Lane line prediction model training method and device and computer equipment

Abstract

The application relates to a lane line prediction model training method and device and a computer device. The method comprises the following steps: obtaining a first lane line image and a second lane line image, the second lane line image being a previous frame image of the first lane line image; extracting an initial feature map of the first lane line image, the first lane line image and the second lane line image both comprising labeled lane lines; performing feature fusion on the initial feature map of the first lane line image and the labeled lane lines in the second lane line image to obtain a fused feature map; performing image generation based on the fused feature map to obtain a lane line prediction image, the lane line prediction image comprising a predicted lane line; determining the difference between the labeled lane line and the predicted lane line, performing model training based on the difference, and obtaining a lane line prediction model. The method can improve the training efficiency and effect and save the training cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a lane line prediction model training method and device, computer equipment, computer readable storage medium and computer program product. BACKGROUND

[0002] Lane lines are lines marked on the surface of the road, used to guide drivers to maintain on the correct driving route. For autonomous vehicles, lane line prediction helps vehicles to identify road boundaries and lane positions, thus realizing functions such as automatic steering and path planning. This not only improves the comfort of driving, but also enhances the safety of driving.

[0003] However, the current lane line prediction model usually needs to manually label the true value (i.e. the real lane line position) of each sequence, and then feature extraction and fusion can be performed. This method not only consumes time and effort, but also further increases the workload because lane lines are static and the same information may appear repeatedly between multiple frames. In addition, the existing methods have deficiencies in handling inter-frame motion relationships, and do not fully utilize the inter-frame correlation information to improve model performance. SUMMARY

[0004] Therefore, it is necessary to provide a lane line prediction model training method, device, computer equipment, computer readable storage medium and computer program product which can improve training efficiency and effect and save training cost.

[0005] In a first aspect, the present application provides a lane line prediction model training method, comprising:

[0006] obtaining a first lane line image and a second lane line image, the second lane line image being a previous frame image of the first lane line image;

[0007] extracting an initial feature map from the first lane line image, the first lane line image and the second lane line image both including labeled lane lines;

[0008] performing feature fusion on the initial feature map of the first lane line image and the labeled lane lines in the second lane line image to obtain a fused feature map;

[0009] generating an image based on the fused feature map to obtain a lane line prediction image, the lane line prediction image including a predicted lane line;

[0010] determining the difference between the labeled lane line and the predicted lane line, and performing model training based on the difference to obtain a lane line prediction model.

[0011] In one of the embodiments, obtaining the first lane line image and the second lane line image comprises:

[0012] obtaining an initial frame lane line image comprising a labeled lane line; determining a homography transformation matrix based on the labeled content of the labeled lane line; transforming the labeled lane line in the initial frame lane line image through the homography transformation matrix to obtain a lane line sequence image simulating lane line transformation caused by vehicle movement, the lane line sequence image comprising a plurality of continuous frame lane line images after the initial frame; selecting the first lane line image and the second lane line image from the plurality of continuous frame lane line images.

[0013] In one of the embodiments, the labeled lane line comprises a labeled angle and a plurality of labeled points; and the predicted lane line comprises a predicted angle and a plurality of predicted points.

[0014] The determination of the difference between the labeled lane line and the predicted lane line comprises:

[0015] determining the position difference between the labeled lane line and the predicted lane line based on the positions of the plurality of labeled points in the first lane line image and the positions of the plurality of predicted points in the lane line prediction image; and determining the angle difference between the labeled lane line and the predicted lane line based on the labeled angle and the predicted angle.

[0016] In one of the embodiments, the determination of the position difference between the labeled lane line and the predicted lane line based on the positions of the plurality of labeled points in the first lane line image and the positions of the plurality of predicted points in the lane line prediction image comprises:

[0017] for each labeled point in the first lane line image, searching for a target predicted point matching the labeled point from the plurality of predicted points in the lane line prediction image; determining the horizontal distance and the vertical distance between the labeled point and the target predicted point in the horizontal direction and the vertical direction; determining the Euclidean distance between the labeled point and the target predicted point based on the horizontal distance and the vertical distance; and taking the Euclidean distance between the labeled point and the target predicted point as the position difference between the labeled lane line and the predicted lane line.

[0018] In one of the embodiments, the model training based on the difference to obtain the lane line prediction model comprises:

[0019] determine a target annotation point from the plurality of annotation points based on a Euclidean distance between the target annotation point and the target prediction point and a preset distance threshold; and adjust parameters of a model based on a number of the target annotation points until the number of the target annotation points is less than a preset annotation point value, to obtain the lane line prediction model.

[0020] In one of the embodiments, the initial feature map of the first lane line image and the annotated lane line in the second lane line image are fused to obtain a fused feature map, including:

[0021] performing convolution operation on the initial feature map of the first lane line image by using a first convolution kernel to obtain a first feature map; performing convolution operation on the annotated lane line in the second lane line image by using a second convolution kernel to obtain a second feature map; performing linear combination on the first feature map and the second feature map to obtain a combined feature map; compressing feature values of the combined feature map by using a prediction activation function to obtain an output feature map, wherein the feature values in the output feature map are used to represent information retention ratios at each position of the annotated lane line in the second lane line image; and generating a fused feature map based on the output feature map, the initial feature map and the annotated lane line in the second lane line image.

[0022] In a second aspect, the present application further provides a lane line prediction model training device, including:

[0023] an obtaining module configured to obtain a first lane line image and a second lane line image, wherein the second lane line image is a previous frame image of the first lane line image;

[0024] an extracting module configured to extract an initial feature map from the first lane line image, wherein the first lane line image and the second lane line image both include an annotated lane line;

[0025] a fusing module configured to fuse the initial feature map of the first lane line image and the annotated lane line in the second lane line image to obtain a fused feature map;

[0026] a generating module configured to generate an image based on the fused feature map to obtain a lane line prediction image, wherein the lane line prediction image includes a predicted lane line;

[0027] a training module configured to determine a difference between the annotated lane line and the predicted lane line, and train a model based on the difference to obtain a lane line prediction model.

[0028] In a third aspect, the present application further provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0029] Obtain a first lane line image and a second lane line image, wherein the second lane line image is the previous frame of the first lane line image;

[0030] An initial feature map is extracted from the first lane line image, and both the first lane line image and the second lane line image include labeled lane lines.

[0031] The initial feature map of the first lane line image and the labeled lane lines in the second lane line image are fused to obtain a fused feature map;

[0032] Based on the fused feature map, an image is generated to obtain a lane line prediction image, which includes predicted lane lines.

[0033] The difference between the labeled lane lines and the predicted lane lines is determined, and the model is trained based on the difference to obtain the lane line prediction model.

[0034] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0035] Obtain a first lane line image and a second lane line image, wherein the second lane line image is the previous frame of the first lane line image;

[0036] An initial feature map is extracted from the first lane line image, and both the first lane line image and the second lane line image include labeled lane lines.

[0037] The initial feature map of the first lane line image and the labeled lane lines in the second lane line image are fused to obtain a fused feature map;

[0038] Based on the fused feature map, an image is generated to obtain a lane line prediction image, which includes predicted lane lines.

[0039] The difference between the labeled lane lines and the predicted lane lines is determined, and the model is trained based on the difference to obtain the lane line prediction model.

[0040] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0041] Obtain a first lane line image and a second lane line image, wherein the second lane line image is the previous frame of the first lane line image;

[0042] An initial feature map is extracted from the first lane line image, and both the first lane line image and the second lane line image include labeled lane lines.

[0043] The initial feature map of the first lane line image and the labeled lane lines in the second lane line image are fused to obtain a fused feature map;

[0044] Based on the fused feature map, an image is generated to obtain a lane line prediction image, which includes predicted lane lines.

[0045] The difference between the labeled lane lines and the predicted lane lines is determined, and the model is trained based on the difference to obtain the lane line prediction model.

[0046] The aforementioned lane prediction model training method, apparatus, computer equipment, computer-readable storage medium, and computer program product obtain a first lane line image and a second lane line image, where the second lane line image is the preceding frame of the first lane line image. By acquiring image data from consecutive frames (i.e., the previous frame and the current frame), a foundation is provided for subsequent feature extraction and fusion. Utilizing information in the temporal dimension, it can better capture changes during vehicle movement, thereby improving the accuracy and robustness of lane line detection.

[0047] An initial feature map is extracted from the first lane line image. Both the first and second lane line images include the labeled lane lines. By extracting features from a single frame, key lane line features can be identified, forming the basis for subsequent feature fusion and helping to more accurately understand the position of the lane lines in three-dimensional space.

[0048] The initial feature map of the first lane line image and the labeled lane lines in the second lane line image are fused to obtain a fused feature map. By combining the features of the two frames, the model not only considers information within a single frame but also incorporates changes over time. This approach effectively overcomes potential occlusion or interference issues that may occur during single-frame analysis and fully utilizes inter-frame correlation information to improve model performance.

[0049] Image generation is performed based on fused feature maps to obtain lane line prediction images, which include predicted lane lines. Generating prediction images using the fused feature maps can simulate the real lane line distribution to a certain extent. This method is more flexible and efficient than traditional rule-based post-processing algorithms, and maintains high detection accuracy even in complex environments.

[0050] The differences between labeled lane lines and predicted lane lines are identified, and a model is trained based on these differences to obtain a lane line prediction model. By comparing the predicted results with the actual ground truth and adjusting the model parameters accordingly, the model's performance can be continuously optimized. Furthermore, this approach avoids the time-consuming and laborious manual labeling process of traditional methods, significantly reducing costs and improving efficiency. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a diagram illustrating the application environment of a lane prediction model training method in one embodiment.

[0053] Figure 2 This is a flowchart illustrating a lane prediction model training method in one embodiment;

[0054] Figure 3 This is a planar schematic diagram of the initial frame lane line image in one embodiment;

[0055] Figure 4 This is an architecture diagram of the lane line prediction model training in one embodiment;

[0056] Figure 5 This is a structural block diagram of a lane line prediction model training device in one embodiment;

[0057] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] The lane prediction model training method provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 through a model. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed in the cloud or on another model server. Terminal 102 generates a lane prediction model training request and sends it to server 104, enabling server 104 to determine the difference between the labeled lane lines and the predicted lane lines, and train the model based on the difference to obtain the lane prediction model. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0060] In one exemplary embodiment, such as Figure 2 As shown, a lane line prediction model training method is provided, which is then applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 202 to 208. Wherein:

[0061] Step 202: Obtain the first lane line image and the second lane line image, wherein the second lane line image is the previous frame of the first lane line image.

[0062] The first lane line image refers to the currently processed image frame; in this application, it refers to the image frame used to extract the initial feature map. This image frame contains road information, including but not limited to the position and shape of the lane lines.

[0063] The second lane line image refers to the frame that immediately precedes the first lane line image, i.e., the previous frame. This frame also contains road information and has already been labeled with lane lines.

[0064] Specifically, firstly, single-frame images are manually annotated, and complete labeled lane lines are obtained by connecting the points. Then, continuous homography transformations are performed on the single-frame images to simulate the changing perspective effects of vehicle movement at different time points, thereby generating a series of continuous time-series images. This simulated time-series image includes lane line images from multiple consecutive frames. Understandably, the lane line images from multiple consecutive frames in the time-series image also include the labeled lane lines.

[0065] Then, from multiple consecutive frames of lane line images in the time series image, the next frame out of two consecutive frames is selected as the "first lane line image," representing the latest road scene to be analyzed. This image contains the latest road information and possible lane line markings. The frame preceding the above two frames is defined as the "second lane line image." This image represents the road conditions at the time point immediately preceding the "first lane line image," and also contains the corresponding lane line markings.

[0066] Step 204: Extract the initial feature map from the first lane line image. Both the first lane line image and the second lane line image include the labeled lane lines.

[0067] The initial feature map refers to the feature representation extracted from the first lane line image (current frame).

[0068] Annotated lane lines refer to lane lines that have been manually or automatically labeled in the second lane line image (previous frame). The annotations in the annotated lane lines include, but are not limited to, the start and end points, angles, and distances between key points. Understandably, these annotations are used to fuse with the feature map extracted from the current frame (first lane line image) to leverage information from historical frames and enhance the lane line detection accuracy of the current frame. For example, during feature fusion, the displacement differences of lane lines between adjacent frames are considered as part of the supervisory signal to help optimize the model training process.

[0069] Specifically, firstly, lane line images from multiple consecutive frames in the time-series image are fed into the feature extraction unit for feature extraction, resulting in an initial feature map for each lane line image. Then, all initial feature maps for each lane line image are sent to a data pool for storage. Next, the initial feature map corresponding to the first lane line image is retrieved from this data pool. Understandably, the data pool acts as an intermediate storage area, holding the feature information extracted from images at various time points. The feature extraction unit processes the input lane line images using a deep learning model, identifying and extracting key features that help identify lane lines. These features can be edges, textures, shapes, etc., which together constitute a description of the lane line's position and shape.

[0070] Step 206: The initial feature map of the first lane line image and the labeled lane lines in the second lane line image are fused to obtain a fused feature map.

[0071] Here, the fused feature map refers to a new feature representation generated by combining the initial feature map of the current frame (the first lane line image) with the labeled lane line information from the previous frame (the second lane line image). Understandably, this feature fusion process aims to leverage information in the temporal dimension to enhance the accuracy and robustness of lane line detection.

[0072] Specifically, the latest feature extraction result (i.e., the initial feature map of the first lane line image) and the detection result of the previous frame (i.e., the labeled lane lines in the second lane line image) are obtained from the data pool and fed into the convolutionally gated recurrent unit for feature fusion. Understandably, the convolutionally gated recurrent unit is a recurrent neural network structure specifically designed for processing sequential data (time-series images in this application). During processing, the convolutionally gated recurrent unit combines the features of the current frame with the detection result of the previous frame to perform feature fusion, obtaining a new fused feature map.

[0073] Step 208: Generate an image based on the fused feature map to obtain a lane line prediction image, which includes predicted lane lines.

[0074] The lane prediction image refers to an image generated based on a fused feature map. In this lane prediction image, the lane prediction model predicts the lane positions in the current frame based on learned patterns and rules. In other words, it is a new image inferred by the lane prediction model based on input data (i.e., the fused feature map obtained by fusing the initial feature map of the first lane image with the labeled lane information in the second lane image). This image aims to reflect the lane positions in the actual scene as accurately as possible.

[0075] Predicted lane lines refer to the location and shape of lane lines identified in a lane prediction image. It is the result of the lane prediction model analyzing input data and represents the model's best guess at the location of lane lines in the current frame. Predicted lane lines typically appear as lines or curves on the lane prediction image, indicating the path that vehicles should follow.

[0076] Specifically, the fused feature map obtained from the above steps is used as input and fed into a pre-trained deep learning model, namely the lane prediction model. This lane prediction model has been sufficiently trained and is able to predict the position and shape of the lane lines based on the input feature map.

[0077] Based on the fused feature map, a new image containing predicted lane lines, i.e., a lane line prediction image, is generated using a lane line prediction model. In this process, the lane line prediction model analyzes the information in the fused feature map and marks the parts in the image that it believes are most likely lane lines.

[0078] In the generated lane prediction image, the predicted lane lines appear as lines or curves, representing the lane line positions inferred by the lane prediction model. These predicted lane lines reflect the lane prediction model's best estimate of the lane line positions in the current frame.

[0079] Step 210: Determine the difference between the labeled lane lines and the predicted lane lines, and train the model based on the difference to obtain the lane line prediction model.

[0080] Here, the difference refers to the gap between the labeled lane lines (i.e., the actual or expected lane line locations, determined based on manual annotation or other precise methods) and the lane lines predicted by the model (i.e., the estimated lane line locations obtained after processing the image using the current model). Optionally, this gap is quantified by a predefined loss function, which measures the error between the model output and the actual target.

[0081] A lane prediction model is a deep learning model that, after training, can identify and predict the position of lanes from an input lane image. This model uses feature extraction techniques to obtain an initial feature map from a first lane image and fuses it with labeled lane information from a second lane image to generate a fused feature map. Based on this fused feature map, the model generates a lane prediction image containing the lane positions inferred by the model. By continuously comparing the predicted lane positions with the actual lane positions and optimizing the model's parameters based on the differences, a more accurate and robust lane prediction model is ultimately obtained. Understandably, this lane prediction model is the core of the entire temporal supervised modeling method, and its performance directly affects the accuracy and stability of lane detection in the autonomous driving system.

[0082] Specifically, refer to Figure 3First, the predicted lane lines in the lane prediction image are compared with their corresponding labeled lane lines. A predefined loss function is used to quantify the difference between the predicted and labeled lane lines. Understandably, a loss function is a metric used to measure the error between the lane prediction model's output and the actual situation. Differences include, but are not limited to, lane line positional deviations and shape mismatches. For example, the Euclidean distance between the keypoints of the predicted lane line (such as start point, end point, angle, and keypoints, where keypoints are points located on the line connecting the start and end points) and their corresponding points on the labeled lane line is calculated as part of the difference.

[0083] Then, based on the difference (loss value) calculated in the previous step, the parameters of the lane prediction model are updated using the backpropagation algorithm. The purpose of this step is to minimize the loss function, enabling the lane prediction model to predict lane positions more accurately. Understandably, this process involves multiple iterations, each using a new dataset or different batches of data from the same dataset to optimize the model parameters in the lane prediction model. To further improve model performance, the above process needs to be repeated continuously, starting with acquiring a new batch of two consecutive frames of lane images from the time-series images, proceeding through feature extraction, feature fusion, lane prediction, until the difference is calculated and the model parameters of the lane prediction model are adjusted.

[0084] As the training iterations increase, the lane prediction model gradually learns to more accurately identify lane lines in different scenarios, maintaining high detection accuracy and robustness even in the presence of occlusion or interference. After multiple training rounds, when the lane prediction model achieves satisfactory performance metrics on the validation or test set, the training is considered complete, and the final lane prediction model is obtained. This lane prediction model can be used to process road images captured by the vehicle's front camera in real time and accurately predict the position of lane lines, providing crucial support for autonomous driving systems.

[0085] In one embodiment, an initial frame lane line image is acquired, which includes labeled lane lines; a homography transformation matrix is ​​determined based on the labeling content of the labeled lane lines; the labeled lane lines in the initial frame lane line image are transformed using the homography transformation matrix to obtain a lane line sequence image that simulates the transformation caused by vehicle movement, the lane line sequence image including multiple consecutive frame lane line images located after the initial frame; a first lane line image and a second lane line image are selected from the multiple consecutive frame lane line images.

[0086] The initial frame lane line image refers to the first road image containing lane lines captured by the vehicle's camera. Understandably, the initial frame lane line image will serve as the basis for subsequent transformations.

[0087] Annotated lane lines refer to the lane line positions marked in the initial frame lane line image. Optionally, this annotation is done manually to indicate the exact location, shape, and other information of the lane lines.

[0088] The annotation content refers to various information recorded for the annotated lane lines, such as the start point, end point, angle, and distances between key points. Understandably, this detailed information is crucial for determining how to apply the homography transformation matrix.

[0089] The homography transformation matrix is ​​a mathematical tool used to simulate the effect of changing perspective caused by vehicle movement. This matrix can transform lane lines in an initial frame into lane lines in a series of consecutive frames, thus generating a series of images that appear to be captured while the vehicle is moving.

[0090] Lane line sequence images refer to a set of continuous time-series images obtained by transforming the labeled lane lines in the initial frame lane line image using a homography transformation matrix. This set of lane line sequence images simulates the lane line conditions seen by a vehicle at different time points due to movement, including multiple consecutive frame lane line images following the initial frame.

[0091] Specifically, a road image is first acquired from the vehicle's front-facing camera or other sensors as the initial frame. Optionally, the initial frame should contain clearly visible and labeled lane lines. The labels include, but are not limited to, information such as start and end points, angles, and distances between key points.

[0092] Then, based on the specific annotations of the lane lines in the initial frame (such as start point, end point, angle, etc.), a homography transformation matrix is ​​calculated to simulate the effect of vehicle movement. The homography transformation matrix will be used to generate a series of consecutive frames that can mimic the different lane line positions seen by the vehicle due to changes in viewing angle during driving.

[0093] Next, the calculated homography transformation matrix is ​​used to transform the labeled lane lines in the initial frame, generating a series of continuous time-series images. These generated images represent the lane line views seen by the vehicle at different points in time as it moves, i.e., lane line sequence images. Each sequence image includes the changes in the position of the lane lines, reflecting the dynamic process of the vehicle traveling on the road.

[0094] Finally, two adjacent frames are selected from the generated consecutive lane line images as the first lane line image (current frame) and the second lane line image (previous frame). Understandably, these two images will be used for subsequent feature extraction and fusion processing to train the lane line prediction model.

[0095] This method, which starts from a single initial frame and uses mathematical transformation techniques to automatically generate a series of consecutive frames, then selects specific frames for further analysis, not only reduces the workload of manual annotation but also improves the diversity and accuracy of training data for lane detection models.

[0096] In one embodiment, the positional difference between the labeled lane line and the predicted lane line is determined based on the positions of multiple labeled points in the first lane line image and the positions of multiple predicted points in the lane line prediction image; the angular difference between the labeled lane line and the predicted lane line is determined based on the labeled angle and the predicted angle.

[0097] Here, the annotation points refer to specific location points that are manually or automatically marked in the initial frame lane line image. Optionally, these annotation points are distributed along the actual lane lines and include, but are not limited to, the start and end points of the lane lines and key feature points, to accurately describe the position and shape of the lane lines.

[0098] A prediction point is a location point on the lane line identified in the lane line prediction image generated by the lane line prediction model.

[0099] Location discrepancy refers to the spatial distance difference between the marked point (the actual lane line location) and the predicted point (the lane line location predicted by the lane line prediction model). Understandably, calculating this discrepancy is to evaluate the accuracy of the lane line prediction model in predicting lane line locations. The smaller the location discrepancy, the more accurate the lane line prediction model is.

[0100] The annotation angle refers to the angle value determined when annotating lane lines in the initial frame of the lane line image. This angle can be the angle of the lane line relative to a certain reference direction (such as the vehicle's direction of travel or the direction of the image coordinate system), used to describe the directionality of the lane line.

[0101] The predicted angle refers to the lane line angle calculated from the lane line prediction image based on the output of the lane line prediction model. It reflects the lane line prediction model's prediction result regarding the lane line directionality.

[0102] Angular difference refers to the difference between the labeled angle and the predicted angle. Calculating angular difference is to measure the accuracy of the lane prediction model in predicting the direction of lane lines. A smaller angular difference means that the lane prediction model can better capture the directional characteristics of the lane lines.

[0103] Specifically, the first step is to analyze the labeled lane lines in the initial frame lane line image. This includes: the labeling angle, which determines the angle of the lane line relative to a reference direction (such as the vehicle's direction of travel or the direction of the image coordinate system), and multiple labeling points, which identify and record the positional information of multiple key points on the lane line. These points include, but are not limited to, the start point, the end point, and other feature points.

[0104] Then, in the generated lane prediction image, the lane lines predicted by the lane prediction model are analyzed. Specifically, this includes: the predicted angle, which is the angle of the lane line calculated based on the prediction results of the lane prediction model, and multiple predicted points, which are the location information of key points on the predicted lane line corresponding to the actual marked points.

[0105] Next, the positional differences between the labeled points and the predicted points are compared. Specifically, for each labeled point, the corresponding predicted point in the lane line prediction image is located. The spatial distance between each labeled point and its corresponding predicted point, i.e., the positional difference, is calculated. Optionally, this can be achieved using metrics such as Euclidean distance. The positional differences of all points are then combined to evaluate the overall positional accuracy of the lane line.

[0106] Finally, the difference between the labeled angle and the predicted angle is compared. Specifically, this involves directly comparing the labeled angle of the lane lines with the predicted angle of the lane lines, and calculating the difference to determine the angle discrepancy. Understandably, this difference reflects the accuracy of the lane line prediction model in predicting the direction of the lane lines.

[0107] Determining positional and angular differences provides specific error metrics for model training. These metrics can be used as part of the loss function to guide the backpropagation algorithm in adjusting model parameters to minimize the gap between predicted and true values, thereby improving model performance. Furthermore, by accurately measuring the positional and directional differences of lane lines and improving the model accordingly, erroneous decisions caused by inaccurate lane line detection, such as incorrect steering or lane departure, can be reduced. This is crucial for improving the safety and stability of autonomous driving systems.

[0108] In one embodiment, for each marked point in the first lane line image, a target predicted point matching the marked point is found among multiple predicted points in the lane line prediction image; the horizontal distance between the marked point and the target predicted point in the horizontal direction and the vertical distance in the vertical direction are determined; based on the horizontal and vertical distances, the Euclidean distance between the marked point and the target predicted point is determined; and the Euclidean distance between the marked point and the target predicted point is used as the positional difference between the marked lane line and the predicted lane line.

[0109] Here, the target predicted point refers to the best-matching predicted point found in the lane line prediction image for each labeled point in the first lane line image. Optionally, a distance metric or similarity comparison algorithm can be used to determine which predicted point is closest to a given labeled point.

[0110] The horizontal direction refers to the X-axis direction in the image coordinate system. In this application, the horizontal direction is used to describe the relative positional relationship between the labeled point and the target predicted point in the horizontal direction.

[0111] The vertical direction refers to the Y-axis direction in the image coordinate system. In this application, the vertical direction is used to describe the relative positional relationship between the labeled point and the target predicted point in the vertical direction.

[0112] Horizontal distance refers to the difference in the X-axis coordinates between a specific labeled point and its corresponding predicted target point in an image coordinate system. In other words, it's the horizontal distance between two points that are in the same row but different columns.

[0113] Vertical distance refers to the difference in the Y-axis coordinate between a specific labeled point and its corresponding target prediction point in an image coordinate system. In other words, it's the vertical distance between two points that are in the same column but not in the same row.

[0114] Euclidean distance refers to the straight-line distance between two points calculated based on the Pythagorean theorem. In this application, the actual spatial distance between the marked point and its corresponding target prediction point is calculated by combining horizontal and vertical distances.

[0115] Specifically, for each labeled point in the first lane line image, the best-matching target predicted point is found in the lane line prediction image. Optionally, the closest predicted point is determined by comparing the similarity or distance between each labeled point and all predicted points. This step is fundamental to ensuring the accuracy of subsequent calculations.

[0116] Then, for each pair of labeled points and target predicted points found, their horizontal distance (X-axis) and vertical distance (Y-axis) are calculated. Specifically, the horizontal distance is calculated as: horizontal distance = |target predicted point's X-coordinate - labeled point's X-coordinate|; vertical distance is calculated as: |target predicted point's Y-coordinate - labeled point's Y-coordinate|. These distance values ​​understandably reflect the offset between the two points along their respective coordinate axes, providing the necessary parameters for the next step of calculating the Euclidean distance.

[0117] Next, using the horizontal and vertical distances obtained in the above steps, the actual spatial distance between the marked point and the target predicted point is calculated using the Euclidean distance formula. Understandably, Euclidean distance provides an intuitive way to quantify the straight-line distance between two points, taking into account deviations in both the X and Y axes, thus reflecting the differences between them more comprehensively.

[0118] Finally, the Euclidean distance between each pair of labeled points and the target predicted point is used as the positional difference between the labeled lane line and the predicted lane line at that point. This positional difference can be used to evaluate the accuracy of the model in lane line position prediction. For example, by analyzing the positional differences of multiple points, the excellent parts of the model and the areas that need improvement can be identified, thereby guiding further optimization of the model.

[0119] By utilizing detailed comparisons between multiple labeled and predicted points, rather than a single overall metric, a more comprehensive evaluation of model performance is provided. This helps identify local anomalies and ensures the model maintains good performance under various conditions. Furthermore, incorporating Euclidean distance as part of the loss function directly guides parameter tuning during model training, making the training objective more explicit: minimizing the spatial distance difference between the actual and predicted lane lines.

[0120] In one embodiment, a target annotation point is determined from multiple annotation points based on the Euclidean distance between the target annotation point and the target prediction point and a preset distance threshold; the model parameters are adjusted based on the number of target annotation points until the number of target annotation points is less than the preset annotation point value, thereby obtaining the lane line prediction model.

[0121] The preset distance threshold is a pre-defined value used to determine whether a given marker is considered a "target marker." Specifically, if the Euclidean distance between a marker and its corresponding target prediction point exceeds this threshold, the marker may be considered an example of a lane prediction model failing to accurately predict a lane line. This threshold helps define the acceptable error range for the lane line prediction model's prediction results.

[0122] Target markers are those markers whose Euclidean distance to their corresponding target predicted points exceeds a preset distance threshold. In essence, target markers are the locations where lane prediction models make inaccurate predictions. Identifying these target markers helps pinpoint areas where lane prediction models need improvement.

[0123] The preset label point value refers to a pre-defined threshold for the maximum allowable number of erroneous label points. During the training process of the lane prediction model, if the number of target label points (i.e., the number of inaccurately predicted points) is less than or equal to this value, the lane prediction model is considered to have reached the expected performance standard. This value is an important indicator for measuring the performance of the lane prediction model, reflecting the overall prediction accuracy requirements of the model.

[0124] A lane prediction model refers to a trained final model capable of predicting the position and shape of lane lines based on an input image. In this application, "obtaining a lane prediction model" means reducing the number of target annotation points by continuously adjusting the model parameters until a preset standard is met (i.e., the number of target annotation points is less than the preset annotation point value), thereby obtaining a model with the expected performance.

[0125] Specifically, for each marker in the first lane image, the Euclidean distance between it and the corresponding target prediction point is calculated. A preset distance threshold is used to filter markers whose Euclidean distance exceeds the threshold as target markers. These target markers represent locations where the lane prediction model makes inaccurate predictions. Understandably, the selection of target markers is based on their large deviation from the predicted points, indicating that the lane prediction model performs poorly at these locations and needs improvement.

[0126] Then, the number of all identified target annotation points is counted. The purpose is to understand the extent of the prediction error of the current lane line prediction model. The number of target annotation points directly reflects the accuracy of the lane line prediction model in the lane line detection task.

[0127] Next, the model parameters are adjusted based on the number of target annotation points. If the number of target annotation points exceeds the preset number (i.e., the maximum allowed number of erroneous annotation points), performance needs to be optimized by adjusting the model parameters, including but not limited to updating weights, changing the network architecture, or adjusting hyperparameters. The goal is to reduce the prediction error of the lane line prediction model so that more annotation points can be correctly predicted.

[0128] Finally, repeat the above steps until the number of target markers drops below the preset marker value, resulting in an optimized lane prediction model. At this point, the lane prediction model has reached the expected performance level and can accurately predict the position and shape of lane lines.

[0129] By calculating the Euclidean distance between the labeled points and the target predicted points, and using a preset distance threshold to filter out the target labeled points, it is possible to accurately identify where the model's predictions have significant deviations. This method allows developers to specifically improve the model, rather than blindly making global adjustments. Furthermore, adjusting model parameters based on the number of target labeled points until the number of target labeled points falls below a preset value ensures that training is only considered complete when the model can accurately predict lane lines in most situations. This significantly reduces the model's false prediction rate and improves overall detection accuracy.

[0130] In one embodiment, a first convolution kernel is applied to the initial feature map of the first lane line image to obtain a first feature map; a second convolution kernel is applied to the labeled lane lines in the second lane line image to obtain a second feature map; the first and second feature maps are linearly combined to obtain a combined feature map; the feature values ​​of the combined feature map are compressed using a predictive activation function to obtain an output feature map, where the feature values ​​represent the information retention ratio at each position of the labeled lane lines in the second lane line image; and a fused feature map is generated based on the output feature map, the initial feature map, and the labeled lane lines in the second lane line image.

[0131] Here, the first convolution kernel refers to a filter used to perform a convolution operation on the initial feature map of the first lane line image. By applying this convolution kernel, specific feature information can be extracted from the initial feature map to form the first feature map.

[0132] The first feature map refers to the result obtained by applying the first convolution kernel to the initial feature map of the first lane line image. It contains the feature representations extracted from the original image, which is helpful for the subsequent feature fusion process.

[0133] The second convolution kernel refers to another filter used to perform a convolution operation on the labeled lane lines in the second lane line image. Its purpose is to extract useful feature information from the labeled lane lines and generate a second feature map.

[0134] The second feature map refers to the result obtained by applying a second convolution kernel to the labeled lane lines in the second lane line image. It reflects the important feature information of the labeled lane lines.

[0135] A combined feature map is a new feature map obtained by linearly combining the first and second feature maps (e.g., by weighted summation). The purpose of this step is to integrate the feature information from the two images, providing a richer data foundation for subsequent processing.

[0136] Predictive activation functions, as non-linear functions, are used in neural networks to enhance the expressiveness of the model. In this application, predictive activation functions are used to compress or adjust the feature values ​​in the combined feature map to better reflect the information retention ratio at each location.

[0137] The output feature map refers to the combined feature map after processing by the predictive activation function. The feature values ​​represent the proportion of information retained at each location in the second lane image, that is, the degree to which information at that location is preserved during the fusion process.

[0138] Feature values ​​refer to the numerical values ​​of each pixel in a feature map, representing the feature intensity or the probability of a certain feature existing within the corresponding region. In this application, after processing by the prediction activation function, the feature values ​​can be understood as the importance or retention ratio of information at each location.

[0139] Information retention ratio refers to the proportion of information at each location that is retained relative to the original input (such as the labeled lane lines in the second lane line image) during the fusion process. Understandably, a higher ratio means that more original information is retained in the final fused feature map.

[0140] Specifically, firstly, a first convolution kernel is applied to the initial feature map of the first lane line image for convolution. This operation extracts key feature information from the original image, resulting in a first feature map. Then, a second convolution kernel is applied to the labeled lane lines in the second lane line image for convolution, resulting in a second feature map. This second feature map is a feature representation extracted based on the labeled lane lines in the second lane line image.

[0141] The obtained first feature map is then linearly combined with the second feature map (e.g., by weighted summation) to generate a combined feature map. This combined feature map integrates key feature information from the two images, providing a rich data foundation for subsequent processing.

[0142] Next, the feature values ​​in the combined feature map are compressed or adjusted using a predictive activation function to obtain the output feature map. The adjusted feature values ​​in the combined feature map characterize the proportion of information retained at each location in the second lane image, i.e., the degree to which information at that location is preserved during the fusion process.

[0143] Finally, based on the obtained output feature map, the initial feature map, and the labeled lane lines in the second lane line image, a final fused feature map is generated. This feature fusion process completes the final fused feature map, which integrates information from different frames, enhancing the model's ability to detect lane lines and improving accuracy and robustness.

[0144] This process integrates multi-source information by generating the final fused feature map based on the output feature map, the initial feature map, and the labeled lane lines, making lane line detection more accurate and reliable. In particular, modeling the displacement difference between consecutive frames helps capture dynamic changes, further improving the overall performance of lane line detection. Furthermore, fusing information from different frames (i.e., the first lane line image and the second lane line image) effectively addresses the problem of insufficient information in a single frame due to occlusion, lighting changes, or other interference factors. This temporal information supplementation enhances the model's robustness and stability in complex environments.

[0145] In one embodiment, referenceFigure 4 The training method for lane prediction models mainly includes three steps:

[0146] Step 1: Data generation, which mainly includes: using a single frame image as input, using homography transformation to generate a time series of lane line images from the labeled single frame image, and outputting an image sequence.

[0147] The second step is feature extraction, which mainly includes: using the image sequence as input, extracting features from the image sequence using the feature extraction unit to obtain feature maps, and then adding the obtained feature maps to the data pool.

[0148] The third step, temporal supervision, mainly includes: using feature maps as input, processing feature fusion using convolutional gated recurrent units to optimize the temporal supervision process. The model parameters are then adjusted using a loss function to minimize the prediction error (i.e., keypoint distance), outputting a trained lane prediction model.

[0149] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0150] Based on the same inventive concept, this application also provides a lane prediction model training device for implementing the lane prediction model training method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more lane prediction model training device embodiments provided below can be found in the limitations of the lane prediction model training method described above, and will not be repeated here.

[0151] In one exemplary embodiment, such as Figure 5 As shown, a lane line prediction model training device 500 is provided, including: an acquisition module 502, an extraction module 504, a fusion module 506, a generation module 508, and a training module 510, wherein:

[0152] The module 502 is used to obtain a first lane line image and a second lane line image, wherein the second lane line image is the previous frame image of the first lane line image.

[0153] Extraction module 504 is used to extract an initial feature map from the first lane line image. Both the first lane line image and the second lane line image include labeled lane lines.

[0154] The fusion module 506 is used to fuse the initial feature map of the first lane line image and the labeled lane lines in the second lane line image to obtain a fused feature map.

[0155] The generation module 508 is used to generate an image based on the fused feature map to obtain a lane line prediction image, which includes predicted lane lines.

[0156] Training module 510 is used to determine the difference between the labeled lane lines and the predicted lane lines, and to train the model based on the difference to obtain the lane line prediction model.

[0157] In one embodiment, the obtaining module 502 is used to obtain an initial frame lane line image, which includes labeled lane lines; determine a homography transformation matrix based on the labeling content of the labeled lane lines; transform the labeled lane lines in the initial frame lane line image using the homography transformation matrix to obtain a lane line sequence image that simulates the transformation caused by vehicle movement, the lane line sequence image including multiple consecutive frame lane line images located after the initial frame; and select a first lane line image and a second lane line image from the multiple consecutive frame lane line images.

[0158] In one embodiment, the training module 510 is used to determine the positional difference between the labeled lane lines and the predicted lane lines based on the positions of multiple labeled points in the first lane line image and the positions of multiple predicted points in the lane line prediction image; and to determine the angular difference between the labeled lane lines and the predicted lane lines based on the labeled angle and the predicted angle.

[0159] In one embodiment, the training module 510 is configured to, for each marked point in the first lane line image, find a target prediction point that matches the marked point among a plurality of prediction points in the lane line prediction image; determine the horizontal distance in the horizontal direction and the vertical distance in the vertical direction between the marked point and the target prediction point; determine the Euclidean distance between the marked point and the target prediction point based on the horizontal and vertical distances; and use the Euclidean distance between the marked point and the target prediction point as the positional difference between the marked lane line and the predicted lane line.

[0160] In one embodiment, the training module 510 is used to determine the target annotation point among multiple annotation points based on the Euclidean distance between the target annotation point and the target prediction point and a preset distance threshold; and to adjust the parameters of the model based on the number of target annotation points until the number of target annotation points is less than the preset annotation point value, thereby obtaining the lane line prediction model.

[0161] In one embodiment, the fusion module 506 is configured to apply a first convolution kernel to the initial feature map of the first lane line image to obtain a first feature map; apply a second convolution kernel to the lane lines already marked in the second lane line image to obtain a second feature map; linearly combine the first feature map and the second feature map to obtain a combined feature map; compress the feature values ​​of the combined feature map using a predictive activation function to obtain an output feature map, wherein the feature values ​​in the output feature map are used to characterize the information retention ratio at each position of the lane lines already marked in the second lane line image; and generate a fused feature map based on the output feature map, the initial feature map, and the lane lines already marked in the second lane line image.

[0162] Each module in the aforementioned lane prediction model training device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0163] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores data related to lane prediction model training. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a model connection. When the computer program is executed by the processor, it implements a lane prediction model training method.

[0164] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0165] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0166] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0167] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0168] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0169] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0170] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0171] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for training a lane line prediction model, characterized in that, The method comprises: acquiring an initial frame lane line image, wherein the initial frame lane line image comprises a labeled lane line; determining a homographic transformation matrix based on the labeled content of the labeled lane line; transforming the labeled lane line in the initial frame lane line image through the homographic transformation matrix to obtain a lane line sequence image that is transformed due to simulated vehicle movement, wherein the lane line sequence image comprises a plurality of continuous frame lane line images after the initial frame; selecting a first lane line image and a second lane line image from the plurality of continuous frame lane line images, wherein the second lane line image is a previous frame image of the first lane line image; extracting an initial feature map of the first lane line image, wherein the first lane line image and the second lane line image both comprise a labeled lane line; performing feature fusion on the initial feature map of the first lane line image and the labeled lane line in the second lane line image to obtain a fused feature map; generating an image based on the fused feature map to obtain a lane line prediction image, wherein the lane line prediction image comprises a predicted lane line; determining a difference between the labeled lane line and the predicted lane line, and training a model based on the difference to obtain a lane line prediction model.

2. The method of claim 1, wherein, The labeled lane line comprises a labeled angle and a plurality of labeled points; and the predicted lane line comprises a predicted angle and a plurality of predicted points. The determination of the difference between the labeled lane line and the predicted lane line comprises: determining a position difference between the labeled lane line and the predicted lane line based on the positions of the plurality of labeled points in the first lane line image and the positions of the plurality of predicted points in the lane line prediction image; and determining an angle difference between the labeled lane line and the predicted lane line based on the labeled angle and the predicted angle.

3. The method of claim 2, wherein, The determination of the position difference between the labeled lane line and the predicted lane line based on the positions of the plurality of labeled points in the first lane line image and the positions of the plurality of predicted points in the lane line prediction image comprises: for each labeled point in the first lane line image, searching for a target predicted point that matches the labeled point from the plurality of predicted points in the lane line prediction image; determining a horizontal distance and a vertical distance between the labeled point and the target predicted point in a horizontal direction and a vertical direction, respectively; determining an Euclidean distance between the labeled point and the target predicted point based on the horizontal distance and the vertical distance; and taking the Euclidean distance between the labeled point and the target predicted point as the position difference between the labeled lane line and the predicted lane line.

4. The method of claim 3, wherein, The training of the model based on the difference to obtain the lane line prediction model comprises: determining a target labeled point from the plurality of labeled points based on the Euclidean distance between the labeled point and the target predicted point and a preset distance threshold; and adjusting parameters of the model based on the number of target labeled points until the number of target labeled points is less than a preset number of labeled points to obtain the lane line prediction model.

5. The method according to any one of claims 1 to 4, characterized in that, The feature fusion of the initial feature map of the first lane line image and the labeled lane line in the second lane line image comprises: performing convolution operation on the initial feature map of the first lane line image by applying a first convolution kernel to obtain a first feature map; performing convolution operation on the labeled lane line in the second lane line image by applying a second convolution kernel to obtain a second feature map; performing linear combination on the first feature map and the second feature map to obtain a combined feature map; compressing feature values of the combined feature map by a prediction activation function to obtain an output feature map, wherein the feature values in the output feature map are used to represent information retention ratios at each position of the labeled lane line in the second lane line image; generating a fusion feature map based on the output feature map, the initial feature map and the labeled lane line in the second lane line image. 6.A lane line prediction model training apparatus, characterized in that, The device comprises: An obtaining module is configured to obtain an initial frame lane line image, wherein the initial frame lane line image comprises a labeled lane line; determine a homographic transformation matrix based on the labeled content of the labeled lane line; transform the labeled lane line in the initial frame lane line image by using the homographic transformation matrix to obtain a lane line sequence image that is transformed due to vehicle movement, wherein the lane line sequence image comprises a plurality of continuous frame lane line images after the initial frame; and select a first lane line image and a second lane line image from the plurality of continuous frame lane line images, wherein the second lane line image is a previous frame image of the first lane line image; An extraction module is configured to extract an initial feature map from the first lane line image, wherein the first lane line image and the second lane line image both comprise a labeled lane line; A fusion module is configured to perform feature fusion on the initial feature map of the first lane line image and the labeled lane line in the second lane line image to obtain a fusion feature map; A generation module is configured to perform image generation based on the fusion feature map to obtain a lane line prediction image, wherein the lane line prediction image comprises a predicted lane line; A training module is configured to determine a difference between the labeled lane line and the predicted lane line, perform model training based on the difference, and obtain a lane line prediction model.

7. The apparatus of claim 6, wherein, The labeled lane line comprises a labeled angle and a plurality of labeled points; and the predicted lane line comprises a predicted angle and a plurality of predicted points. The training module is further configured to determine a position difference between the labeled lane line and the predicted lane line based on positions of the plurality of labeled points in the first lane line image and positions of the plurality of predicted points in the lane line prediction image; and determine an angle difference between the labeled lane line and the predicted lane line based on the labeled angle and the predicted angle.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor implements the steps of the method of any one of claims 1 to 5 when executing the computer program.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 5.

10. A computer program product comprising a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 5.

Images