KR102599330B1 - Image-based fog area detection and visibility calculation system - Google Patents

Abstract

The fog area detection and visibility distance calculation system of the present invention includes an image collection unit, a fog area detection unit, and a visibility distance calculation unit. The video collection unit is installed in a fixed location and collects original video. The fog area detection unit detects the fog area in the original image using a fog area detection model learned in advance based on an artificial intelligence algorithm. Each of the plurality of reference distance lines is defined as a line marking points on the original image that are separated by the reference distance from the image collection unit. The visibility distance calculation unit calculates the visibility distance according to whether each of the plurality of reference distance lines is included in the fog area. Each of the first and second reference distance lines is one of a plurality of reference distance lines. The first reference distance, which is the reference distance corresponding to the first reference distance line, is longer than the second reference distance, which is the reference distance corresponding to the second reference distance line.

Description

Image-based fog area detection and visibility range calculation system {IMAGE-BASED FOG AREA DETECTION AND VISIBILITY CALCULATION SYSTEM}

The present invention relates to a system and method for detecting fog areas and calculating visibility distance by analyzing images. Specifically, the present invention relates to a system and method for detecting a fog area using an artificial intelligence algorithm and calculating a visibility distance based on the area of the fog area.

Fog is a meteorological phenomenon that is formed when fine particles such as water vapor or dust in the atmosphere condense to more than a certain volume under specific conditions (high humidity and low temperature). When fog forms, visibility is obscured and it is impossible to see long distances, which greatly affects the operation and safety of cars, ships, airplanes, etc. In situations where visibility is obscured by fog, it is difficult to respond to rapid changes, greatly increasing the risk of accidents.

For this reason, technology to detect fog areas and calculate visibility is very important for the operation of cars, ships, and airplanes. However, conventional related technologies are limited to simply detecting whether fog has occurred or the concentration of fog, and there is an urgent need to develop technology that can detect visibility in foggy situations.

The present invention is intended to solve the conventional technical limitations described above. It determines whether fog has occurred based on the collected images, calculates the visibility distance, and provides a system that can be used for user decision-making. The purpose is to

In addition, the purpose of the present invention is to provide a highly usable system that can calculate the visibility distance not only in fog but also in various low-visibility situations (land fog, sea fog, yellow dust, smoke, etc.).

Additionally, the purpose of the present invention is to provide a system that can accurately evaluate the performance of a learned model based on area.

In addition, the purpose of the present invention is to provide a system that allows users to understand and explain the visibility distance calculation process, so that users can optimize performance according to the situation and environment and analyze the visibility distance calculation results after the fact. do.

The fog area detection and visibility distance calculation system according to an embodiment of the present invention may include an image collection unit, a fog area detection unit, and a visibility distance calculation unit.

The video collection unit can be installed in a fixed location to collect original video.

The fog area detection unit can detect the fog area in the original image using a fog area detection model learned in advance based on an artificial intelligence algorithm.

The visibility distance calculation unit may calculate the visibility distance depending on whether each of a plurality of reference distance lines is included in the fog area. Each of the plurality of reference distance lines is defined as a line marking points on the original image that are separated by a reference distance from the image collection unit.

Each of the first and second reference distance lines may be one of the plurality of reference distance lines. The first reference distance corresponding to the first reference distance line may be longer than the second reference distance corresponding to the second reference distance line.

When the first reference distance line and the second reference distance line are not included in the fog area, the visibility distance calculation unit may determine that the visibility distance is longer than the first reference distance.

When only the first reference distance line among the first reference distance line and the second reference distance line is included in the fog area, the visibility distance calculation unit determines that the visibility distance is longer than the second reference distance and the first reference distance line is included in the fog area. It can be judged to be shorter than the distance.

When the first reference distance line and the second reference distance line are included in the fog area, the visibility distance calculation unit may determine that the visibility distance is shorter than the second reference distance.

The artificial intelligence algorithm of the fog area detection and visibility distance calculation system according to an embodiment of the present invention is at least one of the YOLO (You Only Look Once) algorithm, the R-CNN algorithm, the Mask R-CNN algorithm, and the Transformer algorithm. It could be any one.

The fog area detection and visibility distance calculation system according to an embodiment of the present invention may further include a data set generator. The dataset generator may generate a dataset for learning and evaluating the fog area detection model.

The dataset generator may select a plurality of second images from a plurality of first images collected in advance. Additionally, the dataset generator may generate metadata by performing annotation on each of the plurality of second images.

The dataset may include the plurality of second images and metadata corresponding to each of the plurality of second images.

The dataset can be divided into a learning-dataset for training the fog area detection model and a test-dataset for evaluating the performance of the fog area detection model.

The metadata generated in the dataset generator of the fog area detection and visibility range calculation system according to an embodiment of the present invention includes bounding box, polygon, semantic segmentation, and poly It can be at least one of the lines (polyline).

In the fog area detection and visibility distance calculation system according to an embodiment of the present invention, the area of the object classified as fog in each of the plurality of second images may be 50% or more and 99.9% or less of the total area.

In the process of generating the metadata, the dataset generator of the fog area detection and visibility range calculation system according to an embodiment of the present invention may perform annotation only for fog in each of the plurality of second images.

In the fog area detection and visibility distance calculation system according to an embodiment of the present invention, the artificial intelligence algorithm may be the YOLO (You Only Look Once) algorithm. The dataset generator may use the Label Smoothing technique in the process of generating the metadata.

In the fog area detection and visibility distance calculation system according to an embodiment of the present invention, the artificial intelligence algorithm is the Mask R-CNN algorithm, and the metadata may be a polygon.

The fog area detection and visibility distance calculation system according to an embodiment of the present invention may further include a detection model evaluation unit. The detection model evaluation unit may calculate the fog area detection agreement rate based on the test-data set. Additionally, the detection model evaluation unit may evaluate the performance of the fog area detection model based on the fog area detection agreement rate.

The test-dataset may include a plurality of third images that are part of the plurality of second images and metadata corresponding to each of the third images.

The first area is defined as an area that the dataset generator annotated on the third image and classified as a fog area. The first area is defined as the area of the first region.

The second area is defined as an area detected by the fog area detection model of the fog area detection unit as a fog area in the third image. The second area is defined as the area of the second region.

The third area is defined as an area where the first area and the second area overlap. The third area is defined as the area of the third region.

The detection model evaluation unit may calculate a fog area detection agreement rate based on the first area, the second area, and the third area.

The detection model evaluation unit of the fog area detection and visibility distance calculation system according to an embodiment of the present invention can calculate the fog area detection agreement rate using Equation 1 below.

[Equation 1]

According to one embodiment of the present invention, the purpose is to provide a system that can be used for user decision-making by determining whether fog has occurred based on collected images and calculating and providing a visibility distance.

In addition, according to an embodiment of the present invention, the visibility distance can be calculated not only in fog but also in various low visibility situations (land fog, sea fog, yellow dust, smoke, etc.), thereby providing a highly usable system.

Additionally, according to an embodiment of the present invention, it is possible to provide a system that can accurately evaluate the performance of a learned model based on area.

In addition, according to an embodiment of the present invention, the user can understand and explain the visible distance calculation process, so the user can optimize performance according to the situation and environment and analyze the visible distance calculation results after the fact. can be provided.

Figure 1 exemplarily shows a block diagram of a fog area detection and visibility distance calculation system according to an embodiment of the present invention.
Figures 2a to 2d are shown to explain the process by which the analysis module of the present invention calculates the viewing distance based on the original image.
Figure 3 is a diagram to explain the process by which the detection model evaluation unit evaluates the performance of the fog area detection model.
Figure 4 exemplarily shows the results of calculating the visibility distance using the fog area detection and visibility distance calculation system according to an embodiment of the present invention.
Figure 5 is a flowchart illustrating the operation process of the fog area detection and visibility distance calculation system according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. In the drawings, the proportions and dimensions of components may be exaggerated for effective explanation of technical content.

Terms such as "include" are intended to designate the presence of features, numbers, steps, operations, components, parts, or a combination thereof described in the specification, but do not include one or more other features, numbers, steps, operations, or configurations. It should be understood that this does not exclude in advance the possibility of the presence or addition of elements, parts, or combinations thereof.

Additionally, when a component is described as “above” it means above or below the component, and does not necessarily mean that it is located above the direction of gravity.

Additionally, when a component is described as being “connected” or “coupled” to another component, it does not only mean that the component is directly connected or coupled to the other component, but also indirectly through another component. It may also include cases where it is connected or combined.

In addition, terms such as first and second may be used to describe a certain component, but these terms are only used to distinguish the component from other components, and the essence or order of the component is determined by the term. It is not intended to limit the order or the like.

Figure 1 exemplarily shows a block diagram of a fog area detection and visibility distance calculation system according to an embodiment of the present invention. Figures 2a to 2d are shown to explain the process by which the analysis module of the present invention calculates the viewing distance based on the original image.

Referring to Figures 1 and 2, the fog area detection and visibility range calculation system (FVS) according to an embodiment of the present invention includes an analysis module (DM), a dataset generation unit (DG), and a detection model evaluation unit (VL). ) may include.

The analysis module (DM) may include a video collection unit (VR), a fog area detection unit (FD), and a visible distance calculation unit (VC). The analysis module (DM) can collect the original image (AA), detect the fog area (FS), and calculate the visibility distance based on this.

The video collection unit (VR) is installed in a fixed location and can collect original video (AA). The video collection unit (VR) may be at least one of CCTV, a camera, and a black box that can capture images of areas where fog frequently occurs. However, the video collection unit (VR) of the present invention is not limited to this, and any device capable of filming a specific area is sufficient. In addition, since the image collection unit (VR) is sufficient as a device capable of filming a specific area, the fog area detection and visibility range calculation system (FVS) of the present invention has the advantage of being able to be built at a low cost.

The original video (AA) collected from the video collection unit (VR) can be used to calculate the presence of eye development and visibility in the area in real time. The image collection unit (VR) can transmit the collected original image (AA) to the fog area detection unit (FD) (see FIGS. 2A and 2B).

The fog area detection unit (FD) can detect the fog area (FS) in the original image (AA) (see FIGS. 2A and 2C). In this process, the fog area detection unit (FD) can use a fog area detection model. The fog area detection model may be learned in advance based on an artificial intelligence algorithm. The fog area detection unit (FD) may transmit the first image (BB), which displays the fog area (FS) in the original image (AA), to the visible distance calculation unit (VC).

In one embodiment of the present invention, the artificial intelligence algorithm may be at least one of the You Only Look Once (YOLO) algorithm, the R-CNN algorithm, the Mask R-CNN algorithm, and the Transformer algorithm. The YOLO (You Only Look Once) algorithm is a deep learning algorithm used to classify objects in images or videos. The Mask R-CNN algorithm is based on the Faster R-CNN algorithm and is a deep learning algorithm that performs object detection and segmentation. Since the above-described algorithms are well-known techniques, detailed descriptions are omitted.

The visibility distance calculation unit (VC) may display a plurality of reference distance lines (SL) on the first image (BB) received from the fog area detection unit (FD) to generate the second image (CC). The visibility distance calculation unit (VC) calculates the visibility distance based on the second image (CC) depending on whether each of the plurality of reference distance lines (SL) is included in the fog area (FS) (FIGS. 2A and 2D reference) can be done.

Each of the plurality of reference distance lines (SL) is defined as a line marking points on the original image (AA) that are separated by the reference distance from the image collection unit (VR). Each of the plurality of reference distance lines (SL) may be secured in advance.

In the present invention, each of the plurality of reference distance lines (SL) can be secured in advance by estimating a non-linear function that defines the correlation between the original image (AA) and the reference distance. For this purpose, the installation location, height, and direction of the video collection unit (VR) can be investigated in advance.

Hereinafter, a method for estimating a nonlinear function will be briefly described for understanding of the present invention. First, two-dimensional coordinates (x, y) are assigned to each pixel of the original image (AA). Afterwards, the distance from the video collection unit (VR) to the point corresponding to the coordinates is collected using a laser distance measuring device. Based on the collected separation distance, a plurality of non-linear functions defining the correlation between the y-coordinate value and the above-mentioned separation distance are estimated for each of the plurality of vertical lines (y-axis) of the original image (AA). A reference distance line can be displayed on the original image (AA) using the plurality of nonlinear functions described above.

Hereinafter, the process by which the visibility distance calculation unit (VC) calculates the visibility distance using the reference distance lines (SL) will be described with reference to FIG. 2D.

The visibility distance calculation unit (VC) determines the visibility according to whether the fog area (FS) includes each of the first reference distance line (SL1), the second reference distance line (SL2), and the third reference distance line (SL3). Distance can be judged.

Each of the first reference distance line SL1, the second reference distance line SL2, and the third reference distance line SL3 may be one of the plurality of reference distance lines SL described above. The first reference distance is a reference distance corresponding to the first reference distance line SL1. The second reference distance is a reference distance corresponding to the second reference distance line (SL2). The third reference distance is a reference distance corresponding to the third reference distance line SL3. The first reference distance may be longer than the second reference distance, and the second reference distance may be longer than the third reference distance. In FIG. 2D, the first reference distance may be 1000m, the second reference distance may be 500m, and the third reference distance may be 200m. However, the above description is only an example and the reference distances of each of the first reference distance line (SL1), the second reference distance line (SL2), and the third reference distance line (SL3) of the present invention are not limited thereto.

If the first reference distance line (SL1) and the second reference distance line (SL2) are not included in the fog area (FS), the visibility distance calculation unit (VC) may determine that the visibility distance is longer than the first reference distance. there is.

When only the first standard distance line (SL1) of the first standard distance line (SL1) and the second standard distance line (SL2) is included in the fog area (FS), the visibility distance calculation unit (VC) determines that the visibility distance is the second standard distance line (SL1). It can be judged to be longer than the reference distance and shorter than the first reference distance.

When the first reference distance line (SL1) and the second reference distance line (SL2) are included in the fog area (FS), the visibility distance calculation unit (VC) may determine that the visibility distance is shorter than the second reference distance. .

Referring to FIG. 2D, among the first standard distance line (SL1), the second standard distance line (SL2), and the third standard distance line (SL3), the second standard distance line (SL2) and the third standard distance line (SL3) ) is included in the above-mentioned fog area (FS). Accordingly, the visibility distance calculation unit (VC) may determine that the visibility distance is longer than the third reference distance and shorter than the second reference distance through the above-described process. In one embodiment of the present invention, the reference distance of the first reference distance line (SL1) is 1,000 m, the reference distance of the second reference distance line (SL2) is 500 m, and the reference distance of the third reference distance line (SL3) is 200 m. You can. Therefore, the visibility distance calculation unit (VC) can ultimately determine that the visibility distance is 200 m or more and 500 m or less. By introducing a greater number of reference distance lines (SL), the visibility distance calculation unit (VC) can calculate the visibility distance more accurately.

The dataset generator (DG) can create a dataset for learning and evaluating the fog area detection model. The dataset generator DG may select a plurality of second images from a plurality of first images collected in advance. The first images may be images collected in advance for the purpose of learning and evaluating the fog area detection model of the present invention. Specifically, the first images are selected from images collected from ports, beaches, etc., only data where there are points where visibility is not secured.

Thereafter, the dataset generator DG may generate metadata by performing annotation on each of the plurality of second images. In one embodiment of the present invention, the dataset generator (DG) may be omitted.

In one embodiment of the present invention, the area of the object classified as fog in each of the plurality of second images may be 50% or more and 99.9% or less of the total area. That is, the dataset generator DG may select a plurality of second images in which the area of an object classified as fog is 50% to 99.9% of the total area from among the plurality of first images collected in advance. Through this, it is possible to create a high-quality dataset for learning and evaluating the fog area detection model. In addition, the dataset generator (DG) selects only images where the area of an object classified as fog is 50% to 99.9% of the total area and uses them as a dataset, so the dataset can be secured quickly.

In one embodiment of the present invention, metadata may be at least one of a bounding box, polygon, semantic segmentation, and polyline.

In one embodiment of the present invention, the dataset generator (DG) may perform annotation only for fog in each of the plurality of second images in the process of generating metadata. In other words, the dataset generator (DG) performs annotation only on fog, excluding objects other than fog, so the dataset can be secured quickly.

The dataset may include a plurality of second images and metadata corresponding to each of the plurality of second images. Datasets can be divided into training-datasets and test-datasets. The learning-dataset may be a dataset for training a fog area detection model. The test-dataset is a dataset to evaluate the performance of the fog area detection model, and is not used to learn the fog area detection model. The process of evaluating the performance of the fog area detection model will be described later.

In the fog area detection and visibility range calculation system (FVS) according to an embodiment of the present invention, the artificial intelligence algorithm used to learn the fog area detection model may be the YOLO (You Only Look Once) algorithm. Additionally, the dataset generator (DG) can use the Label Smoothing technique in the process of generating metadata.

The YOLO algorithm is an artificial intelligence algorithm for identifying objects in images and generating bounding boxes as metadata as a result. Since the object to be identified in the present invention is fog, that is, an atypical object, it is desirable to generate metadata in the form of a polygon. Therefore, when the YOLO algorithm is used in the fog area detection and visibility range calculation system (FVS) according to an embodiment of the present invention, the dataset generator (DG) generates polygon-shaped metadata using the label smoothing technique. It is desirable to do so.

Label Smoothing is one of the technologies for making artificial intelligence models general. Specifically, label smoothing refers to converting the class label used in the model's loss function into a smooth probability value. Label Smoothing can reduce the phenomenon of artificial intelligence models being excessively biased toward specific classes and encourage them to consider all classes equally.

In the fog area detection and visibility range calculation system (FVS) according to an embodiment of the present invention, the artificial intelligence algorithm used to learn the fog area detection model is the Mask R-CNN algorithm, and the metadata can be a polygon. there is. For example, the backbone of MASK R-CNN may be at least one of ResNet-101-FPN or ResNeXt-101-FPN. However, the backbone of MASK R-CNN used in the present invention is not limited to this.

Figure 3 is a diagram to explain the process by which the detection model evaluation unit (VL) evaluates the performance of the fog area detection model. Hereinafter, it will be described with reference to FIG. 3.

The detection model evaluation unit (VL) can calculate the fog area detection agreement rate based on the test-data set. The detection model evaluation unit (VL) can evaluate the performance of the fog area detection model based on the fog area detection agreement rate. The unit of fog area detection agreement rate is %, and the closer it is to 100%, the higher the performance of the fog area detection model can be judged to be. On the other hand, as the fog area detection agreement rate is less than 100%, the performance of the fog area detection model may be judged to be lower. In one embodiment of the present invention, the detection model evaluation unit (VL) may be omitted.

Unlike conventional techniques (which aim to classify types of objects), the fog area detection model of the present invention aims to detect an area (fog area (FS)) where atypical objects (fog) are distributed. . Therefore, in order to evaluate the performance of the fog area detection model, the annotation results of the dataset generator (DG) for the test-dataset and the detection results of the fog area detection model must be compared. Hereinafter, the above-described process will be described in detail.

The test-dataset may include a plurality of third images that are part of the plurality of second images and metadata corresponding to each of the third images. The test-dataset may be a dataset for evaluating the performance of the fog area detection model.

The first area (AR1) is defined as an area that the dataset generator (DG) annotated on the third image and classified as a fog area (FS). The first area is defined as the area of the first area AR1.

The second area AR2 is defined as an area detected as a fog area in the third image by the fog area detection model. The second area is defined as the area of the second area AR2.

The third area is defined as an area where the first area AR1 and the second area AR2 overlap. The third area is defined as the area of the third region.

The detection model evaluation unit (VL) may calculate the fog area detection agreement rate based on the first area, second area, and third area.

In the fog area detection and visibility range calculation system (FVS) according to an embodiment of the present invention, the detection model evaluation unit (VL) can calculate the fog area detection coincidence rate using Equation 1 below. Using Equation 1, the performance of the fog area detection model can be evaluated by reflecting not only the case where the first area AR1 is larger than the second area AR2 but also the case where the first area AR1 is smaller than the second area AR2.

[Equation 1]

Figure 4 exemplarily shows the results of calculating the visibility distance using a fog area detection and visibility distance calculation system (FVS) according to an embodiment of the present invention.

Figure 5 is a flow chart illustrating the operation process of the fog area detection and visibility range calculation system (FVS) according to an embodiment of the present invention.

Referring to FIG. 5, the operation process of the fog area detection and visibility range calculation system (FVS) of the present invention will be briefly described. First, the fog area detection unit (FD) can detect the fog area (FS) in the original image (AA). The visibility distance calculation unit VC may determine the visibility depending on whether each of the plurality of reference distance lines SL1, SL2, and SL3 is included in the fog area FS. Specifically, the visibility distance calculation unit (VC) sequentially compares the fog area (FS) with the first reference distance line (SL1), the second reference distance line (SL2), and the third reference distance line (SL3) to provide visibility. Distance can be judged.

For example, when the first reference distance line SL1 is not included in the fog area FS, the visibility distance calculation unit VC may determine that the visibility distance is longer than 1 km. When the first reference distance line (SL1) is included in the fog area (FS) and the second reference distance line (SL2) is not included in the fog area (FS), the visibility distance calculation unit (VC) determines that the visibility distance is 500m. It can be judged to be longer than 1Km and shorter than 1Km. When the first reference distance line (SL1) and the second reference distance line (SL2) are included in the fog area (FS), and the third reference distance line (SL3) is not included in the fog area (FS), the visibility distance is calculated. VC can be judged to have a visibility longer than 100m and shorter than 500m. When the first reference distance line (SL1), the second reference distance line (SL2), and the third reference distance line (SL3) are included in the fog area (FS), the visibility distance calculation unit (VC) determines that the visibility distance is 100 m. It can be judged to be shorter.

Although the description has been made with reference to examples, those skilled in the art will understand that various modifications and changes can be made to the present invention without departing from the spirit and scope of the present invention as set forth in the following patent claims. There will be. In addition, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention, and all technical ideas within the scope of the following patent claims and equivalents should be construed as being included in the scope of the present invention. .

FVS: Fog area detection and visibility range calculation system
DM: Analysis module VR: Video collection department
VR: Fog area detection unit VC: Visible distance calculation unit
DG: Dataset creation unit VL: Detection model evaluation unit
FS: Fog area SL: Base distance line

Claims (10)

A video collection unit installed at a fixed location to collect original video;
a fog area detection unit that detects a fog area in the original image using a fog area detection model learned in advance based on an artificial intelligence algorithm; and
Each of the plurality of reference distance lines is defined as a line marking points on the original image that are separated by a reference distance from the image collection unit, and the visibility distance depends on whether each of the plurality of reference distance lines is included in the fog area. It includes a visible distance calculation unit that calculates,
Each of the first reference distance line and the second reference distance line is one of the plurality of reference distance lines, and the first reference distance, which is a reference distance corresponding to the first reference distance line, is a reference distance corresponding to the second reference distance line. It is longer than the second reference distance, which is the distance,
The visible distance calculation unit,
When the first reference distance line and the second reference distance line are not included in the fog area, it is determined that the visibility distance is longer than the first reference distance,
When only the first reference distance line among the first reference distance line and the second reference distance line is included in the fog area, the visibility distance is determined to be longer than the second reference distance and shorter than the first reference distance; ,
When the first reference distance line and the second reference distance line are included in the fog area, it is determined that the visibility distance is shorter than the second reference distance,
The artificial intelligence algorithm is at least one of the YOLO (You Only Look Once) algorithm, the R-CNN algorithm, the Mask R-CNN algorithm, and the Transformer algorithm,
It further includes a dataset generator that generates a dataset for learning and evaluating the fog area detection model,
The dataset generator selects a plurality of second images from a plurality of first images collected in advance, and generates metadata by performing annotation on each of the plurality of second images,
The dataset includes the plurality of second images and metadata corresponding to each of the plurality of second images,
The dataset is divided into a learning-dataset for training the fog area detection model and a test-dataset for evaluating the performance of the fog area detection model,
It further includes a detection model evaluation unit that calculates a fog area detection agreement rate based on the test-data set and evaluates the performance of the fog area detection model based on the fog area detection agreement rate,
The test-dataset includes a plurality of third images that are part of the plurality of second images and metadata corresponding to each of the third images,
The area of the first area classified as a fog area by the dataset generator annotating the third image is defined as the first area,
The area of the second area detected by the fog area detection model of the fog area detection unit as a fog area in the third image is defined as the second area,
The area of the third area where the first area and the second area overlap is defined as the third area,
A fog area detection and visibility distance calculation system wherein the detection model evaluation unit calculates a fog area detection agreement rate based on the first area, the second area, and the third area. delete delete According to claim 1,
The metadata generated by the dataset generator is at least one of a bounding box, polygon, semantic segmentation, and polyline. Fog area detection and visibility range calculation system . According to clause 4,
A fog area detection and visibility distance calculation system in which the area of an object classified as fog in each of the plurality of second images is 50% to 99.9% of the total area. According to clause 4,
A fog area detection and visibility distance calculation system in which the dataset generator annotates only fog in each of the plurality of second images in the process of generating the metadata. According to clause 4,
The artificial intelligence algorithm is the YOLO (You Only Look Once) algorithm,
The dataset generator is a fog area detection and visibility distance calculation system that uses a label smoothing technique in the process of generating the metadata. According to clause 4,
The artificial intelligence algorithm is the Mask R-CNN algorithm, and the metadata is a polygon. Fog area detection and visibility distance calculation system. delete According to claim 1,
A fog area detection and visibility distance calculation system in which the detection model evaluation unit calculates the fog area detection agreement rate using Equation 1 below.
[Equation 1]