KR102468753B1 - Fine dust detecting solution and system by computing RGB channel volume residual based on AI - Google Patents

Abstract

The present invention relates to a fine dust reading solution and system through artificial intelligence-based RGB channel residual calculation, and the fine dust reading solution through artificial intelligence-based RGB channel residual calculation according to the present invention acquires video indoors or outdoors. a video acquisition step; a continuous image extraction step of extracting a plurality of continuous still images from the video; an image conversion step of dividing data of each of the still images by RGB channels to generate an R channel image, a G channel image, and a B channel image; a residual calculation step of calculating residual data between two consecutive images in at least one of the R channel image, the G channel image, and the B channel image; and a fine dust level calculation step of calculating the level of fine dust based on the residual data.
Accordingly, it is possible to economically measure the level of fine dust in various places.

Description

Fine dust detecting solution and system by computing RGB channel volume residual based on AI}

The present invention relates to a fine dust reading solution and system through artificial intelligence-based RGB channel residual calculation, and more particularly, to an artificial intelligence-based RGB channel residual calculation capable of economically measuring fine dust levels using images. It is about a fine dust reading solution and system through

Recently, fine dust has become a social issue, and various media such as news provide the concentration of fine dust for each region.

Existing fine dust measurement methods include an indirect measurement method using physical properties of radiation or light (beta absorption method or light scattering method, etc.) and a method of measuring the mass of fine dust with a scale (weight concentration method, etc.). , In order to measure fine dust in this way, expensive measuring equipment is required.

On the other hand, the measurement of fine dust is measured for each administrative district, but even in the same administrative district, there is a difference in the actual fine dust measurement value depending on the topographical characteristics or the density of buildings.

Therefore, there is a need for a method for measuring the concentration of fine dust in various areas without expensive fine dust measuring equipment.

KR 10-2020-0009707 A

Therefore, an object of the present invention is to solve such a conventional problem, and a fine dust reading solution through artificial intelligence-based RGB channel residual calculation capable of economically measuring the fine dust level at various locations by using an image, and in providing the system.

The problem to be solved by the present invention is not limited to the above-mentioned problem, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

The above object is, according to the present invention, a video acquisition step of acquiring a video indoors or outdoors; a continuous image extraction step of extracting a plurality of continuous still images from the video; an image conversion step of dividing data of each of the still images by RGB channels to generate an R channel image, a G channel image, and a B channel image; a residual calculation step of calculating residual data between two consecutive images in at least one of the R channel image, the G channel image, and the B channel image; It is achieved by a fine dust reading solution through artificial intelligence-based RGB channel residual calculation including; and a fine dust degree calculating step of calculating the degree of fine dust based on the residual data.

In the step of calculating the degree of fine dust, the degree of fine dust may be divided into classes and calculated.

In the step of calculating the degree of fine dust, the degree of fine dust may be calculated through a pair of influence variables consisting of two variables in which the dominance of frequencies in different classes is reversed.

The pair of influence variables may be determined as a value when the resultant value of Equation 1 below becomes maximum.

[Equation 1]

: 두 변수가 x, y일 때 미세먼지 정도가 낮은 클래스에서의 관심도와 미세먼지 정도가 높은 클래스에서의 관심도 차이의 절대값,

: 두 변수가 x, y일 때 미세먼지 정도가 낮은 클래스에서의 관심도,

: 두 변수가 x, y일 때 미세먼지 정도가 높은 클래스에서의 관심도)(

: When the two variables are x and y, the absolute value of the difference between the degree of interest in the class with a low degree of fine dust and the degree of interest in a class with a high degree of fine dust,

: When the two variables are x and y, the degree of interest in the class with a low level of fine dust,

: When the two variables are x and y, the degree of interest in the class with a high level of fine dust)

If the result values of Equation 1 are the same, the two variables in the case where the sum of the frequencies of the two variables in each class is maximum can be determined as the pair of influence variables.

The pair of influence variables may be derived by learning the residual data through an artificial intelligence analysis model.

In the step of calculating the degree of fine dust, the degree of fine dust may be calculated using three or more odd-numbered pairs of influence variables.

In the step of calculating the degree of fine dust, the degree of fine dust calculated through the pair of influence variables is determined again through a pair of second influence variables composed of two second variables in which the dominance of frequencies between each other in different subclasses is reversed. It may include a detailed class calculation step.

The fine dust reading solution through the artificial intelligence-based RGB channel residual calculation according to the present invention is residual data excluding components of the residual data that are greater than or equal to a predetermined value, which are progressed between the residual calculation step and the fine dust degree calculation step. A filtering step may be further included.

According to another embodiment of the present invention, a video acquisition unit for obtaining a video indoors or outdoors; a continuous image extraction unit extracting a plurality of continuous still images from the video; an image conversion unit configured to separate data of each of the still images by RGB channels to generate an R channel image, a G channel image, and a B channel image; a residual calculation unit calculating residual data between two consecutive images in at least one of the R channel image, the G channel image, and the B channel image; There is provided a fine dust reading system through artificial intelligence-based RGB channel residual calculation including; and a fine dust degree calculation unit that calculates the degree of fine dust based on the residual data.

The fine dust degree calculation unit calculates the degree of fine dust by dividing it into classes, and the degree of fine dust can be calculated using a pair of influence variables consisting of two variables in which the dominance of frequency between each other in different classes is reversed.

According to the fine dust reading solution through artificial intelligence-based RGB channel residual calculation according to the present invention, the level of fine dust can be economically measured, and the fine dust level can be measured in several places due to this economy, so that local fine dust to know the level.

In addition, in the process of measuring the fine dust level, it is possible to shorten the measurement time by separating each still image for each RGB channel.

In addition, in the process of measuring the level of fine dust, it is possible to easily determine the degree of fine dust using a pair of influencing variables.

1 is a flowchart of a fine dust reading solution through artificial intelligence-based RGB channel residual calculation according to the present invention;
Figure 2 is an explanatory view related to the image conversion step constituting the fine dust reading solution through artificial intelligence-based RGB channel residual calculation according to the present invention;
3 is an explanatory diagram of an analysis model implementing an algorithm of influencing variable pairs used in a fine dust reading solution through artificial intelligence-based RGB channel residual calculation according to the present invention;
4 is an explanatory view related to the detailed class calculation step constituting the fine dust reading solution through artificial intelligence-based RGB channel residual calculation according to the present invention;
5 is a schematic configuration diagram of a fine dust reading system through artificial intelligence-based RGB channel residual calculation according to the present invention.

Hereinafter, specific embodiments of the present invention will be described in detail with reference to the drawings.

1 shows a flow chart of a fine dust reading solution through artificial intelligence-based RGB channel residual calculation according to the present invention.

The fine dust reading solution through artificial intelligence-based RGB channel residual calculation according to the present invention includes a video acquisition step (S10), a continuous image extraction step (S20), an image conversion step (S30), a residual calculation step (S40) and fine dust A degree calculation step (S50) is included.

In the video acquisition step (S10), a video of a place where the level of fine dust is to be measured is obtained.

The video may have a length of about 5 seconds and may be captured by a digital video camera or mobile phone. In the present invention, the fine dust reading solution through artificial intelligence-based RGB channel residual calculation can be applied both indoors and outdoors.

In the continuous image extraction step (S20), a plurality of continuous still images are extracted from the video taken in the video acquisition step (S10).

A moving picture actually consists of a number of still images, each of which appears in sequence for a short period of time, with minute changes. In the continuous image extraction step (S20), each still image constituting the video is extracted. For example, when a video is captured at 15 fps, 15 frames per second may be extracted from a still image.

Light may collide with fine dust and be scattered, and the possibility of light colliding with fine dust increases as the concentration of fine dust increases. And although it is not easy to check with the naked eye, light scattering by fine dust can be captured through a camera.

However, since fine dust is not stationary in the air, light is scattered differently by fine dust moment by moment, and light in different states is captured in each still image extracted from a video. The higher the concentration of fine dust, the larger the difference between each still image.

In the image conversion step (S30), as shown in FIG. 2, each still image data is separated by RGB (Red, Green, Blue) to generate an R channel image, a G channel image, and a B channel image.

While a still image has R channel pixel values between 0 and 255, G channel pixel values, and B channel pixel values, R channel image, G channel image, and B channel image each have only one channel pixel value, so R The data size of each of the channel image, G channel image, and B channel image is reduced compared to the still image.

Therefore, when any one of the R channel image, the G channel image, and the B channel image is used to measure the fine dust level, it is possible to shorten the image processing or analysis time during the measurement process.

Since there are many still images extracted from a moving image, and each still image is composed of a large number of pixels, conversion to each channel image can have a very effective effect.

In the residual calculation step (S40), residual data between two consecutive images in at least one of the R channel image, the G channel image, and the B channel image is calculated.

As described above, light in a different state is captured in each still image due to light scattering by fine dust, and this state of light is reflected in each channel image generated by converting the still image. Therefore, a difference, that is, a residual occurs between two consecutive images of the same channel.

Residual data can be calculated by the following equation. That is, it can be calculated by obtaining a value difference between pixels at the same location in two consecutive images (t-th image and t+1-th image) of the same channel.

: i행, j열의 잔차,

: t번째 이미지에서 i행, j열의 값,

: t+1번째 이미지에서 i행, j열의 값)(

: Residuals in row i, column j,

: Value in row i and column j in the tth image,

: Value of row i, column j in the t+1th image)

As described above, the higher the concentration of fine dust, the larger the difference between each still image, so that the residual appears clearly when the concentration of fine dust is high.

For example, 75 still images may be extracted from a 5-second video shot at 15 fps, and 74 residual data may be calculated from 75 images of any one channel generated by converting each still image.

In the fine dust level calculation step (S50), the level of fine dust is calculated based on the residual data.

When the value of the residual is small, the concentration of fine dust is low, and when the value of the residual is large, it can be determined that the concentration of fine dust is high.

Reference data for determining the degree of fine dust can be created by subjecting moving images taken at different concentrations of fine dust to a process of calculating still images, generating images for each RGB channel, and calculating residual data. It is possible to increase the accuracy of the reference data by repeating the process several times.

Depending on which channel image of the R channel image, G channel image, and B channel image the residual clearly appears, in the fine dust level calculation step (S50), the degree of fine dust is calculated based on the residual data of the specific channel image, The work can be done quickly and easily.

Alternatively, it is also possible to increase the accuracy of fine dust calculation by calculating the degree of fine dust based on the residual data of all channel images.

According to the fine dust reading solution through the calculation of RGB channel residuals based on artificial intelligence of the present invention described above, it is economical because the level of fine dust can be measured using a general digital video camera without expensive measuring equipment. Since the level of fine dust can be measured in several places by this method, it is possible to know the local level of fine dust.

It is also possible to measure the level of fine dust without installing separate hardware equipment by using CCTV as a camera to obtain video.

In the fine dust degree calculation step (S50), the degree of fine dust may be divided into classes and calculated.

In this case, it is possible to intuitively determine whether fine dust is good or bad.

The criterion for classifying may be determined by applying a binary classification algorithm to a plurality of standard residual data.

Classes of fine dust can be divided into, for example, two classes of 'good' and 'bad', or four classes of 'good', 'normal', 'bad', and 'very bad'. .

In the fine dust degree calculation step (S50), the degree of fine dust may be calculated through a pair of influence variables consisting of two variables in which the dominance of the frequency between each other in different classes is reversed.

In this case, it is possible to easily calculate the class of the degree of fine dust by determining which of the two variables constituting the pair of influencing variables has a high frequency.

All of the residual data can be determined through a pair of influence variables, and a class of the degree of fine dust calculated with a relatively high frequency can be determined as the final result.

As an algorithm for deriving an influence variable pair, an influence variable pair algorithm may be used. By the influence variable pair algorithm, the influence variable pair can be determined as a value when the result value of Equation 1 below becomes maximum.

: 두 변수가 x, y일 때 미세먼지 정도가 낮은 클래스에서의 관심도와 미세먼지 정도가 높은 클래스에서의 관심도 차이의 절대값,

: 두 변수가 x, y일 때 미세먼지 정도가 낮은 클래스에서의 관심도,

: 두 변수가 x, y일 때 미세먼지 정도가 높은 클래스에서의 관심도)(

: When the two variables are x and y, the absolute value of the difference between the degree of interest in the class with a low degree of fine dust and the degree of interest in a class with a high degree of fine dust,

: When the two variables are x and y, the degree of interest in the class with a low level of fine dust,

: When the two variables are x and y, the degree of interest in the class with a high level of fine dust)

At this time, the degree of interest in each class can be calculated by the following equation.

,

,

: 해당 클래스(c)에서 변수

가

보다 낮을 확률,

: 변수 x의 빈도수,

: 변수 y의 빈도수)(

: variable in the class (c)

go

lower probability,

: frequency of variable x,

: frequency of variable y)

For example, if there is standard residual data with the following values for fine dust (c = low) and bad (c = high),

The frequency counts for variables constituting the standard residual data of each class are as follows.

< Frequency distribution table of variables constituting standard residual data of each class >

For example, if the two variables x and y are +1 and +5 based on the frequency distribution table above, the degree of interest in the class with a low degree of fine dust is 0, and the degree of interest in the class with a high degree of fine dust is 1 becomes

Therefore, when the two variables x and y are +1 and +5, the absolute value of the difference between the degree of interest in the class with a low degree of fine dust and the degree of interest in a class with a high degree of fine dust becomes 1, which is the maximum.

Therefore, the two variables +1 and +5 can be set as a pair of influence variables.

By these pairs of influence variables, if the frequency of the variable +1 is high in the residual data to be determined, the degree of fine dust can be determined to be low, and if the frequency of +5 is high, the degree of fine dust is judged to be high. can do.

Meanwhile, there may be a case where the absolute value of the difference between the degree of interest in a class with a low degree of fine dust and the degree of interest in a class with a high degree of fine dust is the same.

For example, based on the frequency distribution table above, the degree of interest in the class with a low level of fine dust is 0 and the degree of interest in the class with a high level of fine dust becomes 1,

The absolute value of the difference between the degree of interest in a class with a low degree of fine dust and the degree of interest in a class with a high degree of fine dust is 1.

In this case, the accuracy of calculating the degree of fine dust can be increased by setting the two variables in the case where the sum of the frequencies of the two variables is the maximum in each class as a pair of influencing variables.

The influence variable pair algorithm can be implemented by an analysis model to which artificial intelligence technology is applied. The analysis model may be various artificial intelligence models such as an artificial neural network (ANN), machine learning, and deep learning.

3 shows an example of an analysis model implementing the influence variable pair algorithm. The analysis model may include an input layer, a hidden layer (hidden, hidden 2), and an output layer. The number of hidden layers may differ from that shown.

In the input layer, values of components constituting residual data when fine dust is good (c = low) and values constituting residual data when fine dust is bad (c = high) are respectively input.

In the hidden layer (hidden, hidden 2), the component values of the residual data of each class are learned, and the absolute value of the difference between the degree of interest when the fine dust is good (c = low) and the degree of interest when the fine dust is bad (c = high) is maximum Find the two components (variables) that are

Then, the two components obtained in this way are output through an output layer to form a pair of influence variables.

The influence variable pair may be derived through the analysis model as described above, but the influence variable pair may be derived using another analysis model without being limited thereto.

In the fine dust degree calculation step (S50), the degree of fine dust may be calculated using three or more odd-numbered pairs of influence variables.

In this case, the class of fine dust degree calculated through more than half of the influencing variable pairs may be determined as the result of the final degree of fine dust. For example, when the degree of fine dust is calculated using three pairs of influencing variables, the degree of fine dust is determined to be high through two pairs of influencing variables, and the degree of fine dust is determined through one influencing variable pair. When is determined to be low, it can be finally determined that the degree of fine dust is high.

Since the method of calculating the degree of fine dust through pairs of influencing variables is based on probability, it is possible to increase the accuracy of the calculation by calculating the degree of fine dust through several pairs of influencing variables.

The fine dust level calculation step (S50) may include a detailed class calculation step.

In the detailed class calculation step, as shown in FIG. 4 , the degree of fine dust calculated through the pair of influence variables is divided into detailed classes and calculated. Classification into sub-classes can be made through a second influencing variable pair consisting of two second variables in which the dominance of frequencies in different sub-classes is reversed. That is, the method of classifying the degree of fine dust into detailed classes is different from the second variable constituting the second influencing variable pair compared to the method of calculating the degree of fine dust through the influencing variable pair, but the principle is the same. same.

It is natural that the second influencing variable pairs applicable to the cases in which the degree of fine dust is determined to be high and the degree of fine dust to be low by the influencing variable pair are different from each other.

If the degree of fine dust is classified as 'low' and 'high' by the pair of influencing variables, 'good', 'normal', 'bad', and 'very bad' by the second pair of influencing variables ' can be further subdivided.

Between the residual calculation step ( S40 ) and the fine dust degree calculation step ( S50 ), a residual data filtering step may be further performed.

In the residual filtering step, an operation of excluding components having a predetermined value or more from among residual data components is performed.

In addition to the scattering of light by fine dust, moving objects can be captured in the video, and the change in light caused by the movement of the object is large enough to be seen with the naked eye, resulting in a large residual compared to scattering of light by fine dust. will cause In addition, the residual due to the motion of the object acts as noise in calculating the degree of fine dust.

Therefore, the accuracy of calculating the degree of fine dust can be improved by excluding components having a size greater than or equal to a predetermined value among the residual data components.

Hereinafter, the fine dust reading system 1 through artificial intelligence-based RGB channel residual calculation according to the present invention will be described. 5 shows a schematic configuration diagram of the system 1 of the present invention.

While explaining the fine dust reading system 1 through artificial intelligence-based RGB channel residual calculation according to the present invention, detailed descriptions will be given of the parts mentioned in the description of the fine dust reading solution through artificial intelligence-based RGB channel residual calculation. omit it.

The fine dust reading system (1) through artificial intelligence-based RGB channel residual calculation according to the present invention includes a video acquisition unit (10), a continuous image extraction unit (20), an image conversion unit (30), and a residual calculation unit (40). and a fine dust level calculator 50 .

The video acquiring unit 10 acquires a video indoors or outdoors. The video acquisition unit 10 may be, for example, a digital video camera or a mobile phone.

The continuous image extraction unit 20 extracts a plurality of continuous still images from the video obtained through the video acquiring unit 10 . Due to the scattering of light by fine dust, different states of light may be captured in each still image.

The image conversion unit 30 separates each of the still image data by RGB channel to generate an R channel image, a G channel image, and a B channel image. Compared to a still image expressed in RGB (Red, Green, Blue), any one channel image has small data.

The residual calculator 40 calculates residual data between two consecutive images in at least one of the R channel image, the G channel image, and the B channel image. Depending on the degree of fine dust, the degree of light scattering varies, and the residual between two consecutive images in the same channel image changes according to the degree of light scattering. That is, the residual data varies according to the degree of fine dust.

The fine dust level calculator 50 calculates the level of fine dust based on the residual data. Since the residual data varies according to the degree of fine dust, it is possible to calculate the degree of fine dust through the residual data.

The fine dust degree calculation unit 50 calculates the degree of fine dust by dividing it into classes, and the degree of fine dust can be calculated using a pair of influence variables consisting of two variables in which the dominance of frequency between each other in different classes is reversed. .

In this case, it is possible to easily calculate the class of the degree of fine dust by determining which of the two variables constituting the pair of influencing variables has a high frequency.

The fine dust level calculator 50 includes an influence variable pair derivation unit (not shown) that derives an influence variable pair through an analysis model to which artificial intelligence technology is applied, and analyzes reference residual data from numerous videos to influence a pair of influence variables. can be derived. And the influence variable pair derived from the influence variable pair derivation unit is used to calculate the degree of fine dust.

Components of the fine dust reading system 1 through artificial intelligence-based RGB channel residual calculation may be implemented through one or more hardware or one or more software. Alternatively, it may be implemented through hardware and software.

The scope of the present invention is not limited to the above-described embodiments, but may be implemented in various forms of embodiments within the scope of the appended claims. Anyone with ordinary knowledge in the art to which the invention pertains without departing from the subject matter of the invention claimed in the claims is considered to be within the scope of the claims of the present invention to various extents that can be modified.

1: Fine dust reading system through AI-based RGB channel residual calculation
10: video acquisition unit 20: continuous image extraction unit
30: image conversion unit 40: residual calculation unit
50: Fine dust degree calculation unit

Claims (11)

A video acquisition step of acquiring a video indoors or outdoors;
a continuous image extraction step of extracting a plurality of continuous still images from the video;
an image conversion step of dividing the data of each of the still images by RGB channels to generate an R channel image, a G channel image, and a B channel image;
a residual calculation step of calculating residual data between two consecutive images in at least one of the R channel image, the G channel image, and the B channel image; and
Including; fine dust degree calculation step of calculating the degree of fine dust based on the residual data,
In the fine dust degree calculation step, the degree of fine dust is calculated by dividing into classes, and the degree of fine dust is calculated through a pair of influence variables consisting of two variables in which the dominance of frequency between each other in different classes is reversed. Fine dust reading solution through artificial intelligence-based RGB channel residual calculation.
delete delete According to claim 1,
The pair of influence variables is a fine dust reading solution through artificial intelligence-based RGB channel residual calculation, characterized in that the value when the resultant value of Equation 1 below becomes maximum.
[Equation 1]

(

: When the two variables are x and y, the absolute value of the difference between the degree of interest in the class with a low degree of fine dust and the degree of interest in a class with a high degree of fine dust,

: When the two variables are x and y, the degree of interest in the class with a low level of fine dust,

: When the two variables are x and y, the degree of interest in the class with a high level of fine dust)
According to claim 4,
If the result of Equation 1 is the same, the two variables when the sum of the frequencies of the two variables in each class are maximum are set as the influencing variable pair through artificial intelligence-based RGB channel residual calculation. Fine dust reading solution.
According to claim 4,
The influencing variable pair is a fine dust reading solution through artificial intelligence-based RGB channel residual calculation, characterized in that derived by learning the residual data through an artificial intelligence analysis model.
According to claim 1,
In the fine dust degree calculation step, the fine dust reading solution through artificial intelligence-based RGB channel residual calculation, characterized in that for calculating the degree of fine dust using three or more odd number of influence variable pairs.
According to claim 1,
In the step of calculating the degree of fine dust,
A detailed class calculation step of re-determining the degree of fine dust calculated through the pair of influencing variables through a second influencing variable pair consisting of two second variables in which the dominance of frequency between each other in different subclasses is reversed A fine dust reading solution through artificial intelligence-based RGB channel residual calculation, characterized in that.
According to claim 1,
Proceeding between the residual calculation step and the fine dust degree calculation step,
Fine dust reading solution through artificial intelligence-based RGB channel residual calculation, characterized in that further comprising a residual data filtering step of excluding components of a predetermined value or more among the components of the residual data.
a video acquisition unit that acquires video indoors or outdoors;
a continuous image extraction unit extracting a plurality of continuous still images from the video;
an image conversion unit configured to separate data of each of the still images by RGB channels to generate an R channel image, a G channel image, and a B channel image;
a residual calculation unit calculating residual data between two consecutive images in at least one of the R channel image, the G channel image, and the B channel image; and
A fine dust degree calculation unit for calculating the degree of fine dust based on the residual data; includes,
The fine dust degree calculation unit calculates the degree of fine dust by dividing it into classes,
The degree of fine dust is calculated using a pair of influence variables consisting of two variables in which the dominance of frequency between each other in different classes is reversed. A fine dust reading system through artificial intelligence-based RGB channel residual calculation.
delete

Images