Visibility estimation method based on digital twinborn technology

Abstract

The invention provides a visibility estimation method and system based on a digital twinborn technology, and relates to the technical field of meteorological detection. The method comprises the following steps: simulating atmospheric conditions through a computer based on a digital twinning idea, and generating a medium-low visibility image data set; wherein each data sample in the medium-low visibility image data set adopts visibility as a label; performing iterative training on the visibility estimation network by using the medium-low visibility image data set; and acquiring an image acquired in real time, and inputting the image into the trained visibility estimation network to obtain a visibility estimation result. According to the method, on the basis of a digital twinning concept, atmospheric conditions such as smoke are simulated through a computer, and various and sufficient samples can be created by taking accurate visibility as a label and adopting a domain randomization method in a scene building process in a data set generation stage, so that the precision of a trained model is fundamentally ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological detection, and particularly relates to a visibility estimation method and system based on digital twin technology. Background Art

[0002] Visibility is generally defined as an index that describes the maximum distance at which light or an object can be clearly visible to the human eye or a sensor. This index generally depends on meteorological conditions, air quality, human eye vision, or sensor performance. In meteorology, this index is generally used to reflect the turbidity of the atmosphere, with the unit being meters or kilometers. Visibility is widely used in fields such as transportation, outdoor operations, atmospheric science, and environmental monitoring, and its timeliness, accuracy, and versatility have important impacts on aspects such as disaster warning, safety assurance, and decision-making. Existing measurement methods generally rely on professional hardware devices, which are costly to produce and maintain, require frequent calibration, and their lifespan and accuracy are also affected by harsh working conditions such as temperature and humidity.

[0003] Measuring atmospheric visibility can generally be done by visual observation or by using measuring instruments such as an atmospheric transmissometer or a laser visibility automatic measuring instrument. Currently, most visibility observations are still mainly based on manual visual observation, with relatively poor standardization and objectivity. An atmospheric transmissometer directly measures the transmittance of an air column by passing a light beam through the air column between two fixed points, and then estimates the visibility value based on this. This method requires the light beam to pass through a sufficiently long air column, and the reliability of the measurement is affected by the stability of the light source and other hardware systems. It is generally only applicable to observations of medium and below visibility, and in low visibility weather such as rain and fog, large errors will be caused due to complex conditions such as water vapor absorption. The laser visibility automatic measuring instrument estimates visibility by measuring the atmospheric extinction coefficient with a laser. Relatively speaking, it is more objective and accurate, but this instrument is expensive, has high maintenance costs, is complex to operate, and moreover, it is difficult to conduct normal observations in rainy and foggy days, so it is difficult to promote.

[0004] Machine learning is an effective method to avoid hardware dependence. With the continuous development of machine learning, neural networks have made significant improvements in reasoning ability, especially performing quite well in the field of computer vision. Image classification is one of the most basic and mature tasks, and after adaptation, it can be used as the basic model for visibility estimation. By fitting an image with visibility, the neural network can infer the average visibility in the environment in this direction in real time based on the collected image (hereinafter, this task will be abbreviated as "visibility estimation"). The nature of the task determines that constructing a dataset using traditional methods may face problems such as strict technical requirements, high economic costs, and a long production cycle. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a visibility estimation method and system based on digital twin technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The first aspect of the present invention discloses a visibility estimation method based on digital twin technology, and the method includes:

[0008] Step S1, based on the digital twin concept, simulate the atmospheric conditions by computer and generate a medium and low visibility image dataset; wherein, each data sample in the medium and low visibility image dataset uses visibility as a label;

[0009] Step S2, iteratively train the visibility estimation network using the medium and low visibility image dataset;

[0010] Step S3, obtain the real-time collected image and input it into the trained visibility estimation network to obtain the visibility estimation result.

[0011] In the step S1, based on the digital twin concept, simulate the atmospheric conditions by computer and generate a medium and low visibility image dataset, which specifically includes:

[0012] S11, initialize the scene and import the basic components that make up the scene, and then randomly set the parameters of the basic components that make up the scene based on the domain randomization principle to obtain depth maps under different scenes; for each depth map under a scene, execute step S12;

[0013] S12, set the volume fog parameters to generate volume fog, and then render the depth map under the volume fog at a certain step to obtain medium and low visibility images of the current scene under different visibilities, and based on the visibility fitting function, obtain the visibility corresponding to the medium and low visibility image to generate a visibility label;

[0014] S13, repeat steps S11 - S12 until a medium and low visibility image dataset that meets the quantity requirements is obtained.

[0015] In the step S11, the basic components that make up the scene include a digital twin simulation model, texture maps, light sources, particle emitters, physical fields, and virtual cameras.

[0016] In step S12, the steps of obtaining the visibility fitting function specifically include:

[0017] S121, initialize the scene and import a reference object and volume fog;

[0018] S122, adjust the volume fog parameters, obtain the visible boundary, and calculate the depth value of the visible boundary;

[0019] S123. Repeat steps S121 - S122 until the parameters of the volume fog in S122 are fully traversed within the specified range to obtain a parameter - visibility table for any volume fog parameters, and then obtain a visibility fitting function based on the parameter - visibility table.

[0020] In step S123, based on the parameter - visibility table, perform hash query or multivariate function fitting to obtain a visibility fitting function, that is:

[0021] D vis = f(ρ, i e , r, g, b)

[0022] where D vis is the visibility, ρ, i e , r, g, b are the volume fog parameters, corresponding to concentration, luminous intensity, and the RGB three - element values respectively.

[0023] The second aspect of the present invention discloses a visibility estimation system based on digital twin technology, including:

[0024] The first processing module is configured to simulate atmospheric conditions by computer based on the digital twin concept and generate a medium - low visibility image dataset; among them, each data sample in the medium - low visibility image dataset uses visibility as a label;

[0025] The second processing module is configured to iteratively train a visibility estimation network using the medium - low visibility image dataset;

[0026] The third processing module is configured to acquire a real - time captured image and input it into the trained visibility estimation network to obtain a visibility estimation result.

[0027] Based on the digital twin concept, simulating atmospheric conditions by computer and generating a medium - low visibility image dataset specifically includes:

[0028] Initialize the scene and import the basic components that make up the scene, and then randomly set the parameters of the basic components that make up the scene based on the domain randomization principle to obtain depth maps under different scenes; for each depth map under a scene, execute the next process, that is:

[0029] Set the volume fog parameters to generate volume fog, and then render the depth map under this volume fog at a certain step to obtain medium - low visibility images of the current scene under different visibilities, and obtain the visibility corresponding to this medium - low visibility image based on the visibility fitting function to generate a visibility label;

[0030] Execute the above process of obtaining the depth map and generating the visibility label multiple times until a medium and low visibility image dataset that meets the quantity requirements is obtained.

[0031] The basic components that make up the scene include a digital twin simulation model, texture maps, light sources, particle emitters, physical fields, and virtual cameras.

[0032] A third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps in any one of the visibility estimation methods based on digital twin technology in the first aspect of the present disclosure are implemented.

[0033] A fourth aspect of the present invention discloses a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps in any one of the visibility estimation methods based on digital twin technology in the first aspect of the present disclosure are implemented.

[0034] In summary, the solution proposed by the present invention has the following technical effects:

[0035] Improved robustness under complex working conditions: The present invention uses visible light images as inputs, and the dataset simulates scenes under various complex lighting conditions, and its accuracy is hardly affected by complex factors such as the turbidity of the atmosphere, atmospheric composition, and color difference.

[0036] Improved accuracy of visibility estimation: Based on the concept of digital twin, the present invention can simulate atmospheric conditions such as smoke through computer, and can use accurate visibility as a label during the scene construction in the dataset generation stage, and adopts the domain randomization method to create a variety of samples with sufficient quantity, thus fundamentally ensuring the accuracy of the trained model.

[0037] Reduced economic cost and operation difficulty: The present invention abandons costly sensors and hardware devices, uses a general image acquisition device as the input source (such as a digital camera, etc.), the core program can run on most general-purpose computers, and is user-friendly, without the need for professional training for the operation and maintenance of the instrument. Description of the Drawings

[0038] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0039] Figure 1Flowchart of a visibility estimation method based on digital twin technology according to an embodiment of the present invention;

[0040] Figure 2 Flowchart of batch generating a dataset according to an embodiment of the present invention;

[0041] Figure 3 Flowchart of generating a fitting visibility function according to an embodiment of the present invention;

[0042] Figure 4 Structural diagram of a visibility estimation system based on digital twin technology according to an embodiment of the present invention;

[0043] Figure 5 Structural diagram of an electronic device according to an embodiment of the present invention. Detailed implementation manners

[0044] To make the objectives, 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. Apparently, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0045] It can be understood that the terms "first", "second", etc. used in the present application may be used herein to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of the present application, a first image may be referred to as a second image, and similarly, a second image may be referred to as a first image. Both the first image and the second image are images, but they are not the same image.

[0046] Digital twin is an increasingly mature simulation concept in recent years, that is, representing objects, systems or processes in the physical world through computer simulation. The concept aims to accurately reproduce the corresponding entities in the real world in a virtual manner and perform real-time updates. In the field of computer vision, one of the basic means of digital twin is to build a scene and render images and generate labels based on the principles of computer graphics using 3D modeling to construct a "photo-realistic" simulation dataset. This technology can effectively reduce the data acquisition cost, improve the acquisition efficiency, and solve the problem that it is difficult to create some scenes manually in reality. More importantly, through the scene information in computer simulation, the label annotation work of dataset samples can be directly completed, which can effectively reduce the labor cost and improve the accuracy of labels.

[0047] According to an embodiment of the present invention, in a first aspect, a visibility estimation method based on digital twin technology is provided. Please refer to Figure 1 , the method includes:

[0048] Step S1, based on the digital twin concept, simulate atmospheric conditions by computer and generate a medium and low visibility image dataset; wherein, each data sample in the medium and low visibility image dataset uses visibility as a label;

[0049] Please refer to Figure 2 , in the step S1, based on the digital twin concept, simulate atmospheric conditions by computer and generate a medium and low visibility image dataset, specifically including:

[0050] S11, initialize the scene and import the basic components that make up the scene, and then randomly set the parameters of the basic components that make up the scene based on the domain randomization principle to obtain depth maps under different scenes; for each depth map under a scene, execute step S12;

[0051] In the step S11, the basic components that make up the scene include a digital twin simulation model, a texture map, a light source, a particle emitter, a physical field, and a virtual camera. Among them,

[0052] The digital twin simulation model refers to a "3D model", which is a data structure used to represent three-dimensional objects and is responsible for providing shape, texture, normal, and other information for 3D modeling software. The texture map is used to further improve the authenticity of the model and define information other than the vertices on each triangular face of the model, including base color, roughness, metallicity, normal direction, and so on. The light source is used to simulate different light intensities. The particle emitter is used to generate particles in 3D modeling software, and usually these particles are used to simulate various special effects, including flames, smoke, and fluids. The physical field is a tool in 3D modeling software used to control the physical simulation of specified objects and particles in a certain 3D software scene, and it can interact with objects in the form of neutral, magnetic force, wind field, eddy current, orbit constraint, etc.

[0053] S12, set the volume fog parameters to generate volume fog, and then render the depth map under the volume fog at a certain step length to obtain medium and low visibility images of the current scene under different visibilities, and based on the visibility fitting function, obtain the visibility corresponding to the medium and low visibility image to generate a visibility label; the visibility label is the image performance of a certain basic scene under different visibilities.

[0054] By taking the partial differences of the volume fog parameters in the following formula (1) respectively, the influence law of visibility on these variables can be obtained, and based on this, the volume fog parameters can be set to achieve a linear change in visibility. Since the correlation between the RGB gradients of the pixels near the "visibility boundary" with respect to the change in visibility and the correlation between visibility and the change in volume fog parameters are both differentiable, a relatively large step size can be selected and then the parameter setting can be carried out by means of interpolation.

[0055] Please refer to Figure 3 , in step S12, the steps of obtaining the visibility fitting function specifically include:

[0056] S121, Initialize the scene and import the reference object and volume fog;

[0057] S122, Adjust the volume fog parameters, obtain the visibility boundary, and calculate the depth value of the visibility boundary;

[0058] S123, Repeat steps S121 - S122 until the parameters of the volume fog in S122 are completely traversed within the specified range, so as to obtain a parameter-visibility table under any volume fog parameters, and then obtain the visibility fitting function based on the parameter-visibility table.

[0059] In an example scene, place the example object perpendicular to the lens of the virtual camera and expose the side of the example object to the camera lens to generate a depth map; set the volume fog parameters at a certain step size to generate volume fog, and then perform image rendering on the depth map under this volume fog; calculate the RGB difference of adjacent pixels belonging to the example object along the direction of the example object in the rendering result, and use the pixel points with the RGB difference of adjacent pixels lower than a certain threshold as the visibility boundary, and calculate the values of the nearby pixels within a distance threshold less than the visibility boundary on the depth map to infer the depth of the visibility boundary.

[0060] In step S124, based on the corresponding table of visibility values under any volume fog parameters, perform hash query or multivariate function fitting to obtain the visibility fitting function, that is:

[0061] D vis = f(ρ, i e , r, g, b) (1)

[0062] where D vis is the visibility, ρ, i e , r, g, b are the volume fog parameters, corresponding to the concentration, luminous intensity, and the values of the three RGB elements respectively.

[0063] It should be noted that the steps of obtaining the visibility fitting function are only executed once in the initial process.

[0064] It should be noted that for the depth values that are non-linearly compressed along the pixel direction due to the depth of field, the method can be carried out by changing the field of view angle and focal length of the lens. Here, the non-linear compression means that due to the principle that objects appear larger when they are closer and smaller when they are farther away, in the area with a relatively large depth, a pixel point on the depth map contains a series of points with a very large range of depth changes, and such points are difficult to be effectively quantified by the default 8-bit integer number of the grayscale image. On the premise of ensuring the image resolution remains unchanged, changing the field of view angle and focal length can effectively change the dynamic range of these points, so that the maximum degree of quantization can be carried out locally, reducing the rounding error.

[0065] S13. Repeat steps S11 - S12 until a medium and low visibility image dataset that meets the quantity requirements is obtained.

[0066] In each repetition, clear the previous scene, and import all models, textures, light sources, particle emitters, physical fields, virtual cameras, etc. used for the construction of the current scene as scene assets into the scene. Based on the principle of domain randomization, randomly set the positions, postures, sizes, and some other parameters of these assets, and finally make them form a rather general scene. This step is the fundamental premise to ensure that the rendering results are consistent with the reality in terms of feature distribution. In addition, the heterogeneous and randomized scene design is also the key to the rich variety and diverse features of the dataset.

[0067] Considering the influence of the image acquisition device on visibility, it is necessary to manually adapt the lens parameters, rendering settings, etc. of the virtual camera according to the basic principles of different rendering engines. The lens parameters include lens size, resolution, depth of field, field of view angle, focal length, aperture, lens distortion, etc. The rendering settings include the color space of the image, color jitter degree, dynamic blur degree, exposure, etc.

[0068] In this step, by repeating steps S11 - S12, a series of visibility image renderings are carried out under different scene designs. The purpose of this step is to increase the variety and quantity of scenes, so that the neural network can obtain more comprehensive and rich information, thereby improving the performance of its feature extraction module. In the randomization method, it is generally considered that the number of scene types is more than 8,000, and on the premise of supporting the training method and hardware equipment, the more the better.

[0069] Step S2. Iteratively train the visibility estimation network using the medium and low visibility image dataset;

[0070] In this step, the dataset constructed in step four is used for the training and testing of the basic image classification network for visibility estimation, with visibility as the classification label, and iterate until the test result of the neural network reaches the expected accuracy rate.

[0071] Since the relationship between visibility and pixel changes in images under the same scene is a continuous sequence, which is differentiable compared to text labels and is more friendly to gradient descent, theoretically, it can achieve better training results than general image classification networks.

[0072] Step S3: Obtain the real-time collected images and input them into the trained visibility estimation network to obtain the visibility estimation result.

[0073] The present invention provides a visibility estimation method based on digital twin. This method is based on the concept of digital twin and completes scene construction and the generation of medium and low visibility image datasets through 3D modeling and simulation. This method creates different meteorological conditions by building a virtual scene and parameterizing the generation of volume fog, and conducts visibility calibration and image rendering. Among them, the basis for visibility calibration is the quantitative relationship between volume fog-related parameters and visibility, where the parameters include the concentration, luminous intensity, RGB value, etc. of the volume fog. Under the same volume fog parameters, the obtained visibility is also the same.

[0074] Since the parameters are manually adjustable, this method can generate images with arbitrary parameters and visibility, and can customize the scene by changing the positions of entities such as models, light sources, and virtual cameras in the scene. While ensuring that the generated images are "photo-realistic", it effectively increases the variety and quantity of data. This dataset is used for the training of the visibility estimation network, which can effectively improve the accuracy and generalization ability of the latter, enabling the neural network to accurately estimate visibility all-weather and under all working conditions only relying on general image acquisition devices.

[0075] A second aspect of the present invention discloses a visibility estimation system based on digital twin technology. Please refer to Figure 4 , the system includes:

[0076] The first processing module 100 is configured to simulate atmospheric conditions by computer based on the concept of digital twin and generate a medium and low visibility image dataset; among them, each data sample in the medium and low visibility image dataset uses visibility as a label;

[0077] The second processing module 200 is configured to iteratively train the visibility estimation network using the medium and low visibility image dataset;

[0078] The third processing module 300 is configured to obtain the real-time collected images and input them into the trained visibility estimation network to obtain the visibility estimation result.

[0079] Based on the concept of digital twin, simulating atmospheric conditions by computer and generating a medium and low visibility image dataset specifically includes:

[0080] Initialize the scene and import the basic components that make up the scene. Then, based on the principle of domain randomization, randomly set the parameters of the basic components that make up the scene to obtain depth maps under different scenes. For each depth map under a scene, perform the next process, that is:

[0081] Set the volume fog parameters to generate volume fog. Then, render the depth map under this volume fog at a certain step length to obtain medium and low visibility images of the current scene under different visibilities, and based on the visibility fitting function, obtain the visibility corresponding to this medium and low visibility image to generate a visibility label;

[0082] Execute the above processes of obtaining depth maps and generating visibility labels multiple times until a medium and low visibility image dataset that meets the quantity requirements is obtained.

[0083] The basic components that make up the scene include a digital twin simulation model, texture maps, light sources, particle emitters, physical fields, and virtual cameras.

[0084] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps in any one of the visibility estimation methods based on digital twin technology in the first aspect of the present disclosure are implemented.

[0085] Figure 5 As shown in the structure diagram of an electronic device according to an embodiment of the present invention, Figure 5 The electronic device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, a carrier network, near field communication (NFC), or other technologies. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the electronic device, or an external keyboard, touchpad, or mouse, etc.

[0086] Those skilled in the art can understand that Figure 5 The structure shown in is only the structure diagram of the part related to the technical solution of the present disclosure, and does not constitute a limitation on the electronic device to which the solution of the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0087] A fourth aspect of the present invention discloses a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps in a visibility estimation method based on digital twin technology according to any one of the first aspects of the present disclosure are implemented.

[0088] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that the technical solutions described in the foregoing embodiments can still be modified, or some or all of the technical features can be equivalently replaced, and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A visibility estimation method based on digital twin technology, characterized in that: The method comprises: Step S1: Based on the concept of digital twins, atmospheric conditions are simulated by computer, and a medium- and low-visibility image dataset is generated; wherein each data sample in the medium- and low-visibility image dataset uses visibility as a label; Step S2, iteratively training the visibility estimation network using a medium and low visibility image dataset; Step S3: Acquire the real-time collected image and input it into the trained visibility estimation network to obtain the visibility estimation result.

2. The method according to claim 1, characterized in that: In step S1, based on the concept of digital twins, atmospheric conditions are simulated by computer, and a medium and low visibility image data set is generated, which specifically includes: S11, initializing the scene and importing the basic components constituting the scene, and then randomly setting the parameters of the basic components constituting the scene based on the domain randomization principle to obtain depth maps under different scenes; for the depth map under each scene, executing step S12; S12, setting volume fog parameters to generate volume fog, and then performing image rendering on the depth map under the volume fog according to a certain step length to obtain medium and low visibility images of the current scene under different over-visibility, and obtaining the visibility corresponding to the medium and low visibility image based on the visibility fitting function to generate a visibility label; S13, repeating steps S11-S12 until a low-visibility image dataset that meets the quantity requirement is obtained.

3. The method according to claim 2, characterized in that In step S11, the basic components constituting the scene include a digital twin simulation model, a texture, a light source, a particle emitter, a physical field, and a virtual camera.

4. The method according to claim 2, characterized in that: In step S12, the step of obtaining the visibility fitting function specifically includes: S121, initialize the scene and import reference objects and volume fog; S122, adjusting volume fog parameters, obtaining a visible boundary, and calculating a depth value of the visible boundary; S123, repeating steps S121-S122 until the volume fog parameters in S122 are completely traversed within the specified range, so as to obtain a parameter-visibility table under any volume fog parameters, and then obtain a visibility fitting function based on the parameter-visibility table.

5. The method according to claim 4, characterized in that In step S123, based on the parameter-visibility table, a hash query or multivariate function fitting is performed to obtain a visibility fitting function, namely: D vis =f(ρ,i e ,r,g,b) Among them, D vis is visibility, ρ,i e , r, g, b are volume fog parameters, corresponding to concentration, luminous intensity and RGB three-element values ​​respectively.

6. A visibility estimation system based on digital twin technology, characterized in that: include: The first processing module is configured to simulate atmospheric conditions through a computer based on the concept of digital twins and generate a medium-low visibility image data set; wherein each data sample in the medium-low visibility image data set uses visibility as a label; A second processing module is configured to iteratively train the visibility estimation network using a medium and low visibility image dataset; The third processing module is configured to obtain the real-time collected image and input it into the trained visibility estimation network to obtain the visibility estimation result.

7. The system according to claim 6, characterized in that Based on the concept of digital twins, atmospheric conditions are simulated by computer and a low- to medium-visibility image dataset is generated, including: Initialize the scene and import the basic components that constitute the scene, and then randomly set the parameters of the basic components that constitute the scene based on the principle of domain randomization to obtain depth maps under different scenes; for the depth map under each scene, perform the next process, namely: Set volumetric fog parameters to generate volumetric fog, and then render the depth map under the volumetric fog according to a certain step size to obtain medium and low visibility images of the current scene under different over-visibility, and obtain the visibility corresponding to the medium and low visibility image based on the visibility fitting function to generate a visibility label; The above process of obtaining the depth map and generating the visibility label is performed multiple times until a medium and low visibility image dataset that meets the quantity requirements is obtained.

8. The system according to claim 7, characterized in that The basic components that make up the scene include digital twin simulation models, textures, light sources, particle emitters, physics fields, and virtual cameras.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps in the visibility estimation method based on digital twin technology described in any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the visibility estimation method based on digital twin technology described in any one of claims 1 to 5 are implemented.

Images