Method for training drivable area detection model, method for detecting driving area

Abstract

The application provides a training method of a drivable area detection model and a driving area detection method. The driving area detection method comprises the following steps: acquiring image data, wherein the image data comprises road images collected by a target vehicle; inputting the image data into a pre-trained drivable area detection model to determine a driving area of the target vehicle in the image data. Training data of the drivable area detection model and a discrimination model comprises a set of labeled image samples and a set of unlabeled image samples. According to the embodiment of the application, the trained discrimination model can be used to assist the training of the drivable area detection model, and only a part of the training data needs to be labeled. The model can be trained by using labeled and unlabeled data at the same time, thereby reducing the labeling cost.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular to a training method for a drivable area detection model and a method for detecting drivable areas. Background Technology

[0002] In recent years, vision-based Advanced Driving Assistance Systems (ADAS) have developed rapidly, and driving safety issues require continuous improvement in vision technology. One of the key functions of ADAS is drivable area detection, which provides information about the drivable area around the vehicle. The accuracy of the drivable area edges largely determines the safety and comfort of autonomous driving.

[0003] Mainstream drivable area detection methods are not effective at detecting the edges of drivable areas; on the other hand, the model training on which detection relies is difficult and costly in terms of sample labeling, which increases the R&D cost of ADAS. Summary of the Invention

[0004] This application provides a training method for a drivable area detection model and a method for detecting drivable areas.

[0005] In a first aspect, embodiments of this application provide a training method for a drivable area detection model, which may include:

[0006] The labeled image samples from the labeled image sample set are input into the drivable area detection model to be trained to obtain the first prediction result of the drivable area;

[0007] The first prediction result, the labeled image sample, and the labeling result of the labeled image sample are input into the discrimination model to calculate the first loss function;

[0008] Input the unlabeled image samples from the unlabeled image sample set into the drivable area detection model to be trained to obtain the second prediction result of the drivable area;

[0009] The second prediction result and the unlabeled image samples are input into the discrimination model to calculate the second loss function;

[0010] The parameters of the drivable area detection model to be trained are iteratively updated using the first loss function and the second loss function.

[0011] Secondly, embodiments of this application provide a method for detecting a driving area, which may include:

[0012] Acquire image data, including road images captured by the target vehicle;

[0013] Image data is input into a pre-trained drivable area detection model to determine the driving area of ​​the target vehicle in the image data;

[0014] The drivable area detection model is trained using the training method for the drivable area detection model provided in the first aspect.

[0015] Compared with the prior art, this application has the following advantages:

[0016] According to the embodiments of this application, a pre-trained discriminative model can be used to assist in the training of a drivable area detection model, and only a portion of the training data needs to be labeled. The model can be trained using both labeled and unlabeled data simultaneously, reducing labeling costs.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application, it can be implemented according to the contents of the specification. In order to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0018] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments according to this application and should not be construed as limiting the scope of this application.

[0019] Figure 1 A flowchart of the method for detecting the drivable area provided in this application;

[0020] Figure 2 This is a schematic diagram illustrating the principle of a method for detecting a drivable area according to an embodiment of this application;

[0021] Figure 3 This is a flowchart of a training method for a drivable area detection model according to an embodiment of this application;

[0022] Figure 4 This is a schematic diagram illustrating the working principle of a discrimination model according to an embodiment of this application; and

[0023] Figure 5 This is a block diagram of an electronic device used to implement embodiments of this application. Detailed Implementation

[0024] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the concept or scope of this application. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0025] To facilitate understanding of the technical solutions of the embodiments of this application, the relevant technologies of the embodiments of this application are described below. The following relevant technologies are optional solutions and can be combined with the technical solutions of the embodiments of this application in any way, and all of them fall within the protection scope of the embodiments of this application.

[0026] This application provides a method for detecting drivable areas, such as... Figure 1 The diagram shown is a flowchart of a method for detecting a driving area according to an embodiment of this application, which may include:

[0027] Step S101: Acquire image data, which includes road images captured by the target vehicle;

[0028] Step S102: Input the image data into the pre-trained drivable area detection model to determine the driving area of ​​the target vehicle in the image data.

[0029] The execution entity in this application embodiment can be a vehicle-mounted system or a cloud server communicating with the vehicle-mounted system. The image data can be road images captured by the target vehicle. Generally, during the vehicle's operation, video can be captured by an image acquisition device, and the video can be decoded to obtain continuous frame images. Therefore, continuous frame images can be used as the acquired image data.

[0030] The processing procedure is the same for each frame of image data; an example will be used to illustrate this. (Combined with...) Figure 2 As shown, after image data is processed by the convolutional layer, max pooling layer (downsampling), and deconvolutional layer (upsampling) of the drivable region detection model, pixel-level image segmentation results can be obtained. In the image segmentation results, different colors can be used to distinguish regions; for example, roads can be represented by red areas, the sky by brown areas, and buildings by blue areas.

[0031] Different results can be obtained from image segmentation for different tasks. For example, in a driving area detection method, the drivable area represented by the red area can be used as the detection result. For instance, in a low-speed automatic parking scenario, the drivable area can be used for obstacle avoidance and parking space validity verification. In a high-speed scenario, path planning can be performed by combining lane detection, vehicle detection, and other methods to improve safe driving performance. The colors mentioned above represent what can be observed in one implementation method. In actual implementation, other colors may be used to distinguish different recognition results, which are not specifically limited here. The accompanying drawings in the embodiments of this application are for illustrative purposes only.

[0032] In this current implementation, unlike conventional techniques that require labeling all training data before training the drivable area detection model, a pre-trained discriminant model can be used to assist in training the drivable area detection model, and only a portion of the training data needs to be labeled. The model can be trained using both labeled and unlabeled data simultaneously, reducing labeling costs.

[0033] This application provides a training method for a drivable area detection model, such as... Figure 3 The diagram shown is a flowchart of a training method for a drivable area detection model according to an embodiment of this application, which may include:

[0034] In one optional implementation, the training process of the drivable area detection model includes:

[0035] Step S201: Input the labeled image samples from the labeled image sample set into the drivable area detection model to be trained to obtain the first prediction result of the drivable area.

[0036] Step S202: Input the first prediction result, the labeled image sample, and the labeling result of the labeled image sample into the discrimination model, and calculate the first loss function.

[0037] Step S203: Input the unlabeled image samples from the unlabeled image sample set into the drivable area detection model to be trained, and obtain the second prediction result of the drivable area.

[0038] Step S204: Input the second prediction result and the unlabeled image samples into the discrimination model and calculate the second loss function.

[0039] Step S205: Iteratively update the parameters of the drivable area detection model to be trained using the first loss function and the second loss function.

[0040] The set of labeled image samples can be represented as: {(X1, Y1), (X2, Y2), ..., (X... i Y i ), ..., (X) m Y m )}。 X i Y can represent the i-th labeled image sample. i This can represent the annotation result of the i-th labeled image sample. Inputting the labeled image samples from the set of labeled image samples into the drivable region detection model to be trained, the first prediction result of the drivable region can be represented as G(X). i ).

[0041] The first prediction result, the labeled image samples, and the annotation results of the labeled image samples are input into the discriminative model to calculate the first loss function. For example, the first loss function can be the difference between the first prediction result and the annotation result in the driving region dimension, the difference between the first prediction result and the annotation result in the driving region boundary dimension, or the difference between the first prediction result and the annotation result in the channel dimension. The training process of the discriminative model will be described in detail later.

[0042] The set of unlabeled image samples can be represented as: {Z1, Z2, ..., Z...} i ... Z m}. m can represent the number of unlabeled image samples. Z i This can be represented by the i-th unlabeled image sample. Inputting the unlabeled image samples from the set of unlabeled images into the drivable region detection model to be trained, the second prediction result of the drivable region can be represented as G(Z). i ).

[0043] The second loss function can be expressed as log(1-D(Zi, G(Zi))). Here, G(Zi) represents the driving region prediction result of the drivable region detection model to be trained on the i-th unlabeled image sample, which can be the result of concatenation along the channel dimension. D(Zi, G(Zi)) represents the discrimination model's judgment on the driving region prediction result of the i-th unlabeled image sample. If the output of the discrimination model to be trained is a "non-realistic labeling result," it indicates that the discrimination model's judgment is correct. Conversely, if the output of the discrimination model to be trained is a "realistic labeling result," it indicates that the discrimination model's judgment is incorrect, and the parameters of the discrimination model to be trained can be iteratively updated based on this. That is, the closer the output of the discrimination model to be trained is to a "non-realistic labeling result," the better the performance.

[0044] The iterative update of the parameters of the drivable region detection model to be trained, using both the first and second loss functions, can be expressed as follows: as well as In the formula, L3 can represent the first loss function. The calculation process of the first loss function will be described in detail later.

[0045] In an optional implementation, step S202, which involves inputting the first prediction result, the labeled image samples, and the annotation results of the labeled image samples into the discrimination model to calculate the first loss function, may include:

[0046] Step S2021: Input the first prediction result and the labeled image samples into the discriminant model and calculate the log loss function.

[0047] Step S2022: Input the first prediction result and the annotation result of the labeled image sample into the discrimination model, and calculate the cross-entropy loss function and the edge-aware loss function.

[0048] Step S2023: Use the logarithmic loss function, cross-entropy loss function, and edge-aware loss function as the first loss function.

[0049] The logarithmic loss function can be expressed as log(1-D(Xi, G(Xi))). Here, G(Xi) represents the driving region prediction result of the drivable region detection model to be trained on the i-th labeled image sample, which can be the result of concatenation along the channel dimension. D(Xi, G(Xi)) represents the discrimination model's judgment of the driving region prediction result on the i-th labeled image sample.

[0050] If the output of the discriminant model to be trained is a "true labeling result", it means that the discrimination of the discriminant model to be trained is correct. Conversely, if the output of the discriminant model to be trained is a "non-true labeling result", it means that the discrimination of the discriminant model to be trained is incorrect, and the parameters of the discriminant model to be trained need to be iteratively updated based on this. That is, the closer the output of the discriminant model is to the "true labeling result", the better the effect. Using the gradient descent principle, the log loss function is expressed as log(1-D(Xi, G(Xi))) instead of log(D(Xi, G(Xi))).

[0051] The cross-entropy loss function can be expressed as L ce (Yi, G(Xi)). This cross-entropy loss function compares the driving region prediction result of the drivable region detection model to be trained on the i-th labeled image sample with the labeled result. The higher the overlap between the predicted result and the labeled result, the better the prediction result. The prediction result can be a pixel classification, such as classifying pixels into drivable regions, curbs, etc.

[0052] The edge-aware loss function can be expressed as L ba (Yi, G(Xi)). This edge-aware loss function compares the drivable region prediction result of the drivable region detection model to be trained for the i-th labeled image sample with the label result. The higher the overlap between the predicted result and the label result, the better the prediction result.

[0053] Based on this, the first loss function can be expressed as:

[0054]

[0055] In one alternative implementation, the target value for the logarithmic loss function, the cross-entropy loss function, and the edge-aware loss function is 1.

[0056] Since the first loss function is trained on labeled images, the closer the model's output is to the labeled results, the better. Therefore, the target values ​​for the logarithmic loss function, cross-entropy loss function, and edge-aware loss function are all set to 1.

[0057] In one alternative implementation, the objective value of the second loss function is 0.

[0058] The second loss function is trained based on unlabeled image samples, so it needs to have a target value of 0.

[0059] In this application, the drivable area detection model and the discriminative model can be trained simultaneously. A Nash equilibrium is reached in the mid-to-late stages of model training, leading to model convergence. Unlabeled image samples can be considered as data for inference in actual deployment. Later in training, the discriminative model becomes unable to distinguish between true and false outputs from the drivable area detection model, meaning the output quality of the drivable area detection model approaches the quality of manually labeled samples. Therefore, a large number of samples can still be used for model training, while only a small portion needs to be labeled. Simultaneous training using both labeled and unlabeled samples reduces labeling costs. Furthermore, no complex attention module is needed; edge detection performance can be improved simply by constructing an edge-aware loss function constraint algorithm, significantly reducing the difficulty of porting and facilitating deployment.

[0060] In one optional implementation, the method further includes: displaying the driving area of ​​the target vehicle, including at least one of image display and voice broadcast display.

[0061] After identifying the driving areas in the image data, they can be displayed. Taking image display as an example, driving areas can be shown in green, and non-driving areas in yellow. Taking voice broadcast display as an example, the driving areas can be announced according to the specific scenario. For example, "The road leading to the right is 100 meters ahead."

[0062] In one alternative implementation, the training process of the discriminant model may include:

[0063] Step S301: Input the labeled data pairs from the labeled image sample set into the discrimination model to be trained, and calculate the third loss function; the labeled data pairs are the labeled image samples and the labeling results of the labeled image samples.

[0064] Step S302: Input the unlabeled data pairs of the unlabeled image sample set into the discrimination model to be trained, and calculate the fourth loss function; the unlabeled data pairs are the unlabeled image samples and the prediction results of the drivable area detection model to be trained on the unlabeled image samples;

[0065] Step S303: Use the third and fourth loss functions to iteratively update the parameters of the discrimination model to be trained.

[0066] The set of labeled image samples can be represented as: {(X1, Y1), (X2, Y2), ..., (X... i Y i ), ..., (X) m Y m )}. m can represent the number of labeled image samples, where m is a positive integer. X i Y can represent the i-th labeled image sample. i This can represent the annotation result of the i-th labeled image sample, where 1 ≤ i ≤ m, and i is a positive integer. The annotation result can be the result of concatenation along the channel dimension. Combined with... Figure 4 The diagram illustrates the working principle of the discriminative model. Here, (I, gt) corresponds to the content in the labeled image sample set, I corresponds to the labeled image sample, and gt corresponds to the annotation result of the labeled image sample. (I, O) represents the driving region detection model predicting the drivable region of the labeled image sample I, with the output prediction result being O. The discriminative model then makes a judgment based on the prediction result O and the annotation result gt.

[0067] The third loss function can be expressed as logD(X) i Y i In the formula, D() represents the discriminant model. The labeled image samples and the annotation results of the i-th labeled image sample are input into the discriminant model to be trained. The discriminant model to be trained needs to determine whether the annotation result of the i-th labeled image sample is a true annotation result. If the output of the discriminant model to be trained is a "true annotation result," it means that the discrimination of the discriminant model to be trained is correct. Conversely, if the output of the discriminant model to be trained is a "non-true annotation result," it means that the discrimination of the discriminant model to be trained is incorrect, and the parameters of the discriminant model to be trained need to be iteratively updated based on this. That is, the closer the output of the discriminant model to be trained is to the "true annotation result," the better the effect.

[0068] The set of unlabeled image samples can be represented as: {Z1, Z2, ..., Z...} i ... Z m}. m can represent the number of unlabeled image samples, where m is a positive integer. Z iIt can represent the i-th unlabeled image sample.

[0069] The fourth loss function can be expressed as log(1-D(Zi, G(Zi))). Here, G() represents the drivable region detection model to be trained. G(Zi) represents the drivable region prediction result of the drivable region detection model for the i-th unlabeled image sample, which can be the result of concatenation along the channel dimension. D(Zi, G(Zi)) represents the discrimination model's judgment of the drivable region prediction result for the i-th unlabeled image sample. If the output of the discrimination model to be trained is a "non-realistic labeling result," it indicates that the discrimination model's judgment is correct. Conversely, if the output of the discrimination model to be trained is a "realistic labeling result," it indicates that the discrimination model's judgment is incorrect, and the parameters of the discrimination model to be trained need to be iteratively updated accordingly. That is, the closer the output of the discrimination model to be trained is to the "non-realistic labeling result," the better the effect.

[0070] Using the third and fourth loss functions, the iterative update of the parameters of the discriminative model to be trained can be expressed as: When both the third and fourth loss functions converge, the training process of the discriminant network is considered complete.

[0071] In one alternative implementation, the objective value of the third loss function is 1, and the objective value of the fourth loss function is 0.

[0072] For the third loss function, the input to the discriminant model to be trained is the labeled image samples and the labeling results of the i-th labeled image sample. Therefore, the expected value is that the discrimination model's judgment result is the "true labeling result". In mathematical terms, when the target value of the third loss function is 1 or close to 1, it means that the discrimination model's judgment result is the "true labeling result". Therefore, when training using the third loss function, if the output of the discriminant model to be trained is not 1, for example, the output is 0 or 0.5, the parameters in the discriminant model can be iteratively adjusted based on the difference until the output of the discriminant model to be trained is 1 or close to 1.

[0073] For the fourth loss function, the input to the discriminant model to be trained is the unlabeled image sample and the result of the driving region prediction of the i-th unlabeled image sample by the driving region detection model to be trained. Therefore, the expected value is that the discrimination model's judgment result is a "non-true labeling result". In mathematical terms, when the target value of the fourth loss function is 0 or close to 0, it means that the discrimination model's judgment result is a "non-true labeling result". If the output of the discriminant model to be trained is not 0, for example, the output is 1 or 0.5, the parameters in the discriminant model can be iteratively adjusted based on the difference until the output of the discriminant model to be trained is 0 or close to 0.

[0074] Figure 5 This is a block diagram of an electronic device used to implement embodiments of this application. For example... Figure 5 As shown, the electronic device includes a memory 510 and a processor 520. The memory 510 stores a computer program that can run on the processor 520. When the processor 520 executes the computer program, it implements the method described in the above embodiments. The number of memories 510 and processors 520 can be one or more.

[0075] The electronic device also includes:

[0076] The communication interface 530 is used to communicate with external devices and exchange and transmit data.

[0077] If the memory 510, processor 520, and communication interface 530 are implemented independently, they can be interconnected via a bus to communicate with each other. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0078] Optionally, in a specific implementation, if the memory 510, processor 520, and communication interface 530 are integrated on a single chip, then the memory 510, processor 520, and communication interface 530 can communicate with each other through an internal interface.

[0079] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method provided in this application.

[0080] This application also provides a chip including a processor for calling and executing instructions stored in a memory, causing a communication device with the chip installed to perform the method provided in this application.

[0081] This application also provides a chip, including: an input interface, an output interface, a processor, and a memory. The input interface, output interface, processor, and memory are connected through an internal connection path. The processor is used to execute code in the memory. When the code is executed, the processor is used to execute the method provided in the application embodiment.

[0082] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. General-purpose processors can be microprocessors or any conventional processor. It is worth noting that the processor can be a processor supporting the Advanced Reduced Instruction Set Computing (ARM) architecture.

[0083] Further, optionally, the aforementioned memory may include read-only memory and random access memory. The memory may be volatile memory or non-volatile memory, or may include both. Non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available. Examples include Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DRRAM).

[0084] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0085] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0086] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0087] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0088] The logic and / or steps described in the flowchart or otherwise herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0089] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. All or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware, the program being stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiments.

[0090] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. This storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0091] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope described in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A training method for a drivable area detection model, characterized in that, include: The labeled image samples from the labeled image sample set are input into the drivable area detection model to be trained to obtain the first prediction result of the drivable area; The process involves inputting the first prediction result, the labeled image sample, and the annotation result of the labeled image sample into a discriminant model to calculate a first loss function, including: inputting the first prediction result and the labeled image sample into the discriminant model to calculate a logarithmic loss function; inputting the first prediction result and the annotation result of the labeled image sample into the discriminant model to calculate a cross-entropy loss function and an edge-aware loss function; and using the logarithmic loss function, the cross-entropy loss function, and the edge-aware loss function as the first loss function. Input the unlabeled image samples from the unlabeled image sample set into the drivable area detection model to be trained to obtain the second prediction result of the drivable area; The second prediction result and the unlabeled image sample are input into the discrimination model to calculate the second loss function; The parameters of the drivable area detection model to be trained are iteratively updated using the first loss function and the second loss function. The training process of the discriminative model includes: The labeled data pairs from the labeled image sample set are input into the discrimination model to be trained, and a third loss function is calculated; the labeled data pairs are the labeled image samples and the labeling results of the labeled image samples. The unlabeled data pairs of the unlabeled image sample set are input into the discrimination model to be trained, and the fourth loss function is calculated; the unlabeled data pairs are the unlabeled image samples and the prediction results of the drivable area detection model to be trained on the unlabeled image samples; The parameters of the discrimination model to be trained are iteratively updated using the third loss function and the fourth loss function.

2. The method according to claim 1, characterized in that, The target value of the logarithmic loss function, the cross-entropy loss function, and the edge-aware loss function is 1.

3. The method according to claim 1, characterized in that, The objective value of the second loss function is 0.

4. The method according to claim 1, characterized in that, The objective value of the third loss function is 1, and the objective value of the fourth loss function is 0.

5. A method for detecting a drivable area, characterized in that, include: Acquire image data, including road images captured by the target vehicle; The image data is input into a pre-trained drivable area detection model to determine the driving area of ​​the target vehicle from the image data; The drivable area detection model is trained using any one of the methods in claims 1 to 4.

Images