Lane line detection model training, lane line detection method and device

Abstract

The present application discloses a lane line detection model training, lane line detection method, and lane line detection device. The method includes: dividing a lane line sample image into a first lane line sample image and a second lane line sample image according to whether the vehicle is in a lane line crossing scene; the first lane line sample image has a normal scene label and a first annotated lane line category that does not include a lane line crossing category, and the second lane line sample image has a lane line crossing scene label and a second annotated lane line category that includes a lane line crossing category. Using the first lane line sample image and the second lane line sample image, a preset detection model including a classification sub-model, a first segmentation sub-model, and a second segmentation sub-model is trained to obtain a lane line detection model; the classification sub-model predicts the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model predicts the lane line category of the first lane line sample image, and the second segmentation sub-model predicts the lane line category of the second lane line sample image.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a lane line detection model training, lane line detection method and device. Background Art

[0002] In autonomous driving scenarios, lane line detection technology is the core technology that provides functional services such as vehicle positioning, lane keeping and lane departure warning; therefore, lane line detection technology has high requirements for accuracy.

[0003] Currently, the lane line detection method involves labeling each lane line with a lane line category based on the relative position of the lane line and the vehicle to distinguish the differences between the lane lines. Based on this, a segmentation model is trained; the trained segmentation model is used to process lane line images without labeled lane line categories to obtain the lane line category of each lane line.

[0004] However, for lane line images in a lane-crossing scenario, the relative position of the vehicle and the lane line is unclear because the vehicle has run over the lane line, making it impossible to accurately label each lane line with its lane line category. This leads to inconsistent lane line category labeling. This situation results in poor training results for the segmentation model, which in turn leads to inaccurate lane line detection when processing lane line images without lane line category labels, which in turn has a serious impact on subsequent functional services of autonomous driving. Summary of the Invention

[0005] In view of this, the embodiments of the present application provide a lane line detection model training, lane line detection method and device, so that the lane line detection model is not only applicable to lane line images of vehicles in non-line-crossing scenarios, but also applicable to detecting lane line images of vehicles in line-crossing scenarios, thereby improving the detection accuracy of the lane line detection model.

[0006] In a first aspect, an embodiment of the present application provides a method for training a lane detection model, the method comprising:

[0007] Acquire a first lane line sample image in a non-lane-crossing scenario and a second lane line sample image in a lane-crossing scenario; the first lane line sample image includes a normal scene label and a first annotated lane line category, the second lane line sample image includes a lane-crossing scene label and a second annotated lane line category, the first annotated lane line category does not include a lane-crossing category, and the second annotated lane line category includes the lane-crossing category;

[0008] According to the first lane line sample image and the second lane line sample image, a preset detection model is trained to obtain a lane line detection model; the preset detection model includes a classification sub-model, a first segmentation sub-model and a second segmentation sub-model, the classification sub-model is used to predict the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model is used to predict the lane line category of the first lane line sample image, and the second segmentation sub-model is used to predict the lane line category of the second lane line sample image.

[0009] Optionally, the training of a preset detection model to obtain a lane line detection model based on the first lane line sample image and the second lane line sample image includes:

[0010] Performing scene classification on the first lane line sample image using the classification sub-model to obtain a first predicted scene label for the first lane line sample image;

[0011] Performing scene classification on the second lane line sample image using the classification sub-model to obtain a second predicted scene label for the second lane line sample image;

[0012] Performing lane line segmentation on the first lane line sample image using the first segmentation sub-model to obtain a first predicted lane line category of the first lane line sample image;

[0013] Performing lane line segmentation on the second lane line sample image using the second segmentation sub-model to obtain a second predicted lane line category for the second lane line sample image;

[0014] Adjusting model parameters of the classification sub-model and model parameters of the first segmentation sub-model according to the first predicted scene label, the first predicted lane line category, the common scene label, the first annotated lane line category, a first loss function, and a second loss function until training of the preset detection model is completed; the first loss function corresponds to the classification sub-model, and the second loss function corresponds to the first segmentation sub-model and the second segmentation sub-model;

[0015] Adjusting the model parameters of the classification sub-model and the model parameters of the second segmentation sub-model according to the second predicted scene label, the second predicted lane line category, the lane crossing scene label, the second labeled lane line category, the first loss function, and the second loss function until the preset detection model training is completed;

[0016] The preset detection model that has been trained is determined as the lane line detection model.

[0017] Optionally, the classification sub-model includes a feature extraction layer and a classifier; and performing scene classification on the first lane line sample image using the classification sub-model to obtain a first predicted scene label for the first lane line sample image includes:

[0018] Performing feature extraction on the first lane line sample image by the feature extraction layer to obtain a first feature image of the first lane line sample image;

[0019] Performing scene classification on the first feature image by the classifier to obtain the first predicted scene label;

[0020] The performing scene classification on the second lane line sample image by the classification sub-model to obtain a second predicted scene label for the second lane line sample image includes:

[0021] Performing feature extraction on the second lane line sample image by the feature extraction layer to obtain a second feature image of the second lane line sample image;

[0022] The second feature image is subjected to scene classification by the classifier to obtain the second predicted scene label.

[0023] Optionally, the first segmentation sub-model includes a first decoding layer and a first segmentation layer; and performing lane line segmentation on the first lane line sample image using the first segmentation sub-model to obtain a first predicted lane line category for the first lane line sample image includes:

[0024] Performing feature decoding on the first feature image through the first decoding layer to obtain a third feature image of the first lane line sample image; the third feature image has the same image size as the first lane line sample image;

[0025] Lane line segmentation is performed on the third feature image through the first segmentation layer to obtain the first predicted lane line category.

[0026] Optionally, the second segmentation sub-model includes a second decoding layer and a second segmentation layer; and performing lane line segmentation on the second lane line sample image using the second segmentation sub-model to obtain a second predicted lane line category for the second lane line sample image includes:

[0027] Performing feature decoding on the second feature image through the second decoding layer to obtain a fourth feature image of the second lane line sample image; the fourth feature image has the same image size as the second lane line sample image;

[0028] Lane line segmentation is performed on the fourth feature image through the second segmentation layer to obtain the second predicted lane line category.

[0029] In a second aspect, an embodiment of the present application provides a lane line detection method, the method comprising:

[0030] Obtain the lane line image to be detected;

[0031] Performing lane line detection on the lane line image to be detected using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected;

[0032] The lane line detection model is the lane line detection model described in the first aspect above.

[0033] Optionally, performing lane line detection on the lane line image to be detected using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected includes:

[0034] Performing scene classification on the lane line image to be detected by the classification sub-model to obtain a third predicted scene label of the lane line image to be detected;

[0035] If the third predicted scene label is the common scene label, performing lane line segmentation on the lane line image to be detected using the first segmentation sub-model to obtain the third predicted lane line category;

[0036] If the third predicted scene label is the lane line crossing scene label, lane line segmentation is performed on the lane line image to be detected using the second segmentation sub-model to obtain the third predicted lane line category.

[0037] Optionally, the method further includes:

[0038] Merging the conditional convolution layer, the first segmentation sub-model, and the second segmentation sub-model to obtain a segmentation sub-model; when the conditional convolution layer indicates that the third predicted scene label is the normal scene label, the segmentation sub-model is the first segmentation sub-model; when the third predicted scene label is the line-pressing scene label, the segmentation sub-model is the second segmentation sub-model;

[0039] The performing lane line detection on the lane line image to be detected by using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected includes:

[0040] Performing scene classification on the lane line image to be detected by the classification sub-model to obtain a third predicted scene label of the lane line image to be detected;

[0041] Based on the third predicted scene label, lane line segmentation is performed on the lane line image to be detected through the segmentation sub-model to obtain the third predicted lane line category.

[0042] In a third aspect, an embodiment of the present application provides a lane detection model training device, the device comprising: a first acquisition unit and a training unit;

[0043] The first acquisition unit is configured to acquire a first lane line sample image in a non-lane-crossing scenario and a second lane line sample image in a lane-crossing scenario; the first lane line sample image includes a normal scene label and a first annotated lane line category, the second lane line sample image includes a lane-crossing scene label and a second annotated lane line category, the first annotated lane line category does not include a lane-crossing category, and the second annotated lane line category includes the lane-crossing category;

[0044] The training unit is used to train a preset detection model to obtain a lane line detection model based on the first lane line sample image and the second lane line sample image; the preset detection model includes a classification sub-model, a first segmentation sub-model and a second segmentation sub-model, the classification sub-model is used to predict the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model is used to predict the lane line category of the first lane line sample image, and the second segmentation sub-model is used to predict the lane line category of the second lane line sample image.

[0045] In a fourth aspect, an embodiment of the present application provides a lane line detection device, the device comprising: a second acquisition unit and a detection unit;

[0046] The second acquisition unit is used to acquire the lane line image to be detected;

[0047] The detection unit is configured to perform lane line detection on the lane line image to be detected using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected;

[0048] The lane line detection model is the lane line detection model described in the first aspect above.

[0049] Compared with the prior art, this application has at least the following advantages:

[0050] By adopting the technical solution of the embodiment of the present application, the lane line sample image is divided into a first lane line sample image and a second lane line sample image according to whether the vehicle is in a lane pressing scene; the first lane line sample image has a normal scene label and a first annotated lane line category that does not include the lane pressing category, and the second lane line sample image has a lane pressing scene label and a second annotated lane line category that includes the lane pressing category; the lane line detection model is obtained by training a preset detection model including a classification sub-model, a first segmentation sub-model and a second segmentation sub-model through the first lane line sample image and the second lane line sample image; wherein the classification sub-model predicts the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model predicts the lane line category of the first lane line sample image, and the second segmentation sub-model predicts the lane line category of the second lane line sample image.

[0051] This method distinguishes lane line sample images based on whether the vehicle is in a lane-crossing scenario, labeling them with different scene labels and lane line categories. A classification sub-model then discriminates whether the vehicle is in a lane-crossing scenario in the lane line sample images, using different first and second segmentation sub-models for model training. Based on this, the preset detection model's machine learning capabilities for the first lane line sample image in which the vehicle is not in a lane-crossing scenario are enhanced for the second lane line sample image in which the vehicle is in a lane-crossing scenario. The resulting lane line detection model is therefore applicable not only to lane line images in which the vehicle is not in a lane-crossing scenario, but also to lane line images in which the vehicle is in a lane-crossing scenario, thereby improving the detection accuracy of the lane line detection model.

[0052] In addition, the lane line image to be detected is input into the above-mentioned lane line detection model for lane line detection to obtain a third predicted lane line category of the lane line image to be detected; since the above-mentioned lane line detection model has a higher detection accuracy when processing lane line images to be detected without labeled lane line categories, the third predicted lane line category of the lane line image to be detected is more accurate, thereby improving the subsequent functional services of autonomous driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0054] Figure 1 A schematic diagram of a system framework involved in an application scenario in an embodiment of the present application;

[0055] Figure 2A flowchart of a lane detection model training method provided in an embodiment of the present application;

[0056] Figure 3 A schematic diagram of a first labeled lane line category of a first lane line sample image and a second labeled lane line category of a second lane line sample image provided in an embodiment of the present application;

[0057] Figure 4 A schematic diagram of a lane detection model training framework provided in an embodiment of the present application;

[0058] Figure 5 A schematic flow chart of a lane detection method provided in an embodiment of the present application;

[0059] Figure 6 A schematic diagram of a lane detection framework provided in an embodiment of the present application;

[0060] Figure 7 A schematic diagram of a lane detection model training and lane detection framework provided in an embodiment of the present application;

[0061] Figure 8 A schematic diagram of the structure of a lane detection model training device provided in an embodiment of the present application;

[0062] Figure 9 A schematic structural diagram of a lane detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to help those skilled in the art better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0064] At present, the lane line detection method is to label each lane line with the lane line category according to the relative position of the lane line and the vehicle to distinguish the differences between the lane lines, and then train the segmentation model based on this; use the trained segmentation model to process the lane line image without the lane line category labeled to obtain the lane line category of each lane line.

[0065] However, after research, the inventors found that for lane line images in a lane-crossing scenario, the relative position of the vehicle and the lane line is unclear because the vehicle runs over the lane line, making it impossible to accurately label the lane line category for each lane line, and there are cases where the lane line categories are inconsistently labeled; this situation leads to poor training results of the segmentation model, which leads to inaccurate lane line detection when processing lane line images without labeled lane line categories, which in turn has a serious impact on subsequent functional services of autonomous driving.

[0066] In order to solve this problem, in an embodiment of the present application, the lane line sample image is divided into a first lane line sample image and a second lane line sample image according to whether the vehicle is in a lane pressing scene; the first lane line sample image has a normal scene label and a first annotated lane line category that does not include a lane pressing category, and the second lane line sample image has a lane pressing scene label and a second annotated lane line category that includes a lane pressing category; through the first lane line sample image and the second lane line sample image, a preset detection model including a classification sub-model, a first segmentation sub-model and a second segmentation sub-model is trained to obtain a lane line detection model; wherein, the classification sub-model predicts the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model predicts the lane line category of the first lane line sample image, and the second segmentation sub-model predicts the lane line category of the second lane line sample image.

[0067] This method distinguishes lane line sample images based on whether the vehicle is in a lane-crossing scenario, labeling them with different scene labels and lane line categories. A classification sub-model then discriminates whether the vehicle is in a lane-crossing scenario in the lane line sample images, using different first and second segmentation sub-models for model training. Based on this, the preset detection model's machine learning capabilities for the first lane line sample image in which the vehicle is not in a lane-crossing scenario are enhanced for the second lane line sample image in which the vehicle is in a lane-crossing scenario. The resulting lane line detection model is therefore applicable not only to lane line images in which the vehicle is not in a lane-crossing scenario, but also to lane line images in which the vehicle is in a lane-crossing scenario, thereby improving the detection accuracy of the lane line detection model.

[0068] For example, one of the scenarios of the embodiment of the present application may be applied to Figure 1In the scenario shown. The scenario includes a camera 101, a terminal device 102, and a server 103; wherein the camera 101 captures a first lane line image of a vehicle in a non-lane-crossing scenario and a second lane line image of a vehicle in a lane-crossing scenario and sends them to the terminal device 102; the terminal device 102 annotates the first lane line image to obtain a first lane line sample image including a normal scene label and a first annotated lane line category; the terminal device 102 annotates the second lane line image to obtain a second lane line sample image including a lane-crossing scene label and a second annotated lane line category; the terminal device 102 sends the first lane line sample image and the second lane line sample image to the server 103. The server 103 executes the lane line detection model training method provided in the embodiment of the present application to obtain a lane line detection model.

[0069] The camera 101 collects the lane line image to be detected and sends it to the terminal device 102. The terminal device 102 sends the lane line image to be detected to the server 103. The server 103 executes the lane line detection method provided in the embodiment of the present application to obtain the lane line category of the lane line image to be detected.

[0070] First, in the above application scenario, although the action description of the implementation method provided by the embodiment of the present application is executed by the server 103; however, the embodiment of the present application is not restricted in terms of the execution subject, as long as the actions disclosed in the implementation method provided by the embodiment of the present application are executed.

[0071] Secondly, the above scenario is only an example scenario provided by the embodiment of the present application, and the embodiment of the present application is not limited to this scenario.

[0072] The specific implementation of the lane line detection model training, lane line detection method and device in the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0073] Exemplary Methods

[0074] First, the specific implementation method of lane line detection model training in the embodiment of the present application is described in detail through examples in conjunction with the accompanying drawings.

[0075] See also Figure 2 , shows a flow chart of a method for lane line detection model training and lane line detection in an embodiment of the present application. In this embodiment, the method may include the following steps:

[0076] Step 201: Acquire a first lane line sample image of a vehicle in a non-line-pressing scene and a second lane line sample image of a vehicle in a line-pressing scene; the first lane line sample image includes a normal scene label and a first annotated lane line category, the second lane line sample image includes a line-pressing scene label and a second annotated lane line category, the first annotated lane line category does not include a line-pressing category, and the second annotated lane line category includes the line-pressing category.

[0077] In related technologies, lane line detection methods involve labeling each lane line with a lane line category based on the relative position of the lane line and the vehicle to distinguish between the lane lines. A segmentation model is then trained based on this. Lane line images without lane line category labels are then processed using the trained segmentation model to determine the lane line category for each lane line. However, for lane line images in which a vehicle is in a lane-crossing scenario, the relative position of the vehicle and the lane line is unclear due to the vehicle's crossing of the lane line, making it impossible to accurately label each lane line with a lane line category. This results in inconsistent lane line category labeling. This poor training of the segmentation model leads to inaccurate lane line detection when processing lane line images without lane line category labels, which in turn has a serious impact on subsequent autonomous driving functional services.

[0078] Therefore, in the embodiment of the present application, the lane line sample images collected are differentiated according to whether the vehicle is in a lane-crossing scenario. For lane line sample images in which the vehicle is in a non-lane-crossing scenario, a normal scene label is annotated, and each lane line is annotated according to the first annotated lane line category based on the relative position of the lane line and the vehicle, thereby obtaining a first lane line sample image. For lane line sample images in which the vehicle is in a lane-crossing scenario, a lane-crossing scene label is annotated, and each lane line is annotated according to the second annotated lane line category, including the lane-crossing category, based on the position of the crossed lane line, thereby obtaining a second lane line sample image.

[0079] As an example, Figure 3 A schematic diagram of a first lane line sample image with a first labeled lane line category and a second lane line sample image with a second labeled lane line category is shown. For the first lane line sample image, each lane line is labeled as Left 1, Left 2, Left 3, ..., Right 1, Right 2, Right 3, ...; for the second lane line sample image, each lane line is labeled as Left 1, Left 2, Left 3, ..., Pressure Line, Right 1, Right 2, Right 3, ....

[0080] Step 202: Based on the first lane line sample image and the second lane line sample image, a preset detection model is trained to obtain a lane line detection model; the preset detection model includes a classification sub-model, a first segmentation sub-model and a second segmentation sub-model, the classification sub-model is used to predict the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model is used to predict the lane line category of the first lane line sample image, and the second segmentation sub-model is used to predict the lane line category of the second lane line sample image.

[0081] In the embodiment of the present application, after obtaining the first lane line sample image and the second lane line sample image in step 201, since the first lane line sample image and the second lane line sample image respectively represent a vehicle in a non-lane crossing scenario and a vehicle in a lane crossing scenario, in order to improve the machine learning capability of the second annotated lane line category of the second lane line sample image based on the machine learning of the first annotated lane line category of the first lane line sample image, it is necessary to design and use two different segmentation models to respectively perform machine learning for different scene categories. Furthermore, in order to distinguish the two segmentation models performing machine learning for different scenes, it is also necessary to design and use a classification sub-model to discriminate the scene category of the first lane line sample image and the second lane line sample image.

[0082] Based on this, a preset detection model is designed, consisting of a classification sub-model for predicting scene labels for the first and second lane line sample images, a first segmentation sub-model for predicting lane line categories for the first lane line sample images, and a second segmentation sub-model for predicting lane line categories for the second lane line sample images. The lane detection model is obtained by iteratively training the preset detection model using the first and second lane line sample images. This lane detection model is used to detect lane line categories in lane line images.

[0083] Specific implementation of step 202: First, the first lane line sample image is input into the classification sub-model for scene classification, which can output a first predicted scene label for the first lane line sample image. Similarly, the second lane line sample image is input into the classification sub-model for scene classification, which can output a first predicted scene label for the second lane line sample image.

[0084] Next, the first lane line sample image is input into the first segmentation sub-model for lane line segmentation, and the first predicted lane line category of the first lane line sample image can be obtained. Similarly, the second lane line sample image is input into the second segmentation sub-model for lane line segmentation, and the second predicted lane line category of the second lane line sample image can be obtained.

[0085] Then, for the first lane line sample image, the first predicted scene label and the normal scene label are substituted into the first loss function of the classification sub-model, and the first predicted lane line category and the first annotated lane line category are substituted into the second loss function of the first segmentation sub-model, thereby adjusting the model parameters of the classification sub-model and the model parameters of the first segmentation sub-model until the preset detection model training is completed. Based on the fact that the loss function of the second segmentation sub-model is also the second loss function, it can be similarly obtained that for the second lane line sample image, the second predicted scene label and the line-crossing scene label are substituted into the first loss function of the classification sub-model, and the second predicted lane line category and the second annotated lane line category are substituted into the second loss function of the second segmentation sub-model, thereby adjusting the model parameters of the classification sub-model and the model parameters of the second segmentation sub-model until the preset detection model training is completed.

[0086] Finally, the preset detection model trained through the above specific steps can be used as a lane line detection model for detecting lane line categories in lane line images. This lane line detection model is not only applicable to lane line images in which the vehicle is not in a lane line crossing scenario, but is also applicable to lane line images in which the vehicle is in a lane line crossing scenario, thereby improving the detection accuracy of the lane line detection model. For example, Figure 4 A schematic diagram of the lane detection model training framework is shown.

[0087] Therefore, in an optional implementation of the embodiment of the present application, step 202 may include, for example, the following steps A to G:

[0088] Step A: Perform scene classification on the first lane line sample image using the classification sub-model to obtain a first predicted scene label for the first lane line sample image.

[0089] Among them, to realize the scene classification of the lane line image, it is necessary to first determine the feature image of the lane line image, and then determine the scene category through the feature image. The classification sub-model can be designed to be divided into a feature extraction layer and a classifier. Correspondingly, the specific implementation method of step A is: first, the first lane line sample image is input into the feature extraction layer for feature extraction, and the first feature image of the first lane line sample image can be output; then, the first feature image is input into the classifier for scene classification, and the first predicted scene label of the first lane line sample image can be output. Therefore, in an optional implementation of the embodiment of the present application, the classification sub-model includes a feature extraction layer and a classifier; the step A can, for example, include the following steps A1-A2:

[0090] Step A1: performing feature extraction on the first lane line sample image through the feature extraction layer to obtain a first feature image of the first lane line sample image.

[0091] Step A2: performing scene classification on the first feature image by the classifier to obtain the first predicted scene label.

[0092] Step B: Perform scene classification on the second lane line sample image using the classification sub-model to obtain a second predicted scene label for the second lane line sample image.

[0093] Based on the specific implementation of step A, it can be similarly derived that in an optional implementation of the embodiment of the present application, step B may include, for example, the following steps B1 and B2:

[0094] Step B1: performing feature extraction on the second lane line sample image through the feature extraction layer to obtain a second feature image of the second lane line sample image.

[0095] Step B2: performing scene classification on the second feature image by the classifier to obtain the second predicted scene label.

[0096] The feature extraction layer may adopt, for example, an Xception network structure that implements multiple feature extractions; and the classifier may adopt, for example, an Argmax function.

[0097] Step C: Perform lane line segmentation on the first lane line sample image using the first segmentation sub-model to obtain a first predicted lane line category of the first lane line sample image.

[0098] It should be noted that, in order to realize lane line segmentation of the lane line image, it is necessary to first determine a feature image with the same image size as the lane line image, and then determine the lane line category through the feature image. In this case, the first segmentation sub-model can be designed to be divided into a first decoding layer and a first segmentation layer. Correspondingly, the specific implementation method of step C is: first, on the basis of step A, the first feature image is input into the first decoding layer for feature decoding, and a third feature image of the first lane line sample image can be output. The third feature image has the same image size as the first lane line sample image; then, the third feature image is input into the first segmentation layer for lane line segmentation, and the first predicted lane line category of the first lane line sample image can be output. Therefore, in an optional implementation of the embodiment of the present application, the first segmentation sub-model includes a first decoding layer and a first segmentation layer; the step C can, for example, include the following steps C1-C2:

[0099] Step C1: performing feature decoding on the first feature image through the first decoding layer to obtain a third feature image of the first lane line sample image; the third feature image has the same image size as the first lane line sample image;

[0100] Step C2: performing lane line segmentation on the third feature image through the first segmentation layer to obtain the first predicted lane line category.

[0101] Step D: Perform lane line segmentation on the second lane line sample image using the second segmentation sub-model to obtain a second predicted lane line category of the second lane line sample image.

[0102] Based on the specific implementation of step C, similarly, in an optional implementation of the embodiment of the present application, the second segmentation sub-model includes a second decoding layer and a second segmentation layer; and step D may include the following steps D1-D2:

[0103] Step D1: performing feature decoding on the second feature image through the second decoding layer to obtain a fourth feature image of the second lane line sample image; the fourth feature image has the same image size as the second lane line sample image;

[0104] Step D2: performing lane line segmentation on the fourth feature image through the second segmentation layer to obtain the second predicted lane line category.

[0105] The first decoding layer and the second decoding layer may adopt a decoder network similar to U-Net, for example; the first segmentation layer and the second segmentation layer may adopt a function similar to Argmax, for example.

[0106] Step E: According to the first predicted scene label, the first predicted lane line category, the common scene label, the first labeled lane line category, the first loss function and the second loss function, adjust the model parameters of the classification sub-model and the model parameters of the first segmentation sub-model until the training of the preset detection model is completed; the first loss function corresponds to the classification sub-model, and the second loss function corresponds to the first segmentation sub-model and the second segmentation sub-model.

[0107] Step F: Adjust the model parameters of the classification sub-model and the model parameters of the second segmentation sub-model according to the second predicted scene label, the second predicted lane line category, the line-crossing scene label, the second labeled lane line category, the first loss function and the second loss function until the preset detection model training is completed.

[0108] Among them, the classification sub-model, the first segmentation sub-model and the second segmentation sub-model can use the same cross entropy as the loss function. The first prediction scene label and the second prediction scene label output by the classification sub-model are recorded as The normal scene label (for example, 0) and the line-crossing scene label (for example, 1) are denoted as C, and the first predicted lane line category output by the first segmentation sub-model and the second predicted lane line category output by the second segmentation sub-model are denoted as The first labeled lane line category and the first labeled lane line category and the labeled lane line category equal to the first labeled lane line category are denoted as S. Then the first loss function and the second loss function are as follows:

[0109]

[0110] The loss function of the preset detection model can be expressed as:

[0111]

[0112] Step G: Determine the trained preset detection model as the lane line detection model.

[0113] Through the various implementation methods provided in this embodiment, the lane line sample image is divided into a first lane line sample image and a second lane line sample image according to whether the vehicle is in a lane pressing scene; the first lane line sample image has a normal scene label and a first annotated lane line category that does not include a lane pressing category, and the second lane line sample image has a lane pressing scene label and a second annotated lane line category that includes a lane pressing category; through the first lane line sample image and the second lane line sample image, a preset detection model including a classification sub-model, a first segmentation sub-model and a second segmentation sub-model is trained to obtain a lane line detection model; wherein, the classification sub-model predicts the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model predicts the lane line category of the first lane line sample image, and the second segmentation sub-model predicts the lane line category of the second lane line sample image.

[0114] This method distinguishes lane line sample images based on whether the vehicle is in a lane-crossing scenario, labeling them with different scene labels and lane line categories. A classification sub-model then discriminates whether the vehicle is in a lane-crossing scenario in the lane line sample images, using different first and second segmentation sub-models for model training. Based on this, the preset detection model's machine learning capabilities for the first lane line sample image in which the vehicle is not in a lane-crossing scenario are enhanced for the second lane line sample image in which the vehicle is in a lane-crossing scenario. The resulting lane line detection model is therefore applicable not only to lane line images in which the vehicle is not in a lane-crossing scenario, but also to lane line images in which the vehicle is in a lane-crossing scenario, thereby improving the detection accuracy of the lane line detection model.

[0115] Based on the lane line detection model obtained in the above embodiment, since the lane line detection model is not only applicable to lane line images of vehicles in non-line-pressing scenarios, but also applicable to lane line images of vehicles in line-pressing scenarios, after obtaining the lane line image to be detected without the lane line category marked, it is necessary to input the lane line image to be detected into the lane line detection model, and the predicted lane line category of the lane line image to be detected can be output as the third predicted lane line category. The third predicted lane line category of the lane line image to be detected obtained in this way is more accurate, laying the foundation for improving the subsequent functional services of autonomous driving. Next, the specific implementation method of the lane line detection method and device in the embodiment of the present application is described in detail through examples in combination with the accompanying drawings.

[0116] See also Figure 5 , shows a flow chart of another lane line detection method in an embodiment of the present application. In this embodiment, the method may include the following steps:

[0117] Step 501: Acquire an image of a lane line to be detected.

[0118] Step 502: Perform lane line detection on the lane line image to be detected using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected.

[0119] In the embodiment of the present application, the specific implementation method of step 502 is as follows: first input the lane line image to be detected into the classification sub-model for scene classification to obtain the third predicted scene label of the lane line image to be detected; then determine whether the third predicted scene label is a normal scene label or a line-pressing scene label, and determine whether the first segmentation sub-model corresponding to the normal scene label or the second segmentation sub-model corresponding to the line-pressing scene label performs lane line segmentation to obtain the third predicted lane line category of the lane line image to be detected. In this way, regardless of whether the lane line image to be detected is a vehicle in a non-line-pressing scene or a vehicle in a line-pressing scene, the lane line detection of the lane line image to be detected can be made more accurate. Therefore, in an optional implementation of the embodiment of the present application, the step 502 may, for example, include the following steps H-step J:

[0120] Step H: performing scene classification on the lane line image to be detected using the classification sub-model to obtain a third predicted scene label for the lane line image to be detected;

[0121] Step I: If the third predicted scene label is the common scene label, perform lane line segmentation on the lane line image to be detected using the first segmentation sub-model to obtain the third predicted lane line category;

[0122] Step J: If the third predicted scene label is the lane crossing scene label, perform lane line segmentation on the lane line image to be detected using the second segmentation sub-model to obtain the third predicted lane line category.

[0123] In addition, in the embodiment of the present application, considering that only one of the first segmentation sub-model and the second segmentation sub-model will be used at the same time for the lane line image to be detected, in order to improve the processing speed of lane line segmentation and thus improve the overall processing speed of lane line detection, the first segmentation sub-model and the second segmentation sub-model can be merged into one segmentation sub-model through a conditional convolution layer. Specifically, the conditional convolution layer ensures that when the third predicted scene label output by the classification sub-model is a normal scene label, the segmentation sub-model actually only uses the first segmentation sub-model corresponding to the normal scene label; when the third predicted scene label output by the classification sub-model is a line-crossing scene label, the segmentation sub-model actually only uses the second segmentation sub-model corresponding to the line-crossing scene label.

[0124] For example, the third predicted scene label output by the classification sub-model is a (a=0 represents a normal scene label, a=1 represents a line-pressing scene label), the input of the conditional convolution layer is X, the convolution kernel weight of the first segmentation sub-model is W0 and the bias is b0, the convolution kernel weight of the second segmentation sub-model is W1 and the bias is b1, then the output under the conditional convolution layer is Y=X×((1-a)W0+a W1)+((1-a)b0+a b1).

[0125] Based on this, the specific implementation of step 502 is as follows: first, the lane line image to be detected is input into the classification sub-model for scene classification to obtain the third predicted scene label of the lane line image to be detected; then, based on the third predicted scene label, the lane line image to be detected is input into the segmentation sub-model for lane line segmentation, and the third predicted lane line category of the lane line image to be detected can be output. For example, Figure 6 A schematic diagram of a lane detection framework is shown.

[0126] Therefore, in an optional implementation of the embodiment of the present application, the method may further include, for example, step K: merging the conditional convolution layer, the first segmentation sub-model, and the second segmentation sub-model to obtain a segmentation sub-model; when the conditional convolution layer indicates that the third predicted scene label is the normal scene label, the segmentation sub-model is the first segmentation sub-model, and when the third predicted scene label is the pressure line scene label, the segmentation sub-model is the second segmentation sub-model; then, step 502 may, for example, include the following steps L-M:

[0127] Step L: performing scene classification on the lane line image to be detected using the classification sub-model to obtain a third predicted scene label for the lane line image to be detected;

[0128] Step M: Based on the third predicted scene label, perform lane line segmentation on the lane line image to be detected through the segmentation sub-model to obtain the third predicted lane line category.

[0129] Through the various implementation methods provided in this embodiment, the lane line image to be detected is input into the above-mentioned lane line detection model for lane line detection to obtain a third predicted lane line category of the lane line image to be detected; because the above-mentioned lane line detection model has a higher detection accuracy when processing lane line images to be detected without labeled lane line categories, the third predicted lane line category of the lane line image to be detected is more accurate, thereby improving the subsequent functional services of autonomous driving.

[0130] Based on the above description, if Figure 7 The following diagram illustrates a lane detection model training and lane detection framework. Model design yields a preset detection model. Image acquisition and annotation yield first and second lane line sample images. The preset detection model is trained on the first and second lane line sample images to yield a lane detection model. The lane line image to be detected is input into the lane detection model for lane detection, resulting in a predicted lane line category.

[0131] Exemplary devices

[0132] The specific implementation of the lane line detection model training device and the lane line detection device in the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0133] See also Figure 8 , shows a schematic structural diagram of a lane detection model training device in an embodiment of the present application. In this embodiment, the device may specifically include: a first acquisition unit 801 and a training unit 802;

[0134] The first acquisition unit 801 is configured to acquire a first lane line sample image in a non-lane-crossing scenario and a second lane line sample image in a lane-crossing scenario; the first lane line sample image includes a normal scene label and a first annotated lane line category, the second lane line sample image includes a lane-crossing scene label and a second annotated lane line category, the first annotated lane line category does not include a lane-crossing category, and the second annotated lane line category includes the lane-crossing category.

[0135] The training unit 802 is used to train a preset detection model to obtain a lane line detection model based on the first lane line sample image and the second lane line sample image; the preset detection model includes a classification sub-model, a first segmentation sub-model and a second segmentation sub-model, the classification sub-model is used to predict the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model is used to predict the lane line category of the first lane line sample image, and the second segmentation sub-model is used to predict the lane line category of the second lane line sample image.

[0136] In an optional implementation of the embodiment of the present application, the training unit 802 includes a first classification subunit, a second classification subunit, a first segmentation subunit, a second segmentation subunit, a first adjustment subunit, a second adjustment subunit, and a determination subunit:

[0137] The first classification subunit is configured to perform scene classification on the first lane line sample image using the classification submodel to obtain a first predicted scene label for the first lane line sample image;

[0138] The second classification subunit is configured to perform scene classification on the second lane line sample image using the classification submodel to obtain a second predicted scene label for the second lane line sample image;

[0139] The first segmentation sub-unit is configured to perform lane line segmentation on the first lane line sample image using the first segmentation sub-model to obtain a first predicted lane line category of the first lane line sample image;

[0140] The second segmentation sub-unit is configured to perform lane line segmentation on the second lane line sample image using the second segmentation sub-model to obtain a second predicted lane line category of the second lane line sample image;

[0141] The first adjustment sub-unit is configured to adjust the model parameters of the classification sub-model and the model parameters of the first segmentation sub-model according to the first predicted scene label, the first predicted lane line category, the common scene label, the first annotated lane line category, a first loss function, and a second loss function until the training of the preset detection model is completed; the first loss function corresponds to the classification sub-model, and the second loss function corresponds to the first segmentation sub-model and the second segmentation sub-model;

[0142] The second adjustment sub-unit is configured to adjust the model parameters of the classification sub-model and the model parameters of the second segmentation sub-model according to the second predicted scene label, the second predicted lane line category, the lane crossing scene label, the second annotated lane line category, the first loss function, and the second loss function until the training of the preset detection model is completed;

[0143] The determination subunit is used to determine the trained preset detection model as the lane line detection model.

[0144] In an optional implementation of the embodiment of the present application, the classification sub-model includes a feature extraction layer and a classifier; the first classification sub-unit includes a first feature extraction module and a first classification module;

[0145] The first feature extraction module is configured to perform feature extraction on the first lane line sample image through the feature extraction layer to obtain a first feature image of the first lane line sample image;

[0146] The first classification module is configured to perform scene classification on the first feature image using the classifier to obtain the first predicted scene label;

[0147] The second classification subunit includes a second feature extraction module and a second classification module;

[0148] The second feature extraction module is configured to perform feature extraction on the second lane line sample image through the feature extraction layer to obtain a second feature image of the second lane line sample image;

[0149] The second classification module is used to perform scene classification on the second feature image through the classifier to obtain the second predicted scene label.

[0150] In an optional implementation of the embodiment of the present application, the first segmentation sub-model includes a first decoding layer and a first segmentation layer; the first segmentation sub-unit includes a first feature decoding module and a first segmentation module;

[0151] The first feature decoding module is configured to perform feature decoding on the first feature image through the first decoding layer to obtain a third feature image of the first lane line sample image; the third feature image has the same image size as the first lane line sample image;

[0152] The first segmentation module is configured to perform lane line segmentation on the third feature image through the first segmentation layer to obtain the first predicted lane line category.

[0153] In an optional implementation of the embodiment of the present application, the second segmentation sub-model includes a second decoding layer and a second segmentation layer; the second segmentation sub-unit includes a second feature decoding module and a second segmentation module;

[0154] The second feature decoding module is configured to perform feature decoding on the second feature image through the second decoding layer to obtain a fourth feature image of the second lane line sample image; the fourth feature image has the same image size as the second lane line sample image;

[0155] The second segmentation module is used to perform lane line segmentation on the fourth feature image through the second segmentation layer to obtain the second predicted lane line category.

[0156] Through the various implementation methods provided in this embodiment, the lane line sample image is divided into a first lane line sample image and a second lane line sample image according to whether the vehicle is in a lane pressing scene; the first lane line sample image has a normal scene label and a first annotated lane line category that does not include a lane pressing category, and the second lane line sample image has a lane pressing scene label and a second annotated lane line category that includes a lane pressing category; through the first lane line sample image and the second lane line sample image, a preset detection model including a classification sub-model, a first segmentation sub-model and a second segmentation sub-model is trained to obtain a lane line detection model; wherein, the classification sub-model predicts the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model predicts the lane line category of the first lane line sample image, and the second segmentation sub-model predicts the lane line category of the second lane line sample image.

[0157] The device distinguishes lane line sample images based on whether the vehicle is in a lane-crossing scenario, labeling them with different scene labels and lane line categories. A classification sub-model determines whether the vehicle is in a lane-crossing scenario in the lane line sample images, and then uses different first segmentation sub-models and second segmentation sub-models for model training. Based on this, the preset detection model's machine learning capabilities for the second lane line sample image in a lane-crossing scenario are enhanced, building on the machine learning of the preset detection model on the first lane line sample image in which the vehicle is not in a lane-crossing scenario. The trained lane line detection model is therefore applicable not only to lane line images in which the vehicle is in a lane-crossing scenario, but also to lane line images in which the vehicle is in a lane-crossing scenario, thereby improving the detection accuracy of the lane line detection model.

[0158] See also Figure 9 , shows a schematic structural diagram of a lane detection model training device in an embodiment of the present application. In this embodiment, the device may specifically include: a second acquisition unit 901 and a detection unit 902;

[0159] The second acquisition unit 901 is used to acquire the lane line image to be detected;

[0160] The detection unit 902 is configured to perform lane line detection on the lane line image to be detected using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected;

[0161] The lane line detection model is the lane line detection model described in the above method embodiment.

[0162] In an optional implementation of the embodiment of the present application, the detection unit 902 includes a third classification subunit, a third segmentation subunit and a fourth segmentation subunit;

[0163] The third classification subunit is configured to perform scene classification on the lane line image to be detected using the classification submodel to obtain a third predicted scene label for the lane line image to be detected;

[0164] The third segmentation sub-unit is configured to, if the third predicted scene label is the common scene label, perform lane line segmentation on the lane line image to be detected using the first segmentation sub-model to obtain the third predicted lane line category;

[0165] The fourth segmentation sub-unit is used to perform lane line segmentation on the lane line image to be detected using the second segmentation sub-model to obtain the third predicted lane line category if the third predicted scene label is the lane crossing scene label.

[0166] In an optional implementation of the embodiment of the present application, the method further includes a merging unit:

[0167] The merging unit is configured to merge the conditional convolution layer, the first segmentation sub-model, and the second segmentation sub-model to obtain a segmentation sub-model; when the conditional convolution layer indicates that the third predicted scene label is the normal scene label, the segmentation sub-model is the first segmentation sub-model; when the third predicted scene label is the line-pressing scene label, the segmentation sub-model is the second segmentation sub-model;

[0168] The detection unit 902 includes a fourth classification subunit and a fifth segmentation subunit;

[0169] The fourth classification subunit is configured to perform scene classification on the lane line image to be detected using the classification submodel to obtain a third predicted scene label for the lane line image to be detected;

[0170] The fifth segmentation sub-unit is configured to perform lane line segmentation on the lane line image to be detected based on the third predicted scene label using the segmentation sub-model to obtain the third predicted lane line category.

[0171] Through the various implementation methods provided in this embodiment, the lane line image to be detected is input into the above-mentioned lane line detection model for lane line detection to obtain a third predicted lane line category of the lane line image to be detected; because the above-mentioned lane line detection model has a higher detection accuracy when processing lane line images to be detected without labeled lane line categories, the third predicted lane line category of the lane line image to be detected is more accurate, thereby improving the subsequent functional services of autonomous driving.

[0172] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0173] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0174] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements that are inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0175] The above description is only a preferred embodiment of the present application and does not constitute any formal limitation to the present application. Although the present application has been disclosed as above with preferred embodiments, it is not intended to limit the present application. Any technician familiar with the art can use the above-disclosed methods and technical contents to make many possible changes and modifications to the technical solution of the present application without departing from the scope of the technical solution of the present application, or modify it into an equivalent embodiment with equivalent changes. Therefore, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application still falls within the scope of protection of the technical solution of the present application.

Claims

1. A lane detection model training method, characterized in that: include: Acquire a first lane line sample image in a non-lane-crossing scenario and a second lane line sample image in a lane-crossing scenario; the first lane line sample image includes a normal scene label and a first annotated lane line category, the second lane line sample image includes a lane-crossing scene label and a second annotated lane line category, the first annotated lane line category does not include a lane-crossing category, and the second annotated lane line category includes the lane-crossing category; According to the first lane line sample image and the second lane line sample image, a preset detection model is trained to obtain a lane line detection model; the preset detection model includes a classification sub-model, a first segmentation sub-model and a second segmentation sub-model, the classification sub-model is used to predict the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model is used to predict the lane line category of the first lane line sample image, and the second segmentation sub-model is used to predict the lane line category of the second lane line sample image, and the classification sub-model is respectively connected to the first segmentation sub-model and the second segmentation sub-model.

2. The method according to claim 1, characterized in that The step of training a preset detection model to obtain a lane line detection model based on the first lane line sample image and the second lane line sample image includes: Performing scene classification on the first lane line sample image using the classification sub-model to obtain a first predicted scene label for the first lane line sample image; Performing scene classification on the second lane line sample image using the classification sub-model to obtain a second predicted scene label for the second lane line sample image; Performing lane line segmentation on the first lane line sample image using the first segmentation sub-model to obtain a first predicted lane line category of the first lane line sample image; Performing lane line segmentation on the second lane line sample image using the second segmentation sub-model to obtain a second predicted lane line category for the second lane line sample image; Adjusting model parameters of the classification sub-model and model parameters of the first segmentation sub-model according to the first predicted scene label, the first predicted lane line category, the common scene label, the first annotated lane line category, a first loss function, and a second loss function until training of the preset detection model is completed; the first loss function corresponds to the classification sub-model, and the second loss function corresponds to the first segmentation sub-model and the second segmentation sub-model; Adjusting the model parameters of the classification sub-model and the model parameters of the second segmentation sub-model according to the second predicted scene label, the second predicted lane line category, the lane crossing scene label, the second labeled lane line category, the first loss function, and the second loss function until the preset detection model training is completed; The preset detection model that has been trained is determined as the lane line detection model.

3. The method according to claim 2, characterized in that The classification sub-model includes a feature extraction layer and a classifier; performing scene classification on the first lane line sample image by the classification sub-model to obtain a first predicted scene label for the first lane line sample image includes: Performing feature extraction on the first lane line sample image by the feature extraction layer to obtain a first feature image of the first lane line sample image; Performing scene classification on the first feature image by the classifier to obtain the first predicted scene label; The performing scene classification on the second lane line sample image by the classification sub-model to obtain a second predicted scene label for the second lane line sample image includes: Performing feature extraction on the second lane line sample image by the feature extraction layer to obtain a second feature image of the second lane line sample image; The second feature image is subjected to scene classification by the classifier to obtain the second predicted scene label.

4. The method according to claim 3, characterized in that The first segmentation sub-model includes a first decoding layer and a first segmentation layer; performing lane line segmentation on the first lane line sample image using the first segmentation sub-model to obtain a first predicted lane line category of the first lane line sample image includes: Performing feature decoding on the first feature image through the first decoding layer to obtain a third feature image of the first lane line sample image; the third feature image has the same image size as the first lane line sample image; Lane line segmentation is performed on the third feature image through the first segmentation layer to obtain the first predicted lane line category.

5. The method according to claim 3, characterized in that The second segmentation sub-model includes a second decoding layer and a second segmentation layer; performing lane line segmentation on the second lane line sample image using the second segmentation sub-model to obtain a second predicted lane line category of the second lane line sample image includes: Performing feature decoding on the second feature image through the second decoding layer to obtain a fourth feature image of the second lane line sample image; the fourth feature image has the same image size as the second lane line sample image; Lane line segmentation is performed on the fourth feature image through the second segmentation layer to obtain the second predicted lane line category.

6. A lane line detection method, characterized in that: include: Obtain the lane line image to be detected; Performing lane line detection on the lane line image to be detected using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected; The lane line detection model is the lane line detection model according to any one of claims 1 to 5.

7. The method according to claim 6, characterized in that The performing lane line detection on the lane line image to be detected by using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected includes: Performing scene classification on the lane line image to be detected by the classification sub-model to obtain a third predicted scene label of the lane line image to be detected; If the third predicted scene label is the common scene label, performing lane line segmentation on the lane line image to be detected using the first segmentation sub-model to obtain the third predicted lane line category; If the third predicted scene label is the lane line crossing scene label, lane line segmentation is performed on the lane line image to be detected using the second segmentation sub-model to obtain the third predicted lane line category.

8. The method according to claim 7, characterized in that The method further comprises: Merging the conditional convolution layer, the first segmentation sub-model, and the second segmentation sub-model to obtain a segmentation sub-model; when the conditional convolution layer indicates that the third predicted scene label is the normal scene label, the segmentation sub-model is the first segmentation sub-model; when the third predicted scene label is the line-pressing scene label, the segmentation sub-model is the second segmentation sub-model; The performing lane line detection on the lane line image to be detected by using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected includes: Performing scene classification on the lane line image to be detected by the classification sub-model to obtain a third predicted scene label of the lane line image to be detected; Based on the third predicted scene label, lane line segmentation is performed on the lane line image to be detected through the segmentation sub-model to obtain the third predicted lane line category.

9. A lane detection model training device, characterized in that: include: a first acquisition unit and a training unit; The first acquisition unit is configured to acquire a first lane line sample image in a non-lane-crossing scenario and a second lane line sample image in a lane-crossing scenario; the first lane line sample image includes a normal scene label and a first annotated lane line category, the second lane line sample image includes a lane-crossing scene label and a second annotated lane line category, the first annotated lane line category does not include a lane-crossing category, and the second annotated lane line category includes the lane-crossing category; The training unit is used to train a preset detection model to obtain a lane line detection model based on the first lane line sample image and the second lane line sample image; the preset detection model includes a classification sub-model, a first segmentation sub-model and a second segmentation sub-model, the classification sub-model is used to predict the scene label of the first lane line sample image and the scene label of the second lane line sample image, the first segmentation sub-model is used to predict the lane line category of the first lane line sample image, and the second segmentation sub-model is used to predict the lane line category of the second lane line sample image, and the classification sub-model is respectively connected to the first segmentation sub-model and the second segmentation sub-model.

10. A lane line detection device, characterized in that: include: a second acquisition unit and a detection unit; The second acquisition unit is used to acquire the lane line image to be detected; The detection unit is configured to perform lane line detection on the lane line image to be detected using a lane line detection model to obtain a third predicted lane line category of the lane line image to be detected; The lane line detection model is the lane line detection model according to any one of claims 1 to 5.

Images