Intelligent Identification and Emissions Estimation of Harmful Gas

Based on Background Segregation Algorithm

Li Leyang

Guo Shuaichen

and Hao Peifeng

Software College of Northeastern Uinversity, ChuangXin Street 195, Shenyang, China

Shenyang No.2 High School , Wuai Street 6 , China

Keywords: Environmental protection, Smog particle recognizing, Background segregation, Emission.

Abstract: Environmental protection has become an important issue in modern enterprises, usually involving harmful

liquid and harmful gas to be detected. In order to address this issue, we propose a new intelligent algorithm to

identify and estimate the emission based on the color and smoke. To be specific, the three-frame difference

method is used to achieve the background segregation, and the area of the smoke is separated after the images

are proceeded by a mask. The color dictionary is employed to determine what color it is. Then, the background

separation algorithm and the physics knowledge are used to establish an estimation model. In the experiment,

we use the proposed approach to estimate harmful gas emissions in the real-world production process of

enterprises. Results validate the effectiveness and efficiency of the proposed algorithm, in terms of the

prediction accuracy.

1 INTRODUCTION

Security is an important issue in the production process

of modern enterprises. It is a basic way to alarm in time

in a dangerous situation, because there are a large

number of necessary chemical reactions, which will

produce much smoke in the producing process.

According to the color of smoke, it can be judged

whether it is a toxic gas. According to the area and

volume of the smoke, it can be estimated how much

gas has been generated. Accordingly, we can activate

an alarm by combining these two conditions. First, we

can use background separation algorithm to separate

the smoke into a set of pictures with different features,

and calculate the area of smoke . Then, the color can

be obtained by a mask. Finally, we use an estimation

model to calculate the volume of gas. The video after

the background separation can be used to display in a

page, and then the user can see the shape and position

of the smoke. The combination of these two methods

can detect harmful gases to ensure a safety.

Therefore, this paper proposes an intelligent smoke

detection algorithm and design a smoke detection

system based on it. Especially, the proposed algorithm

is able to detect the color and area of smoke, and judge

whether it requires an alarm. In fact, we use the video

as experimental data and extract a set of pictures from

the video frame by frame. In order to obtain the

background separation, we use the frame difference

method, which has exhibited a powerful real-time

performance. After background separation, the area

algorithm is used based on the video data to find out

the proportion of white, which is the percentage of

smoke in the image. Based on the above, this paper

combines the knowledge of both physics and

mathematics to establish a set of mathematical models,

which are to calculate the volume. In the process of

production, the camera is fixed and the area being

photographed is also fixed. As long as the total

percentage of smoke is required, the area can be

accurately calculated. When detecting the color, this

system uses a color image separated by the background

separation masked. Generally, the proposed algorithm

judges the main color via a customized color dictionary,

because different production processes produce

different colors. Especially, the color dictionary can

add various colors according to different production.

Finally, this paper is implemented in Python, supported

by OpenCV, and Python+OpenCV reduces code

complexity.

Leyang, L., Shuaichen, G. and Peifeng, H.

Intelligent Identiﬁcation and Emissions Estimation of Harmful Gas Based on Background Segregation Algorithm.

DOI: 10.5220/0008095800050009

In Proceedings of the International Conference on Advances in Computer Technology, Information Science and Communications (CTISC 2019), pages 5-9

ISBN: 978-989-758-357-5

2 BACKGROUND SEPARATION

BASED ON FRAME

DIFFERENCE METHOD

Background separation is the key process to realize the

system function, which it is used for the display in a

page, thereby it is necessary and irreplaceable. There

are two methods to obtain the background separation

of video. The first one is based on the Mixture Gauss

Principle. Especially, the background segmentation

can be realized by using three background dividers

including KNN, MOG2, GMG, which are provided by

the Background Subtractor class in OpenCV. This

method can obtain a high detection rate by using a

concise process. However, it also has some

shortcomings. For example, it would consume too

much system resources due to its inherent principle of

the implementation. When a static object moves

suddenly, some redundant objects can be partitioned

out. Therefore, this method makes the actual system

perform poorly in some real-time application scenarios.

The second one is based on the frame difference

method. In this approach, the frame difference method

can effectively solve the problem of low real-time

performance. However, due to the fact that the

boundary of smoke surface is usually not clear, the

separated image may encounter the dilemma of

burning into ghosting. In order to address the above

problem, we use three frame difference method for

background segmentation in this paper.

Assume that the current frame is S(t), the previous

frame is S (t-1), and the next frame is S(t+1) Then, the

image I

is obtained by S(t) and S(t-1) after using this

algorithm, and the image I

is obtained by S(t+1) and

S(t). We can obtain that I = I

is the final image.

After that, the picture is processed by the morphology,

which can eliminate ghosting. In order to reduce the

image noise and to highlight the edge features, we refer

to the idea of multi-objective optimization and global

optimization. The purpose of these methods is to find

an appropriate way to obtain a satisfactory detection

accuracy.

3 CALCULATION OF SMOKE

AREA BASED ON

BINARIZATION METHOD

The existing methods usually use the image contour to

calculate the area of smoke, which needs

morphological processing. These methods detect the

image contour by using the edge detection function,

and calculate the area of each contour by using self-

defined area function, and then accumulate it.

Because the camera captures only one angle of smoke,

the smoke area refers to the cross-sectional area seen

from the camera angle. The camera is installed above

the position of smoke. In this paper, the area is

calculated by a binarization method, and the final

image is obtained by morphological processing. The

morphological processing is orient to a binary image

and it uses median filter to clear the noises. Then, the

picture is traversed and the number of white pixels is

recorded. Note that, the length h and width w of the

picture are calculated by the called shape method. The

total number of white pixels is recorded as whitenum .

The proportion of white pixels in whole image can also

be called the proportion of smoke, represented as

Eq.(1). Because the camera is fixed, the area of smoke

can regard as whitenum, multiplying area

photographed.

𝑎𝑟𝑒𝑎 = 𝑤ℎ𝑖𝑡𝑒𝑛𝑢𝑚/(ℎ ∗ 𝑤) (1)

Compared with other algorithms, this method has

less codes and a more simple logic. This means the

code complexity is simplified by this method in

background separation.

4 SMOKE VOLUME ESTIMATION

MODEL

The smoke is discharged from a small outlet and it

gradually spreads outward and upward. In order to

estimate the volume of smoke, the shape of smoke can

be roughly regarded as a cylinder. The camera is

installed on the top position of the smoke, so the

bottom area height varies with time. When installing

monocular camera condition, the height of smoke

can’t be measured by image recognition, which

requires a mathematical model to estimate the

relationship between smoke emission and bottom area.

The height of smoke increases gradually at the

beginning of process, then stabilizes, and finally

gradually decreases. The changing process is shown in

Figure 1. Because of the Fourier's Law, the gas

emission can be estimated through the rule of the

concentration change. The concentration C x y z t( , , , )

can be expressed as Eq.(2). The trend of gas emission

with time is shown in Eq.(3) . Among them, Q is the

CTISC 2019 - International Conference on Advances in Computer Technology, Information Science and Communications

gas emission, which is a larger number, can equal 10

and k is a proportional coefficient.

Figure 1: Emissions change with time.

zyx

tzyxC

222

)4(

),,,(









(2)

ktr





ln4

(3)

Therefore, the relationship between smoke area

and time can be expressed as (4).

ktarea







ln4

(4)

The function of gas emission and smoke area can

be expressed as (5).

area









(5)

In conclusion, the smoke volume can be estimated

according to the smoke area.

5 EXPERIMENT ON COMPILING

COLOR DICTIONARY

In the experiment, we usually use the masking to cut

the area of interest in a given image. Masking in the

image processing looks like the process of PCB plate

making, which uses a specific picture to shade the

image. The used mask is a two-dimensional matrix

array. We can get final image by using various

operations. The masking process is as follows.

Figure 2: Mask operation process.

“&” is a common and simple operation, which can

use masks and original images to get the target area.

When judging the color of smoke, it is not necessary

to cut the smoke completely, only the part which

contains a large amount of smoke. The position of

smoke discharged from the production workshop is

fixed. Through the experiment, we selected the region

of 100 pixels around area where the smoke is

discharged, and use it as the mask, we set mask to 1,

the rest of the position is 0.

Some color detection methods by extract smoke

area, and normalize it to RGB space. Then, analyzing

the data of each color component. The way in this

paper of judge color through a customized color

dictionary. First, we define the upper bound and lower

bound of the color. The color dictionary is constructed

in RGB mode. The RGB value of the upper bound and

lower bound of color can be found in the drawing

software. Then, add the upper bound and lower bound

of a color to a list. Finally, take black as an example,

the following code defines a color.

lower_black = np.array([0, 0, 0])

upper_black = np.array([180, 255, 46])

color_list = []

color_list.append(lower_black)

color_list.append(upper_black)

dict['black'] = color_list

Traverse one picture, record the value of each color

pixel in the color dictionary, and compare the number

of each color pixel. The color of largest value is the

color of this picture.

6 EXPERIMENT RESULTS

In chapter two, we analysis some advantages in our

algorithm that others do not have. We found a video

on the Internet, and the comparison of experimental

results is as follows.

Figure 3: Original image. Figure 4: MOG2

Intelligent Identiﬁcation and Emissions Estimation of Harmful Gas Based on Background Segregation Algorithm

Figure 5: GMG. Figure 6: Frame difference method.

It can be found that MOG2 method has a very poor

effect and can hardly separate the smoke. GMG

method can separate smoke contour roughly, but still

can’t separate all the smoke. In contrast, using three

frame difference method to extract the foreground can

separate all the smoke, even slight smoke can be

extracted, which ensures accuracy for later calculation

of smoke area.

According to the experimental results in in terms of

the separated pictures and the resultant data, the

accuracy of color recognition can reach more than 99%.

The deviation between gas emission and actual gas is

shown in the following figure. It can be seen that most

of the estimated values are bigger than the actual

values, because the prediction model is calculated with

the maximum bottom area, so the value is bigger than

actual data, but such errors do not affect the volume

estimation.

Figure 7: Comparison between gas emission estimation and

actual gas production.

From the above, we can get the following observations:

(1) The frame difference method can generate a

better effect on these image set.

(2) The proposed mathematical model can

realize the estimation of smoke emission under the

condition of monocular camera, and it has a merit of

saving the cost.

7 CONCLUSIONS

In the processing of production, the traditional smoke

alarm system only detects whether there is smoke, but

can’t judge the toxicity and emissions of smoke, which

is ineffective to improve the accuracy of alarm. In this

paper, we propose an intelligent algorithm to

identify and estimate the emission based on the

color and smoke. In the experiment, the proposed

algorithm is used to estimate harmful gas emissions

in the real-world production process of enterprises.

The experimental results show that our algorithm can

identify the smoke more accurately, and judge the

color and emission more effectively This means our

algorithm obtain a desired performance, and it also

provides a new idea to identify the same type of defects.

ACKNOWLEDGEMENTS

This paper supported by National Natural Science

Foundation of China (6177021519,61503373), the

Fundamental Research Funds for the Central

Universities (N161705001）.

REFERENCES

Nana Wang 2012.Research on video based fire smoke

detection algorithm.Xi'an University of Science and

Technology.Xian.

Peng Chen 2013.Image background separation based on

Gauss fuzzy logic.Computer and modernization.

Jianbo Li 2014.Research on moving object detection

algorithm in video surveillance.Harbin Institute of

Technology.Harbin.

L. Ma, X. Wang, M.Huang, Z.Lin, L..Tian, H.Chen 2017.

Two-level Master-slave RFID Networks Planning via

Hybrid Multi-objective Artificial Bee Colony Optimizer,

IEEE Transactions on Systems, Man, and Cybernetics:

Systems, PP(99):1-20.

L. Ma, X. Wang, M. Huang, H. Zhang, H. Chen 2017.A

Novel Evolutionary Root System Growth Algorithm for

Solving Multi-objective Optimization Problems, Applied

Soft Computing. 57:379-398.

L. Ma, K. Hu, Y. Zhu, H. Chen 2014.Cooperative Artificial

Bee Colony Algorithm for Multi-objective RFID

Network Planning. Journal of network and computer

applications. 143-162.

L. Ma, H.Chen, X.Li, X.He,X.Liang 2016.Root system

growth biomimicry for global optimization models and

emergent behaviors, Soft computing.21(24):1-18.

H.Chen, L. Ma*, M. He, X. Wang* 2016.Artificial Bee

Colony Optimizer Based on Bee Life-cycle for

CTISC 2019 - International Conference on Advances in Computer Technology, Information Science and Communications

Stationary and Dynamic Optimization. IEEE

Transactions on Systems. Man, and Cybernetics:

Systems, 1-20.

L.Ma, X.Wang, R.Yu 2018.Biomimicry of plant root growth

using bioinspired foraging model for data clustering.

Neural Computing & Applications,29(3):1-18.

Lin Deng 2015.Research on video smoke detection

algorithm.University of Electronic Science and

technology of China.Chengdu.

Yongcai Hu 2017.Smoke detection algorithm based on low

illumination indoor video images.Xi'an University of

Science and Technology.Xian.

Zijian Liu 2014.Fire smoke diffusion model based on Gauss

distribution.Fire science and technology.

Lu Zheng,Junzhou Chen 2010.Video smoke detection

algorithm based on motion and color.Computer

engineering and design.

Yuehua Li, Lei Shan 2016.Fire smoke color detection

algorithm based on video sequence.Semiconductor

photoelectricity.

L. Ma, S.Cheng, X. Wang, M.Huang, H. Shen, X. He, Y. Shi

2017.Cooperative Two-engine Multi-objective Bee

Foraging Algorithm with Reinforcement Learning,

Knowledge-Based Systems,133:278-293.

L. Ma, Y. Zhu,Y. Liu, L. Tian 2015.A novel bionic algorithm

inspired by plant root foraging behaviors, Applied Soft

Computing, 37:95-133.

L. Ma, K. Hu, Y. Zhu, H. Chen 2015.A Hybrid Artificial Bee

Colony Optimizer by combining with life-cycle,

Powell’s search and crossover, Applied Mathematics and

Computation,vol.252, pp.133–154.

L. Ma, Y. Zhu, K. Hu 2014.A Novel Plant Root Foraging

Algorithm for Image Segmentation Problems,

Mathematical Problems in Engineering, Article ID

471209, 16 pages.

L. Ma, Li X, Gao T, et al 2016.Indicator-Based Multi-

objective Bacterial Foraging Algorithm with Adaptive

Searching Mechanism.Bio-Inspired Computing -

Theories and Applications. Springer Singapore, 271-277.

Wang R, Chen S, Ma L, et al 2018.Multi-indicator Bacterial

Foraging Algorithm with Kriging Model for Many-

Objective Optimization. Advances in Swarm Intelligence.

L. Ma, K. Hu, Y. Zhu, H. Chen 2014.Improved Multi-

objective Artificial Bee Colony Algorithm for optimal

power flow problem, Journal of Central South

University, 21(11): 4220-4227.

Chen S, Wang R, Ma L, et al 2018.A Novel Many-Objective

Bacterial Foraging Optimizer Based on Multi-engine

Cooperation Frameworz. International Conference on

Swarm Intelligence. Springer, Cham, 520-529.

Intelligent Identiﬁcation and Emissions Estimation of Harmful Gas Based on Background Segregation Algorithm