CN112070037B - Road extraction method, device, medium and equipment based on remote sensing image - Google Patents

Abstract

Provided are a method, device, medium and equipment for extracting a road based on a remote sensing image. The method comprises the following steps: drawing a road in the remote sensing image, and converting the road into a two-classification image; dividing the remote sensing image and the binary image to manufacture a remote sensing image sample and a binary image sample; carrying out edge symbol distance transformation on the binary image samples to obtain edge symbol distance image samples corresponding to the binary image samples one to one; training a road extraction model by using the corresponding remote sensing image sample, the binary image sample and the edge symbol distance image sample; and carrying out road extraction on the remote sensing image by using the trained road extraction model. When the road is extracted, the road continuity performance is obviously improved, and most of the shielding problems can be overcome.

Description

Road extraction method, device, medium and equipment based on remote sensing image

technical field

The invention belongs to the field of information technology, and in particular, relates to a road extraction method, device, medium and equipment based on remote sensing images.

Background technique

Road information is one of the most important geographic information elements and plays an important role in applications such as autonomous driving and disaster emergency response. With the development of remote sensing technology and the maturity of deep learning technology, road extraction algorithms from remote sensing images with high temporal and spatial resolution based on CNN emerge in an endless stream, allowing large-scale monitoring of roads. Therefore, remote sensing image data has quickly become an important data source for automatic extraction of road networks, and the research on algorithms for automatic extraction of roads from remote sensing images has become the focus.

However, the current research mainly focuses on road extraction in urban areas. Because roads in remote areas are often narrow, variable in width, and have serious tree canopy, shadow occlusion and other problems, these road extraction algorithms for urban areas are often unsatisfactory in remote areas, especially The problem of road discontinuity and fragmentation can be serious.

SUMMARY OF THE INVENTION

The present invention aims to solve the problems described above. Specifically, the present invention provides a method, device, medium and device for road extraction based on remote sensing images.

According to the first aspect of this paper, a road extraction method based on remote sensing images is provided, including:

Draw roads in remote sensing images and convert them into binary images;

Segment remote sensing images and binary images to produce remote sensing image samples and binary image samples, and remote sensing image samples and binary image samples correspond one-to-one;

Perform edge symbol distance transformation on the two-class image samples to obtain edge-symbol distance image samples corresponding to the two-class image samples one-to-one;

Use the corresponding remote sensing image samples, binary image samples, and edge symbol distance image samples to train the road extraction model;

Use the trained road extraction model to extract roads from remote sensing images.

Performing edge signed distance transformation on binary image samples includes:

Determine the distance D _i from each pixel point in the binary image sample to the closest point located on the road edge, and determine the edge symbol distance of all pixel points based on D _i .

其中，x_i为像素点位置，x_j为道路边缘上与x_i最近的像素点位置，ED为欧几里得距离。

Among them, x _i is the pixel position, x _j is the pixel position closest to x _i on the edge of the road, and ED is the Euclidean distance.

Determining the edge sign distance of all pixels based on D _i includes:

其中，Htanh为HardTanh函数，α为比例系数，F为前景区域，B分别背景区域。The edge signed distance is BSD _i , then

Among them, Htanh is the HardTanh function, α is the scale coefficient, F is the foreground area, and B is the background area.

The road extraction model includes the ResNet framework, the distance regression task branch, and the classification task branch.

Using the corresponding remote sensing image samples, binary image samples, and edge symbol distance image samples to train the road extraction model includes:

Input remote sensing image samples into the ResNet framework to obtain remote sensing image feature maps;

The remote sensing image feature map is input into the distance regression task branch, and the output result of the distance regression task branch and the edge symbol distance image sample are used to calculate the first Loss through the first loss function;

The shallow features and deep features obtained in the distance regression task branch are input into the classification task branch, and the output result of the classification task branch and the binary image sample are calculated by the second loss function through the second loss function.

The road extraction method based on remote sensing images also includes:

Based on the first Loss and the second Loss, the final Loss is determined, and the road extraction model is trained using the final Loss.

According to another aspect of this article, a road extraction device based on remote sensing images is provided, comprising:

The two-class image production module is used to draw the road in the remote sensing image and convert it into a two-class image;

The sample making module is used to segment the remote sensing image and the two-class image, and produce the remote sensing image sample and the two-class image sample, and the remote-sensing image sample and the two-class image sample are in one-to-one correspondence;

The edge symbol distance transformation module is used to perform edge symbol distance transformation on the two-class image samples, and obtain the edge symbol distance image samples corresponding to the two-class image samples one-to-one;

The model training module is used to train the road extraction model using the corresponding remote sensing image samples, binary image samples, and edge symbol distance image samples;

The road extraction module is used to extract roads from remote sensing images using the trained road extraction model.

According to another aspect of this document, there is provided a computer-readable storage medium on which a computer program is stored, and when the computer program is executed, implements the steps of a road extraction method based on a remote sensing image.

According to another aspect of this document, a computer device is provided, comprising a processor, a memory, and a computer program stored on the memory, and the processor implements the steps of a road extraction method based on remote sensing images when the processor executes the computer program.

The present invention obtains edge symbol distance image samples by performing edge symbol distance transformation on binary images of remote sensing images, and uses remote sensing image samples, binary classification image samples, and edge symbol distance image samples to train a multi-branch road extraction model. The real-valued distance information in the edge symbol distance image can promote the model to better identify the road boundary, and better classify the road pixels by learning the distance features of the boundary, rather than limited to the spectral feature learning, to correct the shape deformity Or break the purpose of the road.

Other characteristic features and advantages of the present invention will become apparent upon reading the following description of exemplary embodiments with reference to the accompanying drawings.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention. In the figures, like reference numerals are used to refer to like elements. The drawings in the following description are some, but not all, embodiments of the invention. For those of ordinary skill in the art, other drawings can be obtained from these drawings without creative effort.

FIG. 1 is a flowchart of a road extraction method based on remote sensing images according to an exemplary embodiment.

FIG. 2 is a schematic diagram of a remote sensing image according to an exemplary embodiment.

FIG. 3 is a schematic diagram of a two-category image according to an exemplary embodiment.

FIG. 4 is a schematic diagram of an edge symbol distance image according to an exemplary embodiment.

FIG. 5 is a schematic diagram of a road extraction model according to an exemplary embodiment.

FIG. 6 is a schematic diagram illustrating a prediction result of a road extraction model according to an exemplary embodiment.

FIG. 7 is a block diagram of an apparatus for extracting roads based on remote sensing images according to an exemplary embodiment.

FIG. 8 is a block diagram of a computer device for road extraction based on remote sensing images, according to an exemplary embodiment.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. It should be noted that, the embodiments in the present application and the features in the embodiments may be arbitrarily combined with each other if there is no conflict.

FIG. 1 is a flowchart of a method for extracting roads based on remote sensing images according to an exemplary embodiment. Referring to Figure 1, the road extraction method based on remote sensing images includes:

Step S11: draw the road in the remote sensing image and convert it into a binary image;

Step S12 : segment the remote sensing image and the second-class image, and create the remote-sensing image sample and the second-class image sample, and the remote-sensing image sample and the second-class image sample are in one-to-one correspondence;

Step S13: performing edge symbol distance transformation on the two-category image samples to obtain edge symbol distance image samples corresponding to the two-category image samples one-to-one;

Step S14: using corresponding remote sensing image samples, two-class image samples, and edge symbol distance image samples to train a road extraction model;

Step S15: Use the trained road extraction model to extract roads from the remote sensing images.

In step S11, the road objects in the remote sensing image are drawn, and drawing tools such as ArcGIS, labelme, etc. can be selected. Taking ArcGIS as an example, create a new polygon layer element in the remote sensing image, and assign the same geographic coordinate system as the original remote sensing image, and then use the editing tool to manually interpret the road in the remote sensing image to delineate, select Appropriate scaling, drawing along edge pixels, and finally generating a vector file in shapefile format. Then, the vector image is converted into a raster image to obtain a binarized raster image, that is, a binary image. In the binary image, 0 is the background and 1 is the road.

In step S12, a sample segmentation rule is set, including the size of a single sample and the overlap ratio between adjacent samples. According to the set sample segmentation rules, the remote sensing image and the binary image are segmented, and multiple segmented remote sensing image image samples and the same multiple binary image samples can be obtained as the training samples of the subsequent road extraction model.

In step S13, edge-symbol distance transformation is performed on the two-class image samples to obtain edge-symbol distance image samples corresponding to the two-class image samples one-to-one. When only two-class images are used to train the model, the deep classification model determines whether a pixel belongs to the road category or the non-road category, ignoring the high spatial distribution feature correlation between road pixels, so that the road extraction results will often be occluded. Some pixels are classified as non-roads, leading to the problem of road fragmentation and discontinuity.

Therefore, the edge symbol distance is introduced in this paper. In one embodiment, performing edge symbol distance transformation on the binary image sample includes:

Determine the distance D _i from each pixel point in the binary image sample to the closest point located on the road edge, and determine the edge symbol distance of all pixel points based on D _i .

Inspired by the signed distance transformation SDT, given a pixel position x _i , its signed distance value SDT _i is:

where ED is the Euclidean distance, and F and B are the foreground and background regions, respectively. The signed distance value SDT _i represents the distance from a pixel to its closest pixel of different categories.

其中，x_i为像素点位置，x_j为道路边缘上与x_i最近的像素点位置，ED为欧几里得距离。Thus, the distance from a given pixel to the nearest pixel on the edge of the road can be defined as D _i ,

Among them, x _i is the pixel position, x _j is the pixel position closest to x _i on the edge of the road, and ED is the Euclidean distance.

其中，Htanh为HardTanh函数，α为比例系数，F为前景区域，B分别背景区域。由于在二分类图像上进行计算，前景区域为道路，背景区域为非道路。edge sign distance BSD _i ,

Among them, Htanh is the HardTanh function, α is the scale coefficient, F is the foreground area, and B is the background area. Since the calculation is performed on a binary image, the foreground area is a road, and the background area is a non-road.

In the above formula, the distance D _i is scaled by logarithmic mapping, so that the sensitivity of the edge symbol distance to the distance change decreases with the increase of the distance between the pixel point and the road edge, so that the attention can be concentrated on the road edge and the relative distance. Neighboring regions, the attention to background regions far away from the road edge is weakened. Then, the distance value is normalized to [-1, 1] using the HardTanh function and the scale coefficient α, which speeds up the training speed of the model and further weakens the focus on background pixels far away from the road, thereby filtering out some noise. The edge sign distance assigned to road pixels is positive, non-road pixels edge sign distance is negative, and the edge sign distance value of road boundary pixels is zero, which will effectively help the model to better distinguish between roads and non-roads. FIG. 2 is a schematic diagram of a remote sensing image according to an exemplary embodiment. FIG. 3 is a schematic diagram of a two-category image according to an exemplary embodiment. FIG. 4 is a schematic diagram of an edge symbol distance image according to an exemplary embodiment. It can be seen that the schematic diagram of the remote sensing image, the two-class image, and the edge symbol distance image have a one-to-one correspondence.

In step S14, a road extraction model is trained using the corresponding remote sensing image samples, binary classification image samples, and edge symbol distance image samples. The road extraction model is trained by using the remote sensing images, two-class images, and edge symbol distance images prepared in steps S11 to S13 as training samples.

In one embodiment, the road extraction model includes a ResNet framework, a distance regression task branch, and a classification task branch.

FIG. 5 is a schematic diagram of a road extraction model according to an exemplary embodiment. Referring to Figure 5, 51 in the figure is the ResNet framework, 52 is the distance regression task branch, and 53 is the classification task branch.

Using the corresponding remote sensing image samples, binary image samples, and edge symbol distance image samples to train the road extraction model includes:

Input remote sensing image samples into the ResNet framework to obtain remote sensing image feature maps;

The remote sensing image feature map is input into the distance regression task branch, and the output result of the distance regression task branch and the edge symbol distance image sample are used to calculate the first Loss through the first loss function;

The shallow features and deep features obtained in the distance regression task branch are input into the classification task branch, and the output result of the classification task branch and the binary image sample are calculated by the second loss function through the second loss function.

Based on the first Loss and the second Loss, the final Loss is determined, and the road extraction model is trained using the final Loss.

The road extraction model will be described below with reference to FIG. 5 .

In this embodiment, the ResNet framework is selected as the backbone network. Considering that the forest border road is usually small in width and has few pixels, the model needs to maintain the high-frequency details of the input image. To this end, changing the stride of the last two residual blocks of the ResNet framework to 1 can obtain a feature map of 1/8 size of the input image. In addition, dilated convolutions are used to expand the receptive field of convolutional layers, making the spatial context information richer. Input the remote sensing image sample (5c in the figure) into the ResNet framework, and then pass the obtained remote sensing image feature map into 52 (distance regression task branch).

In the distance regression task branch, firstly, a 3*3 convolutional layer (Conv) is used to reduce the dimension of the remote sensing image feature map generated by the backbone network. Then, a batch normalization layer (BN) is used to normalize the distribution of each batch of inputs to a standard normal distribution to reduce the internal covariate offset and speed up the learning process, and finally use Tanh as the activation function to generate shallow Layer feature map. Then, the shallow feature map is sequentially operated by a 1*1 convolution layer (Conv) and the activation function Tanh to generate a deep feature map, and a low-resolution distance value prediction result is obtained. The Tanh function is used to limit the distance value to the range of [-1, 1], so as to keep the label value range of the sample (5a in the figure) consistent with the edge symbol distance transformation. Then perform up-sampling to obtain the edge symbol distance prediction result (Outputs-A) of the same size as the remote sensing image sample (5c in the figure). In the distance regression task branch, the edge-signed distance image samples are used as the ground truth, and the L1 loss function is used to calculate the first Loss (Loss-1 in the figure).

For the classification task branch (53 in the figure), connect the shallow features and deep features directly from the distance regression task branch, input the classification task branch, and combine the shallow feature map and deep feature generated in the distance regression task branch. The graph is shared between the two branches, distance regression task and binary classification task, to achieve the fusion of distance features and semantic features. Then input 3*3 convolutional layers, batch normalization layers and ReLU activation layers to generate semantic feature maps. Then, the classification probability map is obtained through the 1*1 convolutional layer. The classification probability map is up-sampled to obtain the classification prediction results (Outputs-B) of the same size as the remote sensing image samples. In the classification task branch, the binary image sample (5b in the figure) is used as the true value, and the cross-entropy loss function is used to calculate the second Loss (Loss-2 in the figure).

其中，N表示像素的总个数，BSD^gt _i，表示像素i的真值，BSD_i为预测的边缘符号距离值。分类任务分支采用交叉熵损失函数，第二Loss为：

其中N表示像素的总个数，y_i表示真值的类别，p_i表示预测值的类别。As mentioned above, the model adopts a multi-task training strategy, and the regression task branch adopts the L1 loss function. The first Loss is:

Among them, N represents the total number of pixels, BSD ^gt _i represents the true value of pixel i, and BSD _i represents the predicted edge symbol distance value. The classification task branch adopts the cross entropy loss function, and the second Loss is:

where N represents the total number of pixels, _yi represents the category of the true value, and p _i represents the category of the predicted value.

The final Loss of multitasking is calculated as: L _final =L _bsd +λ×L _cls .

where λ is a hyperparameter that balances the magnitude difference between the two types of losses and can also be used to control the weighting ratio for classification and regression tasks.

The road extraction model is trained using the final Loss until convergence, and after the model is validated, the road extraction model is used for road extraction.

Overall, the distance regression task branch can be seen as an intermediate supervision for the classification task branch.

From the above description, this paper obtains edge symbol distance image samples by performing edge symbol distance transformation on binary images of remote sensing images, and uses remote sensing image samples, binary image samples, and edge symbol distance image samples to train the multi-branch road extraction model. . The real-valued distance information in the edge symbol distance image can promote the model to better identify the road boundary, and better classify the road pixels by learning the distance features of the boundary, rather than limited to the spectral feature learning, to correct the shape deformity Or break the purpose of the road.

Fig. 6 is a schematic diagram showing a prediction result of a road extraction model according to an exemplary embodiment. 61 is the remote sensing image, 62 is the ground truth of the road, 63 is the prediction result using only binary classification, and 64 is the prediction result using edge sign distance. Referring to Figure 6, the road extraction method based on remote sensing images provided in this paper introduces the distance regression task into the road extraction model for supervision. When extracting roads, the road continuity performance is significantly improved, and most of the occlusion problems can be overcome.

FIG. 7 is a block diagram of an apparatus for extracting roads based on remote sensing images according to an exemplary embodiment. Referring to FIG. 7 , a road extraction device based on remote sensing images includes: a two-class image production module 701 , a sample production module 702 , an edge symbol distance transformation module 703 , a model training module 704 , and a road extraction module 705 .

The two-class image making module 701 is configured to draw the road in the remote sensing image and convert it into a two-class image.

The sample making module 702 is configured to segment the remote sensing image and the two-class image, and create a remote-sensing image sample and a two-class image sample, and the remote-sensing image sample corresponds to the two-class image sample one-to-one.

The edge symbol distance transformation module 703 is configured to perform edge symbol distance transformation on the two-class image samples to obtain edge symbol distance image samples one-to-one corresponding to the two-class image samples.

The model training module 704 is configured to train a road extraction model using the corresponding remote sensing image samples, the binary image samples, and the edge symbol distance image samples.

The road extraction module 705 is configured to perform road extraction from the remote sensing image using the trained road extraction model.

FIG. 8 is a block diagram of a computer device 800 for road extraction based on remote sensing images, according to an exemplary embodiment. For example, computer device 800 may be provided as a server. Referring to FIG. 8 , the computer device 800 includes a processor 801, and the number of the processors can be set to one or more as required. Computer device 800 also includes memory 802 for storing instructions executable by processor 801, such as application programs. The number of memories can be set to one or more as required. It can store one or more applications. The processor 801 is configured to execute instructions to perform a method for road extraction based on remote sensing images, including:

Draw roads in remote sensing images and convert them into binary images;

Segmenting the remote sensing image and the second-class image to create a remote-sensing image sample and a second-class image sample, where the remote-sensing image sample corresponds to the second-class image sample one-to-one;

Performing edge symbol distance transformation on the two-class image samples to obtain edge symbol distance image samples corresponding to the two-class image samples one-to-one;

Using the corresponding remote sensing image samples, the two-class image samples, and the edge symbol distance image samples to train a road extraction model;

Use the trained road extraction model to extract roads from remote sensing images.

As will be appreciated by those skilled in the art, the embodiments herein may be provided as a method, an apparatus (apparatus), or a computer program product. Accordingly, this document may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this document may take the form of a computer program product embodied on one or more computer-usable storage media having computer-usable program code embodied therein. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data , including but not limited to RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cartridges, magnetic tape, magnetic disk storage or other magnetic storage devices, or may be used for Any other medium that stores desired information and can be accessed by a computer, etc. In addition, communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and can include any information delivery media, as is well known to those of ordinary skill in the art .

Described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (apparatus) and computer program products according to embodiments herein. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The means implements the functions specified in one or more of the flowcharts and/or one or more blocks of the block diagrams

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

As used herein, the terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that an article or device comprising a list of elements includes not only those elements, but also others not expressly listed elements, or elements inherent to the article or equipment. Without further limitation, an element defined by the phrase "comprising" does not preclude the presence of additional identical elements in the article or device comprising said element.

While the preferred embodiments have been described herein, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be construed to include the preferred embodiment and all changes and modifications that fall within the scope of this document.

It will be apparent to those skilled in the art that various modifications and variations can be made in this document without departing from the spirit and scope of this document. Thus, provided that such modifications and variations herein come within the scope of the claims herein and their equivalents, it is intended that such modifications and variations are also included herein.

Claims (7)

1. The road extraction method based on the remote sensing image is characterized by comprising the following steps:

drawing a road in the remote sensing image, and converting the road into a two-classification image;

dividing the remote sensing image and the two classified images to manufacture a remote sensing image sample and two classified image samples, wherein the remote sensing image sample corresponds to the two classified image samples one to one;

performing edge symbol distance conversion on the two classified image samples to obtain edge symbol distance image samples corresponding to the two classified image samples one to one;

training a road extraction model by using the corresponding remote sensing image sample, the two classification image samples and the edge symbol distance image sample;

carrying out road extraction on the remote sensing image by using the trained road extraction model;

the performing edge-symbol distance transformation on the two classified image samples comprises:

determining a distance D from each pixel point in the two classified image samples to a closest point located on a road edge _i Based on D _i Determining the edge symbol distance of all pixel points;

wherein x is _i Is the pixel location, x _j On the road edge with x _i ED is the Euclidean distance at the position of the nearest pixel point;

the base is based on D _i Determining the edge symbol distances of all the pixel points comprises:

edge symbol distance BSD _i Then, then

Wherein Htanh is a HardTanh function, alpha is a proportionality coefficient, F is a foreground area, and B is a background area.

2. The method for extracting a road based on remote sensing images as claimed in claim 1, wherein the road extraction model comprises a ResNet frame, a distance regression task branch and a classification task branch.

3. The method of claim 2, wherein the training of the road extraction model using the corresponding remote-sensing image samples, the two-class image samples and the edge sign distance image samples comprises:

inputting the remote sensing image sample into a ResNet frame to obtain a remote sensing image characteristic diagram;

inputting the remote sensing image feature map into the distance regression task branch, and calculating a first Loss through a first Loss function between the output result of the distance regression task branch and the edge symbol distance image sample;

and inputting the shallow feature and the deep feature obtained from the distance regression task branch into the classification task branch, and calculating a second Loss through a second Loss function by using the output result of the classification task branch and the binary image sample.

4. The method for extracting a road based on a remote sensing image as claimed in claim 3, further comprising:

and determining a final Loss based on the first Loss and the second Loss, and training the road extraction model by using the final Loss.

5. Road extraction element based on remote sensing image, its characterized in that includes:

the second classification image making module is used for drawing the road in the remote sensing image and converting the road into a second classification image;

the sample manufacturing module is used for segmenting the remote sensing image and the two classified images to manufacture a remote sensing image sample and two classified image samples, and the remote sensing image sample corresponds to the two classified image samples one by one;

the edge symbol distance conversion module is used for carrying out edge symbol distance conversion on the binary image samples to obtain edge symbol distance image samples which correspond to the binary image samples one to one;

the model training module is used for training a road extraction model by using the corresponding remote sensing image sample, the two classification image samples and the edge symbol distance image sample;

the road extraction module is used for extracting the road of the remote sensing image by using the trained road extraction model;

the performing edge-symbol distance transform on the two classified image samples comprises:

determining a distance D from each pixel point in the two classified image samples to a closest point located on a road edge _i Based on D _i Determining the edge symbol distance of all pixel points;

wherein x is _i Is the pixel location, x _j On the road edge with x _i ED is the Euclidean distance at the position of the nearest pixel point;

the base is based on D _i Determining the edge symbol distances of all the pixel points comprises:

edge symbol distance BSD _i Then, then

Wherein Htanh is a HardTanh function, alpha is a proportionality coefficient, F is a foreground area, and B is a background area.

6. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed, implements the steps of the method according to any one of claims 1-4.

7. A computer arrangement comprising a processor, a memory and a computer program stored on the memory, characterized in that the processor, when executing the computer program, carries out the steps of the method according to any of claims 1-4.

Images