and precision loss detection based on feature maps obtained by feature extraction on the lane line image to be detected to obtain the lane line detection result; and the lane line detection model is obtained by training based on training lane line images and lane line detection true values obtained by position encoding based on lane lines in the training lane line images. The lane line detection model is used for performing point position probability prediction, point position detection and precision loss detection on the lane line image to be detected, thereby improving the lane line detection precision in a bifurcated lane line scene, significantly helping to correctly determine a route and plan a path, and reducing incorrect driving routes such as missing an intersection and wrong turning.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a lane line detection method and apparatus. Background Technology

[0002] Lane markings are an indispensable component of traditional ADAS (Advanced Driver Assistance Systems) and autonomous driving, especially in applications such as lane departure warning (LDW) and lane keeping assist (LKA), where lane marking detection is crucial.

[0003] The current mainstream lane detection algorithms are mainly divided into the following four types: segmentation-based, anchor-based, row-wise, and poly-regression. All of them use neural networks to output lane line positions. Among them, the segmentation-based method outputs a lane line mask based on image segmentation, the anchor-based method obtains the offset of the lane line from the anchor line based on the anchor line, the row-wise method divides the image into several grids to obtain the center offset, and the poly-regression method directly regresses the lane line fitting parameters.

[0004] However, the aforementioned mainstream lane detection algorithms perform poorly in complex and varied road scenarios, such as forking lanes, including 'V', 'Y', 'V', and inverted 'V' lane patterns. They suffer from both false positives and false negatives, and often detect them as the same lane, which can easily lead to path planning errors and cause the wrong intersection to be missed, such as at highway ramps, forks, and intersections. Summary of the Invention

[0005] This invention provides a lane line detection method and apparatus to address the shortcomings of traditional lane line detection in complex road scenarios, improve lane line detection accuracy in bifurcated lane scenarios, significantly aid in decision-making and route planning to ensure correct routes, and reduce errors such as missing intersections and wrong turns.

[0006] This invention provides a lane line detection method, comprising: acquiring a lane line image to be detected; inputting the lane line image to be detected into a lane line detection model to obtain a lane line detection result output by the lane line detection model; wherein, the lane line detection model is used to perform point probability prediction, point detection, and accuracy loss detection based on the feature map extracted from the lane line image to be detected, respectively, to obtain the lane line detection result; the lane line detection model is trained based on a training lane line image and lane line detection ground values ​​obtained by position encoding of lane lines in the training lane line image.

[0007] According to a lane line detection method provided by the present invention, the lane line detection result includes the confidence level of each point constituting the lane line, the relative coordinates of the lane line points, and the accuracy loss result. The lane line detection model includes: a feature extraction layer, which extracts features based on the input lane line image to be detected to obtain a lane line feature map; a first lane line detection layer, which performs point probability prediction on the lane line feature map to obtain the confidence level of each point constituting the lane line; a second lane line detection layer, which performs point detection on the lane line feature map to obtain the relative coordinates of the lane line points; and a third lane line detection layer, which performs accuracy loss detection on the lane line feature map to obtain the accuracy loss result after floating-point to integer conversion.

[0008] According to a lane line detection method provided by the present invention, the relative coordinates of lane line points include start-point distance coordinates and end-point distance coordinates. The step of detecting points on the lane line feature map to obtain the relative coordinates of lane line points includes: based on the lane line feature map, performing point detection using a first offset channel to obtain start-point distance coordinates, where the start-point distance coordinates are the distance coordinates between the points constituting the lane line and the start point along the direction pointing towards the start point; and based on the lane line feature map, performing point detection using a second offset channel to obtain end-point distance coordinates, where the end-point distance coordinates are the distance coordinates between the points constituting the lane line and the end point along the direction pointing towards the end point.

[0009] According to a lane line detection method provided by the present invention, after obtaining the lane line detection result output by the lane line detection model, the method includes: filtering the relative coordinates of the lane line points based on a preset threshold to obtain all start-point distance coordinates and end-point distance coordinates less than the preset threshold; clustering the filtered start-point distance coordinates and end-point distance coordinates and obtaining the mean of each clustering result to obtain predicted start-point coordinates and predicted end-point coordinates; and obtaining the coordinate positions of all points and their respective lane line IDs based on the predicted start-point coordinates and predicted end-point coordinates, combined with the confidence and accuracy loss results of each point constituting the lane line.

[0010] According to a lane detection method provided by the present invention, after obtaining the lane detection result output by the lane detection model, the method further includes: optimizing the lane detection result based on nonmaximum suppression.

[0011] According to a lane detection method provided by the present invention, training the lane detection model includes: obtaining a training lane image and a lane detection ground truth value corresponding to the training lane image; using the training lane image as input data for training, using the lane detection ground truth value corresponding to the training lane image as a label, and training the network to be trained to obtain a lane detection model for generating lane detection results.

[0012] According to a lane line detection method provided by the present invention, obtaining a training lane line image and a lane line detection ground value corresponding to the training lane line image includes: obtaining a training lane line image; performing lane line sampling on the training lane line image to obtain sampling points that constitute lane lines; and encoding the sampling points to obtain lane line detection ground values.

[0013] The present invention also provides a lane line detection device, comprising: a data acquisition module for acquiring a lane line image to be detected; and a lane line detection module for inputting the lane line image to be detected into a lane line detection model to obtain a lane line detection result output by the lane line detection model; wherein, the lane line detection model is used to perform point probability prediction, point detection, and accuracy loss detection based on the feature map extracted from the lane line image to be detected, respectively, to obtain the lane line detection result; the lane line detection model is trained based on a training lane line image and lane line detection ground values ​​obtained by position encoding of lane lines in the training lane line image.

[0014] The present invention also provides an electronic device, including 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 steps of any of the lane line detection methods described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the lane line detection methods described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the lane line detection methods described above.

[0017] The lane line detection method and apparatus provided by this invention improve the lane line detection accuracy in bifurcated lane line scenarios by using a lane line detection model to perform point probability prediction, point detection, and accuracy loss detection on the lane line image to be detected. This significantly helps in decision-making and path planning to ensure the correct route and reduces errors such as missing intersections and wrong turns. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a schematic flowchart of the lane line detection method provided by the present invention;

[0020] Figure 2 This is a schematic diagram of the framework of the lane line detection model provided by the present invention;

[0021] Figure 3 This is a schematic diagram of lane line coding provided by the present invention;

[0022] Figure 4 This is a schematic diagram of the lane line detection device provided by the present invention;

[0023] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] Figure 1 A schematic flowchart of a lane line detection method according to the present invention is shown, the method comprising:

[0026] S11, acquire the image of the lane line to be detected;

[0027] S12, the lane line image to be detected is input into the lane line detection model to obtain the lane line detection result output by the lane line detection model; wherein, the lane line detection model is used to perform point probability prediction, point detection and accuracy loss detection based on the feature map extracted from the lane line image to be detected to obtain the lane line detection result; the lane line detection model is trained based on the training lane line image and the lane line detection ground value obtained by position encoding of the lane lines in the training lane line image.

[0028] It should be noted that S1N in this manual does not represent the order of lane detection methods. The following details will explain this in conjunction with... Figure 2-3 The lane line detection method of the present invention is described.

[0029] Step S11: Obtain the image of the lane line to be detected.

[0030] In this embodiment, obtaining the lane line image to be detected includes: obtaining an optical image of the road segment to be detected based on a camera device to obtain the lane line image to be detected; or obtaining a point cloud image of the road segment to be detected using a lidar to obtain the lane line image to be detected.

[0031] Step S12: Input the lane line image to be detected into the lane line detection model to obtain the lane line detection result output by the lane line detection model; wherein, the lane line detection model is used to perform point probability prediction, point detection and accuracy loss detection based on the feature map extracted from the lane line image to be detected, and obtain the lane line detection result; the lane line detection model is trained based on the training lane line image and the lane line detection ground value obtained by position encoding of the lane lines in the training lane line image.

[0032] In this embodiment, the lane line detection result includes the confidence level of each point that makes up the lane line, the relative coordinates of the lane line points, and the accuracy loss result. The lane line detection model includes: a feature extraction layer, which extracts features based on the input lane line image to be detected to obtain a lane line feature map; and a lane line detection layer, which performs point probability prediction, point detection, and accuracy loss detection on the lane line feature map.

[0033] It should be added that the lane line detection model also includes a neck network layer, which is used to input the lane line feature map output by the feature extraction layer into the lane line detection layer, so that the lane line detection layer can perform point probability prediction, point detection and accuracy loss detection based on the lane line feature map.

[0034] In addition, the feature extraction layer and the neck network layer can use general convolutional networks such as backbone and neck, or convolutional networks customized based on the needs of lane line feature extraction, without further restrictions.

[0035] In this embodiment, reference Figure 2 The lane detection layer includes: a first lane detection layer, which performs point probability prediction on the lane feature map to obtain the confidence level of each point that makes up the lane line; a second lane detection layer, which performs in-band detection on the lane feature map to obtain the relative coordinates of the lane line points; and a third lane detection layer, which performs precision loss detection on the lane feature map to obtain the precision loss result of floating-point integer conversion.

[0036] Furthermore, the relative coordinates of lane line points include the starting point distance coordinates and the ending point distance coordinates. Point detection is performed on the lane line feature map to obtain the relative coordinates of lane line points, including: based on the lane line feature map, point detection is performed using the first offset channel to obtain the starting point distance coordinates, which are the distance coordinates between the points forming the lane line and the starting point along the direction pointing to the starting point; based on the lane line feature map, point detection is performed using the second offset channel to obtain the ending point distance coordinates, which are the distance coordinates between the points forming the lane line and the ending point along the direction pointing to the ending point.

[0037] In one optional embodiment, training a lane detection model includes: obtaining a training lane image and the lane detection ground truth corresponding to the training lane image; using the training lane image as input data for training, and using the lane detection ground truth corresponding to the training lane image as labels, training the network to be trained to obtain a lane detection model for generating lane detection results.

[0038] Specifically, obtaining the training lane line image and the corresponding lane line detection ground value includes: obtaining the training lane line image; sampling the training lane line image to obtain the sampling points that make up the lane lines; and encoding the sampling points to obtain the lane line detection ground value.

[0039] It should be noted that encoding the sampling points includes: encoding the start and end points separately; and encoding the remaining sampling points sequentially along the direction from the start to the end. For example, refer to... Figure 3 Taking a single lane as an example, where P(x) i ,y i P(x) represents a sampling point on the lane line. s ,y s P(x) represents the starting point of the lane line. e ,y e The sign () indicates the end point of a lane line. A single lane line can be represented as follows:

[0040]

[0041] Among them, P ΔS P ΔeThis represents a vector point pointing to both the start and end points.

[0042] In addition, the network to be trained can be an existing network built into the training device. This existing network typically includes a network structure, or it can be other networks specified by the user, such as various neural networks. The network to be trained typically includes a feature extraction layer, a neck network layer, and a lane detection layer. According to a preset iteration rule, the encoded training lane images are input into the model to be trained for training, resulting in a trained lane detection model.

[0043] In an optional embodiment, after obtaining the lane detection results output by the lane detection model, the process includes: filtering the relative coordinates of lane line points based on a preset threshold to obtain all start-point and end-point distance coordinates less than the preset threshold; clustering the filtered start-point and end-point distance coordinates and obtaining the mean of each clustering result to obtain predicted start-point and predicted end-point coordinates; and obtaining the coordinate positions of all points and their respective lane line IDs based on the predicted start-point and predicted end-point coordinates, combined with the confidence and accuracy loss results of each point constituting the lane line.

[0044] In one optional embodiment, after obtaining the lane detection results output by the lane detection model, the method further includes: optimizing the lane detection results based on nonmaximum suppression. It should be noted that further optimization of the lane detection results improves their accuracy, particularly in bifurcation lane scenarios, significantly aiding in decision-making and path planning to ensure correct routes and reducing errors such as missing intersections or making incorrect turns.

[0045] Furthermore, the lane detection results include the confidence scores of each point that makes up the lane line, the relative coordinates of the lane line points, and the accuracy loss results. Based on non-maximum suppression, the lane detection results are optimized, including: sorting the points that make up the lane line based on their confidence scores; selecting the point corresponding to the highest confidence score in descending order of confidence scores, and obtaining its intersection-over-union (IoU) ratio with the remaining points based on its relative coordinates; retaining the relative coordinates of the point corresponding to the highest confidence score and removing the relative coordinates of the remaining points based on the IoU ratio being greater than a preset threshold; reselecting the point corresponding to the highest confidence score, obtaining its IoU ratio with the remaining points again based on its relative coordinates, and retaining the relative coordinates of the current point corresponding to the highest confidence score and removing the relative coordinates of the remaining points based on the newly obtained IoU ratio being greater than a preset threshold, until all retained relative coordinates of the lane line points are obtained.

[0046] In summary, the embodiments of the present invention improve the accuracy of lane line detection in bifurcated lane line scenarios by using a lane line detection model to perform point probability prediction, point detection, and accuracy loss detection on the lane line image to be detected. This significantly helps in decision-making and path planning to ensure the correct route and reduces errors such as missing intersections or making wrong turns.

[0047] The lane line detection device provided by the present invention is described below. The lane line detection device described below can be referred to in correspondence with the lane line detection method described above.

[0048] Figure 4 A schematic diagram of a lane line detection device according to the present invention is shown. The device includes:

[0049] Data acquisition module 41 acquires the image of the lane line to be detected;

[0050] The lane line detection module 42 inputs the lane line image to be detected into the lane line detection model to obtain the lane line detection result output by the lane line detection model. The lane line detection model is used to perform point probability prediction, point detection, and accuracy loss detection based on the feature map extracted from the lane line image to be detected, and obtain the lane line detection result. The lane line detection model is trained based on the training lane line image and the lane line detection ground value obtained by position encoding of the lane lines in the training lane line image.

[0051] In this embodiment, the data acquisition module 41 includes: a first image acquisition unit, which acquires an optical image of the road segment to be detected based on a camera device to obtain an image of the lane line to be detected; or, a second image acquisition unit, which acquires a point cloud image of the road segment to be detected using a lidar to obtain an image of the lane line to be detected.

[0052] The lane line detection module 42 includes: a data input unit for inputting the lane line image to be detected into the lane line detection model; a detection unit for performing lane line detection on the input lane line image based on the lane line detection model to obtain the lane line detection result; and a data output unit for outputting the lane line detection result obtained by the lane line detection model.

[0053] Furthermore, the lane line detection results include the confidence scores of each point that makes up the lane line, the relative coordinates of the lane line points, and the accuracy loss results. The detection unit includes: a feature extraction subunit, which extracts features based on the input lane line image to be detected to obtain a lane line feature map; and a lane line detection subunit, which performs point probability prediction, point detection, and accuracy loss detection on the lane line feature map.

[0054] It should be added that the detection unit also includes a neck network subunit, which is used to input the lane line feature map output by the feature extraction subunit into the lane line detection subunit, so that the lane line detection subunit can perform point probability prediction, point detection and accuracy loss detection based on the lane line feature map.

[0055] The lane line detection subunit includes: a first lane line detection layer, which performs point probability prediction on the lane line feature map to obtain the confidence level of each point that makes up the lane line; a second lane line detection layer, which performs point detection on the lane line feature map to obtain the relative coordinates of the lane line points; and a third lane line detection layer, which performs precision loss detection on the lane line feature map to obtain the precision loss result of floating-point integer conversion.

[0056] Furthermore, the relative coordinates of lane line points include the starting point distance coordinates and the ending point distance coordinates. The second lane line detection layer includes: based on the lane line feature map, using the first offset channel to perform point detection to obtain the starting point distance coordinates, which are the distance coordinates between the points forming the lane line and the starting point along the direction pointing to the starting point; and based on the lane line feature map, using the second offset channel to perform point detection to obtain the ending point distance coordinates, which are the distance coordinates between the points forming the lane line and the ending point along the direction pointing to the ending point.

[0057] In an optional embodiment, the apparatus further includes a training module for training the lane line detection model before inputting the lane line image to be detected into the lane line detection model. Specifically, the training module includes: a training data acquisition unit for acquiring training lane line images and lane line detection ground values ​​corresponding to the training lane line images; and a training unit for using the training lane line images as input data for training and the lane line detection ground values ​​corresponding to the training lane line images as labels to train the network to be trained, thereby obtaining a lane line detection model for generating lane line detection results.

[0058] In addition, the training data acquisition unit includes: an image acquisition subunit for acquiring training lane line images; a sampling subunit for sampling lane lines in the training lane line images to obtain sampling points that make up the lane lines; and an encoding subunit for encoding the sampling points to obtain lane line detection ground truth.

[0059] It should be noted that the encoding unit includes: a first encoding subunit, which encodes the start point and the end point respectively; and a second encoding subunit, which encodes the remaining sampling points sequentially along the direction from the start point to the end point.

[0060] In an optional embodiment, the device further includes: a filtering module, which filters the relative coordinates of lane line points based on a preset threshold to obtain all start-point distance coordinates and end-point distance coordinates that are less than the preset threshold; a coordinate prediction module, which clusters the filtered start-point distance coordinates and end-point distance coordinates and obtains the mean of each clustering result to obtain predicted start-point coordinates and predicted end-point coordinates; and a position determination module, which, based on the predicted start-point coordinates and predicted end-point coordinates, and combined with the confidence and accuracy loss results of each point constituting the lane line, obtains the coordinate positions of all points and the lane line ID to which they belong.

[0061] In an optional embodiment, the apparatus further includes a result optimization module, which optimizes the lane line detection results based on nonmaximum suppression after obtaining the lane line detection results output by the lane line detection model. In this example, the lane line detection results include the confidence scores of each point constituting the lane line, the relative coordinates of the lane line points, and the accuracy loss results. Furthermore, the result optimization module includes: a sorting unit, which sorts the points that make up the lane lines based on confidence levels; an intersection-over-union ratio (IoU) acquisition unit, which, based on the sorting results, sequentially selects the point corresponding to the highest confidence level from largest to smallest, and obtains its IoU with the remaining points based on the relative coordinates of the corresponding lane line point; an optimization unit, which, based on the IoU being greater than a preset threshold, retains the relative coordinates of the lane line point corresponding to the highest confidence level and removes the relative coordinates of the corresponding remaining points; and a re-selection unit, based on the IoU being greater than a preset threshold, selects the point corresponding to the highest confidence level, and re-obtains its IoU with the remaining points, and re-obtains the relative coordinates of the lane line point corresponding to the highest confidence level, and removes the relative coordinates of the corresponding remaining points, until all retained relative coordinates of the lane line points are obtained.

[0062] In summary, the embodiments of the present invention improve the accuracy of lane line detection in bifurcated lane line scenarios by using a lane line detection model to perform point probability prediction, point detection, and accuracy loss detection on the lane line image to be detected. This significantly helps in decision-making and path planning to ensure the correct route and reduces errors such as missing intersections or making wrong turns.

[0063] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include: a processor 51, a communication interface 52, a memory 53, and a communication bus 54. The processor 51, communication interface 52, and memory 53 communicate with each other via the communication bus 54. The processor 51 can call logical instructions in the memory 53 to execute a lane detection method. This method includes: acquiring an image of a lane line to be detected; inputting the lane line to be detected into a lane detection model to obtain a lane detection result output by the lane detection model; wherein the lane detection model is used to perform point probability prediction, point detection, and accuracy loss detection based on the feature map extracted from the image of the lane line to be detected, respectively, to obtain the lane detection result; the lane detection model is trained based on a training lane line image and lane detection ground values ​​obtained by position encoding of lane lines in the training lane line image.

[0064] Furthermore, the logical instructions in the aforementioned memory 53 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0065] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the lane line detection method provided by the above methods. The method includes: acquiring a lane line image to be detected; inputting the lane line to be detected into a lane line detection model to obtain a lane line detection result output by the lane line detection model; wherein, the lane line detection model is used to perform point probability prediction, point detection, and accuracy loss detection based on the feature map extracted from the lane line image to be detected, respectively, to obtain the lane line detection result; the lane line detection model is trained based on a training lane line image and a lane line detection ground value obtained by position encoding of the lane lines in the training lane line image.

[0066] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the lane detection method provided by the above methods. The method includes: acquiring a lane line image to be detected; inputting the lane line to be detected into a lane detection model to obtain a lane detection result output by the lane detection model; wherein the lane detection model is used to perform point probability prediction, point detection, and accuracy loss detection based on the feature map extracted from the lane line image to be detected, respectively, to obtain the lane detection result; the lane detection model is trained based on a training lane line image and lane detection ground values ​​obtained by position encoding of lane lines in the training lane line image.

[0067] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The 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 the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0068] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A lane line detection method, characterized in that, include: Acquire the image of the lane line to be detected; The lane line image to be detected is input into the lane line detection model to obtain the lane line detection result output by the lane line detection model; The lane detection model is used to perform point probability prediction, point detection, and accuracy loss detection based on the feature map extracted from the lane line image to be detected, to obtain the lane line detection result; the lane line detection model is trained based on the training lane line image and the lane line detection ground value obtained by position encoding of the lane lines in the training lane line image. The lane detection results include the confidence scores of each point constituting the lane line, the relative coordinates of the lane line points, and the accuracy loss results. The lane detection model includes: The feature extraction layer extracts features from the input lane line image to be detected, resulting in a lane line feature map. The first lane detection layer performs point probability prediction on the lane feature map to obtain the confidence level of each point that makes up the lane line. The second lane line detection layer performs point detection on the lane line feature map to obtain the relative coordinates of the lane line points. The third lane detection layer performs precision loss detection on the lane feature map to obtain the precision loss result of floating-point integer conversion; The relative coordinates of the lane line points include the starting point distance coordinates and the ending point distance coordinates. The step of performing point detection on the lane line feature map to obtain the relative coordinates of the lane line points includes: Based on the lane line feature map, point detection is performed using the first offset channel to obtain the starting point distance coordinates. The starting point distance coordinates are the distance coordinates between the points that make up the lane line and the starting point along the direction pointing to the starting point. Based on the lane line feature map, point detection is performed using the second offset channel to obtain the first endpoint distance coordinates. The endpoint distance coordinates are the distance coordinates between the points that make up the lane line and the endpoint along the direction pointing to the endpoint. After obtaining the lane detection results output by the lane detection model, the process includes: The relative coordinates of the lane line points are filtered based on a preset threshold to obtain all start-point and end-point distance coordinates that are less than the preset threshold. Cluster the selected start-point and end-point distance coordinates and obtain the mean of each cluster to get the predicted start-point and end-point coordinates. Based on the predicted starting point coordinates and the predicted ending point coordinates, and combined with the confidence and accuracy loss results of each point that makes up the lane line, the coordinate positions of all points and their respective lane line IDs are obtained.

2. The lane line detection method according to claim 1, characterized in that, After obtaining the lane detection results output by the lane detection model, the process also includes: The lane detection results are optimized based on nonmaximum suppression.

3. The lane line detection method according to claim 1, characterized in that, Training the lane detection model includes: Obtain training lane line images and the lane line detection ground values ​​corresponding to the training lane line images; The training lane line images are used as input data for training, and the ground truth values ​​of lane line detection corresponding to the training lane line images are used as labels to train the network to obtain a lane line detection model for generating lane line detection results.

4. The lane line detection method according to claim 3, characterized in that, The step of obtaining the training lane line image and the corresponding ground truth lane line detection value includes: Obtain training lane line images; The training lane line image is sampled to obtain the sampling points that make up the lane lines; The sampling points are encoded to obtain the lane line detection ground truth.

5. A lane line detection device, characterized in that, include: The data acquisition module acquires images of the lane lines to be detected. The lane line detection module inputs the lane line image to be detected into the lane line detection model to obtain the lane line detection result output by the lane line detection model; The lane detection model is used to perform point probability prediction, point detection, and accuracy loss detection based on the feature map extracted from the lane line image to be detected, to obtain the lane line detection result; the lane line detection model is trained based on the training lane line image and the lane line detection ground value obtained by position encoding of the lane lines in the training lane line image. The lane detection results include the confidence scores of each point constituting the lane line, the relative coordinates of the lane line points, and the accuracy loss results. The lane detection model includes: The feature extraction layer extracts features from the input lane line image to be detected, resulting in a lane line feature map. The first lane detection layer performs point probability prediction on the lane feature map to obtain the confidence level of each point that makes up the lane line. The second lane line detection layer performs point detection on the lane line feature map to obtain the relative coordinates of the lane line points. The third lane detection layer performs precision loss detection on the lane feature map to obtain the precision loss result of floating-point integer conversion; The relative coordinates of the lane line points include the starting point distance coordinates and the ending point distance coordinates. The second lane line detection layer is also used for: Based on the lane line feature map, point detection is performed using the first offset channel to obtain the starting point distance coordinates. The starting point distance coordinates are the distance coordinates between the points that make up the lane line and the starting point along the direction pointing to the starting point. Based on the lane line feature map, point detection is performed using the second offset channel to obtain the first endpoint distance coordinates. The endpoint distance coordinates are the distance coordinates between the points that make up the lane line and the endpoint along the direction pointing to the endpoint. The device further includes: After obtaining the lane line detection results output by the lane line detection model, the filtering module filters the relative coordinates of the lane line points based on a preset threshold to obtain all start-point and end-point distance coordinates that are less than the preset threshold. The coordinate prediction module clusters the selected start-point and end-point distance coordinates and obtains the mean of each cluster result to obtain the predicted start-point and end-point coordinates. The location determination module obtains the coordinates of all points and their respective lane IDs based on the predicted starting point coordinates and the predicted ending point coordinates, combined with the confidence and accuracy loss results of each point that makes up the lane line.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the lane line detection method as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the lane line detection method as described in any one of claims 1 to 4.

Images