Frontiers in Computing and Intelligent Systems 
ISSN: 2832-6024 | Vol. 6, No. 3, 2023 

 

19 

Blowout Fluid Velocity Measurement Method based on 
High-speed Sequence Images 

Zijian Teng, He Zhang 

School of Mechatronic Engineering, Southwest Petroleum University, Chengdu Sichuan, 610500, China 

 

Abstract: Given the high-temperature and high-pressure environment after the blowout, the staff usually cannot calculate and 

analyze various blowout parameters at close range and thus take effective measures to control accidents quickly. In this paper, a 

blowout fluid velocity measurement method is proposed based on high-speed sequence images, and simulation experiments are 

conducted. The scene conditions of blowout accidents are simulated and analyzed. On this basis, cameras are adopted to shoot a 

large number of simulated blowout fluid videos from different angles under outdoor conditions. Afterward, the video is converted 

into a multi-frame image, and the blowout fluid image feature point detection and extraction model are established based on the 

orb (oriented fast and rotated brief) algorithm. On the feature points extracted by the model, the gms algorithm is used to match 

the features of the blowout fluid images in the front and back frames. Additionally, the wrong matching points are eliminated 

using the ransac (random sample consensus) algorithm, and the obtained feature point pairs are employed into the flow velocity 

calculation model for flow velocity calculation. The results demonstrate that the measured fluid velocity is more accurate. The 

method proposed in this study can provide reliable data support for early blowout emergency rescue. 

Keywords: Blowout Velocity; Image Recognition; Feature Matching; Pinhole Imaging; Velocity Calculation. 

 

1. Introduction 

Generally, the operating environment of deep oil wells is 

extremely complex, with many unknown risks and changes. 

These factors reinforce the difficulty of mining in deep well 

environments. Safety risks are increasing with the increasing 

social demand for oil and natural gas and the continuous 

expansion of exploration and development fields. The 

occurrence of blowout accidents severely damages oil and gas 

resources, causes environmental pollution, and easily leads to 

fires, casualties, equipment damage, and even the scrapping 

of oil and gas wells. These also disrupt the normal production 

order [1,2]. When the blowout occurred, it is in a high-

temperature and high-pressure environment with violent and 

harmful reactions. Hence, the staff are unable to calculate and 

analyze various blowout parameters at close range and take 

effective measures to control the accident quickly. In other 

words, in-depth research on measurement techniques and 

equipment such as blowout fluid parameters should be 

conducted in-depth research on measurement techniques and 

equipment such as blowout fluid parameters, which is of great 

significance for early blowout emergency rescue [3-6]. 

Dong et al. carried out the research investigated on the 

equilibrium state of the underwater emergency sealing device, 

and simulated the analysis model, and obtained revealed the 

relationship between its flow velocity and the lifting force [7]. 

Jia et al. utilized the ultrasonic time-of-flight method to 

measure the fluid flow, demonstrating that the ultrasonic 

propagation time and the horizontal component of the 

propagation displacement were inversely and directly 

proportional to the flow velocity, respectively. They also 

calibrated the flow measurements using the tangent of 

refraction versus ultrasonic transit time [8]. Wang et al. 

designed a fluid flow rate measurement system based on the 

lms least mean square algorithm and a single-chip 

microcomputer as the main controller, and verified the 

feasibility of the system with actual test results [9]. Li et al. 

theoretically and numerically studied the lift-off distance and 

blowout limit of the boundary layer diffusion flame, and 

deduced the explicit solution of the flame lift-off distance and 

critical blowout limit [10]. Based on the robot volume ptv 

measurement method, C Jux et al. proposed an experimental 

method to measure the time-averaged fluid flow pressure on 

the surface of a general three-dimensional object for 

evaluating the pressure distribution on the surface of the 

object [11]. Sara Rashed et al. established a new approach to 

measure each component in oil-gas-water mixtures by 

deploying devices such as biplane electrical resistance 

tomography sensors, gradient pressure gauges, and flow 

density meters, so as to estimate fluid volumetric flow rates. 

They further deduce the water content and gas volume 

fraction based on the method [12]. With cameras, Wu Zhou et 

al. obtained the size and position of each imaged droplet using 

the depth-from-focus method and introduced a simple 

algorithm to estimate the size and position from the two-

camera images. The effect of the algorithm on the sensitivity 

of various system parameters was investigated through 

simulations using synthetic images, measurements with 

calibrated equipment, and measurements performed in a 

sparse spray [13]. Moreover, B. Rengel et al. presented an 

advanced equation to predict the self-extinguishing 

conditions of jet flames after blowout accidents and verified 

that it can determine the extinguishing speed of jet flames 

involving various gas fuels, wind conditions, flow regimes, 

and directions [14]. Bang et al. researched the effect of 

pressure on the characteristics of the rising flame in the co-

flow of propane fuel, discovering that the flame rising height, 

ground clearance height, and ejection velocity all decreased 

to varying degrees as the test pressure decreased [15]. 

After the fluid is ejected, the ground staff cannot approach 

and detect fluid parameters owing to the strong pressure and 

high temperature of the formation. Moreover, there is 

currently no instrument that can measure the blowout velocity 

at close range. In this paper, a blowout fluid velocity 

measurement method based on high-speed sequence images 

was proposed to quickly measure the flow velocity of blowout 



 

20 

fluid in the early stage of blowout accidents. This provides 

suggestions for well site staff to rapidly judge dangerous 

situations.  

2. Fluid Jet Flow Analysis and 
Research Ideas 

2.1. Analysis of Fluid Jet Flow Movement 

Generally, free jets have the property of jetting into bundles, 

and a bundled jet is formed above the spout. There is a large 

velocity gradient between the ejected fluid and the 

surrounding medium, and momentum exchange occurs 

between fluid particles and between adjacent fluid particles. 

The momentum of the free jet along the length direction 

remains unchanged, indicating that after the mass point of the 

ejected medium collides with the surrounding medium, the 

collided mass point obtains momentum and moves, though 

the momentum decreases. As a result, the sum of the 

momentum of the two remains constant. The reduction of 

kinetic energy is caused by the collision between the fast-

moving gas and the slow-moving medium that is driven up, 

where the lost kinetic energy is converted into heat. As the 

height and diameter of the jet increase, the jet transfers part of 

its momentum to the incoming fluid, and thus the velocity of 

the jet gradually decreases. Eventually, all the momentum of 

the jet disappears in the space fluid [16]. In order to simplify 

the complex medium motion of well jet flow, subsequent 

experiments used water flow to replace the well jet fluid. 

After the blowout jet forms a stable flow state, its jet shape 

and flow velocity distribution are illustrated in Figure 1. 

Initial stage Basic stage

Fluid flow 

core area

Spout

D

Inner 

boundary

Outer 

boundary

Turning 

section

0V

X
Fluid source

 
Fig 1. Velocity distribution of well jet flow 

 

In Figure 1, the jet boundary is regarded as a linear 

diffusion interface according to the statistical average 

significance. The velocity evolution process of the water jet 

is that the water flow is ejected from the spout with a diameter 

D at the initial velocity 𝑉0. When it moves along the X-axis 

for a certain distance, momentum exchange occurs between 

fluid particles, the jet boundary gradually widens, the velocity 

of the main body of the jet attenuates, and thus the area that 

still maintains the initial velocity gradually becomes smaller. 

These results are induced by the large velocity gradient 

between the ejected fluid and the surrounding medium. 

Generally, the initial velocity boundary of the jet is considered 

the inner boundary of the jet. At a certain height, the jet 

boundary layer expands to the central axis of the jet. At this 

time, only the velocity on the centerline of the jet maintains 

the initial velocity 𝑉0, and the cross-section at the centerline 

is viewed as a turning section. In other words, starting from 

this turning section, the surrounding fluid is entrained and 

drawn to the ejection direction to accelerate, while the jet 

boundary becomes wider and wider. Furthermore, the jet flow 

section from the wellhead to the turning section is defined as 

the jet flow initial section, the jet flow center velocity in this 

section is equal to 𝑉0, and the jet flow core area has an initial 

velocity 𝑉0. The part above the turning section is defined as 

the basic section of the jet flow. This section moves farther 

away from the turning section along the X-axis, and the center 

velocity is continuously attenuated along the central axis. The 

jet source opening angle is 18^°~26^° [17-19]. 

2.2. Research Ideas 

The overall research approach of this article is to simulate 

and analyze the on-site conditions of a blowout accident, 

using water flow to simulate the blowout fluid. Utilize high-

speed cameras to capture a large number of simulated blowout 

fluid videos outdoors, then convert the videos into multiple 

frames of images, and establish an image feature point 

detection and extraction model based on orb (oriented fast 

and rotated brief) algorithm. On the feature points extracted 

from the model, gms algorithm is used to achieve feature 

matching of fluid images before and after frames, and ransac 

(random sample consumus) algorithm is used to eliminate 

erroneous matching points. The final feature point pairs are 

substituted into the flow rate calculation model for flow rate 

calculation and compared with the standard flow rate for error 

analysis. The research idea of this paper is shown in Figure 2. 

Blowout fluid velocity measurement method 

based on high-speed sequence images

Design of blowout fluid 

velocity measurement scheme

Simulated blowout 

fluid image acquisition

Feature extraction of blowout fluid 

image based on ORB

Feature matching of blowout fluid 

image based on GMS

Mismatch elimination based on 

RANSAC

The blowout fluid velocity 

calculation

Calculate the blowout flow rate, compare it with 

the calibrated flow rate and analyze the error
 

Fig 2. Research plan 

3. Image Processing Model 
Establishment 

3.1. ORB Feature Detection Algorithm 

Orb is an image feature detection algorithm optimized 

based on the fast key point detector and the brief descriptor. 

Fast is widely applied for its high efficiency in feature point 

detection, while it is deficient in directionality. Therefore, a 

simple and effective method of intensity center-of-mass angle 

orientation measurement is introduced in the orb algorithm to 

equip the key points of the orb with direction components and 

rotation invariance. Moreover, a scale pyramid is adopted to 

generate fast features at each layer of the pyramid, 

contributing to the scale invariance of the orb algorithm [20].  

The orb algorithm adopts the gray-scale centroid method 



 

21 

to determine the main direction of the feature point. Firstly, 

the gray-scale centroid C in the neighborhood of the feature 

point is determined. Then, the geometric center O of the local 

area of the feature point is constructed to point to the direction 

vector of the gray-scale centroid C, expressed as:17,8.5 

𝑚𝑝𝑞 = ∑ 𝑥𝑝𝑦𝑝𝐼(𝑥, 𝑦)𝑥,𝑦              (1) 

where 𝐼(𝑥, 𝑦) represents the gray value at the pixel point 
(𝑥, 𝑦). 

At this time, the centroid of the image block is obtained by 

equation (2): 

𝐶 = (
𝑚10

𝑚00
,
𝑚01

𝑚00
)                  (2) 

where 𝑚00  denotes the zero moment; 𝑚01  and 𝑚10 

indicate the first-order moment. 

The feature point direction θ is defined as: 

𝜃 = 𝑎𝑟𝑐𝑡𝑎𝑛 (
𝑚01

𝑚10
)                  (3) 

After extracting oriented fast key points, its descriptor is 

calculated for each point. Orb uses a modified brief feature 

description, where brief is a binary descriptor whose 

description vector consists of many zeros and ones. The zeros 

and ones encode the size relationship between two random 

pixels (such as p and q) near key points. If p is larger than q, 

one is taken; otherwise, zero is taken. If 128 such p and q are 

taken, it ends up with a 128-dimensional vector consisting of 

0 and 1. brief takes the comparison of randomly selected 

points, which is fast and easy to store attributed to the use of 

a binary representation, and is suitable for real-time image 

matching. In this study, 256 was selected as the feature 

descriptor string length [21]. 

3.2. GMS Feature Matching Algorithm 

The gms (grid-based motion statistics) algorithm can 

improve the efficiency of the operation while ensuring the 

matching effect. The lack of apparent correct matches is 

induced by the difficulty in distinguishing between correct 

and incorrect matches, rather than the low number of 

matching pairs. Since the feature points in the correctly 

matched neighborhood frequently maintain geometric 

consistency, the correct and wrong matches can be 

distinguished by evaluating the number of matching points 

contained in the adjacent area of the matching pair to be 

judged [22], as detailed in Table 1. 

 

Table 1. GMS algorithm solves the following problems 

Problem Solution 

Grid scoring 
Divide the image into 20*20, 

numbered 1-400 

Robust Neighbor Grouping 

The eight squares surrounding a 

square serve as neighbors, with 

rotation and scale provided. 

Appropriate number of grids 

20*20 is suitable for 10000 

feature points. More feature 

points can divide more grids 

Separation of correct and false 

matches by thresholding 

The threshold is determined by 

the preset threshold and the 

average number of features of the 

9 adjacent squares. 

 

The sum of the total number of matches in the grid 

corresponding to the nearest neighbors (9 grids) in the other 

image corresponding to the grid corresponding to the nearest 

neighbors is calculated, written as:  

𝑆𝑖𝑗 = ∑ |𝑋𝑖𝑘𝑗𝑘|𝐾=9
𝑘=1                 (4) 

The exact and false matches are divided as: 

cell − pair{i, j} {
True, if Sij > τi = α√ni

Flase, otherwise             
      (5) 

where ni  denotes the average number of features 

mentioned in the table, and α represents a hyperparameter, 

which is generally set to 4~6. 

3.3. Random Sampling Consensus (RANSAC) 

Algorithm  

The random sample consensus (ransac) algorithm is a 

model-based robust iterative estimation algorithm, which can 

calculate and give a reasonable model parameter set in the 

case of high outliers in the data set. The calculation process is 

illuminated in Figure 3. The ransac algorithm is used when 

most of the observations match a model, while a small 

proportion of data (outliers) are distributed in the data space 

in the model, and the outliers deviate from the normal data by 

a small margin. The application is the elimination and 

correction of mismatched points after image feature matching 

[23]. 

4 groups of non-collinear data are randomly 

selected from the matching data set

Calculate the homography matrix H and 

record it as model M

Calculate the projection error e between all 

data in the data set and M

e< threshold？

Add data to interior point set I

Finish

Mark as outer point

?Iin  elements ofNumber bestI

K iterations and  Update IIbest =

No

Yes

Yes

 
Fig 3. ransac algorithm flow chart 

 

In, opencv the ransac algorithm eliminates the mismatched 

pairs in the existing image feature matching pairs by finding 

the best homography matrix 𝐻(3 × 3) . The pixel point 

transformation relationship of the two images described by H 

is described as: 

s [
𝑥′

𝑦′

1

] = [

ℎ11 ℎ12 ℎ13

ℎ21 ℎ22 ℎ23

ℎ31 ℎ32 ℎ33

] [
𝑥
𝑦
1
]             (6) 

where (𝑥, 𝑦) represents the corner position of the previous 

image, (𝑥′, 𝑦′)  indicates the corner position of the next 

image, and s refers to the scale factor. ℎ33 = 1  is the 

normalized matrix. 

Therefore, there are 8 unknown parameters in H, and at 

least 8 linear equations need to be established for a solution, 

namely, at least 4 matching pairs. In this paper, 4 groups of 

non-collinear data are randomly selected from the matching 

data set and then employed to calculate the homography 

matrix H. Finally, the matching point logarithm and 

projection error that satisfy this model are calculated, that is, 

the cost function e. The corresponding cost function is the 

smallest when the model reaches the optimum. 



 

22 

𝒆 = ∑ [(𝒙𝒊
′ −

𝒉𝟏𝟏𝒙𝒊+𝒉𝟏𝟐𝒚𝒊+𝒉𝟏𝟑

𝒉𝟑𝟏𝒙𝒊+𝒉𝟑𝟐𝒚𝒊+𝒉𝟑𝟑
)

𝟐

+ (𝒚𝒊
′ −𝒏

𝒊=𝟏

𝒉𝟐𝟏𝒙𝒊+𝒉𝟐𝟐𝒚𝒊+𝒉𝟐𝟑

𝒉𝟑𝟏𝒙𝒊+𝒉𝟑𝟐𝒚𝒊+𝒉𝟑𝟑
)

𝟐

]                (7) 

3.4. Establishment of Flow Velocity 

Calculation Model 

In the current development of camera technology, pinhole 

imaging is still the core idea of camera imaging, where the 

pinhole imaging model is generally adopted [24]. The 

principle of pinhole imaging is exhibited in Figure 4. Before 

the blowout velocity is calculated by applying the pinhole 

imaging method, the internal and external parameters of the 

camera needed to be first determined, as well as converting 

the external parameters from the pixel coordinate system to 

the world coordinate system [25].  

Op

Py

Oy
Oz

Ox

W 

W

Ix

Iy

f

O

Px Oi

Imaging plane

Camera

 
Fig 4. Pinhole imaging model 

 

In Figure 4, the camera coordinate system takes O as the 

camera optical center, the 𝑂𝑧, 𝑂𝑥 and 𝑂𝑦 axis as the optical 

axis, the horizontal axis, and the vertical axis, respectively. 

The coordinate system composed of the coordinate origin 𝑂𝑖, 

the abscissa 𝐼𝑥 , and the ordinate𝐼𝑦  is the image coordinate 

system, where 𝐼𝑥  and 𝐼𝑦  are parallel to 𝑂𝑥  and 𝑂𝑦  in the 

camera coordinate system, respectively and the distance 𝑂𝑂𝑖 

between the optical center O of the camera coordinate system 

and the origin 𝑂𝑖 of the image coordinate system is the focal 

length f. The coordinate system consisting of the coordinate 

origin 𝑂𝑝 , the abscissa 𝑃𝑥 , and the ordinate 𝑃𝑥  is a pixel 

coordinate system, where any point (𝑃𝑥𝑖 , 𝑃𝑦𝑖 ) in the 

coordinate system represents the coordinates of a pixel point 

in the image. However, the pixel coordinates should be 

converted to image coordinates since the pixel coordinate 

system failed to reflect the specific position of the target in 

the image [26]. It is assumed in this paper that the pixel 

coordinates of 𝑂𝑖  in the image coordinate system are 

(𝑃𝑥𝑜, 𝑃𝑦𝑜 ), and the pixel specification of the photosensitive 

device in the camera is 𝑑𝑥 × 𝑑𝑦 . Thus, the conversion 

relationship between the pixel coordinate system and the 

image coordinate system is obtained, expressed as: 

{
𝑰𝒙 = (𝑷𝒙 − 𝑷𝒙𝟎)𝒅𝒙
𝑰𝒚 = (𝑷𝒚 − 𝑷𝒚𝟎)𝒅𝒚

              (8) 

The homogeneous equation of its matrix is: 

[
I𝑥
I𝑦
1

] = [
𝑑𝑥 0 −𝑃𝑥0𝑑𝑥
0 𝑑𝑦 −𝑃𝑦0𝑑𝑦

0 0 1

] [
𝑃𝑥

𝑃𝑦

1

]         (9) 

The image formed in the image is set as point 𝑊’(𝐼𝑥, 𝐼𝑦) 

when the W coordinate of a point of the object in the camera 

coordinate system is (𝑂𝑥, 𝑂𝑦, 𝑂𝑧) . Subsequently, the 

conversion relationship between the image coordinate system 

and the camera coordinate system is described as: 

[

O𝑥

O𝑦

𝑂𝑧

1

] =

[
 
 
 
 
 
𝑂𝑧

𝑓⁄ 0 0

0
𝑂𝑧

𝑓⁄ 0

0 0 𝑂𝑧

0 0 1 ]
 
 
 
 
 

[
𝐼𝑥
𝐼𝑦
1

]          (10) 

The relationship with the pixel coordinate system is written 

as: 

[

O𝑥

O𝑦

𝑂𝑧

1

] =

[
 
 
 
 
 
(𝑂𝑧𝑑𝑥)

𝑓⁄ 0
(−𝑂𝑧𝑃𝑥0𝑑𝑥)

𝑓⁄

0
(𝑂𝑧𝑑𝑦)

𝑓⁄
(−𝑂𝑧𝑃𝑦0𝑑𝑦)

𝑓
⁄

0 0 𝑂𝑧

0 0 1 ]
 
 
 
 
 

[
𝑃𝑥

𝑃𝑦

1

]     (11) 

When camera calibration is performed, the obtained 

camera intrinsic parameters are denoted as 𝑘𝑥, 𝑘𝑦 and 𝑃𝑥0, 

𝑃𝑦0, where 𝑘𝑥 = 𝑓 ∗ (1/𝑑𝑥), 𝑘𝑦 = 𝑓 ∗ (1/𝑑𝑦). 

When the target scene is parallel to the camera plane, the 

target scene point 𝑊0 = (𝑂𝑥0, 𝑂𝑦0, 𝑂𝑧0) moves to 𝑊1 =

(𝑂𝑥1, 𝑂𝑦1, 𝑂𝑧1), which is equivalent to moving from 𝑊′0 =

(𝑃𝑥𝑜, 𝑃𝑦𝑜 ) to 𝑊1 = (𝑃𝑥1, 𝑃𝑦1 ) on the image. Based on the 

mapping relationship, equation (12) is derived as:  

{
𝑃𝑥1 − 𝑃𝑥0 =

𝑘𝑥(𝑂𝑥1−𝑂𝑥𝑜)

𝑂𝑧

𝑃𝑦1 − 𝑃𝑦0 =
𝑘𝑦(𝑂𝑦1−𝑂𝑦𝑜)

𝑂𝑧

           (12) 

The ratio of the pixel to the actual distance can be deduced 

after calibrating 𝑘𝑥  and 𝑘𝑦  by the camera and calculating 

the distance 𝑂𝑧  between the camera and the jet surface. 

Assuming that the pixel moving by the object on the imaging 

plane is (∆𝑚, ∆𝑛), the actual moving distance ∆𝑌 should be: 

∆𝑌 = √(
∆𝑚𝑂𝑧

𝑘𝑥
)

2

+ (
∆𝑛𝑂𝑧

𝑘𝑦
)

2

           (13) 

Finally, the time interval ∆t of the front and rear frames is 

deduced according to the frame rate of the camera. Hence, the 

initial velocity of the well jet can be calculated by: 

{
∆𝑌 = 𝜔0∆𝑡 −

1

2
𝑔(∆𝑡)2

𝜔0 = (∆𝑌 +
1

2
𝑔(∆𝑡)2)/∆𝑡

          (14) 

In this paper, the calculation of flow velocity is based on 

the front and rear frame images, and its feature points are 

extracted and matched. With the method of physical size 

conversion coefficient and lens imaging, the flow velocity is 

calculated through the pixel difference of the feature point 

displacement of the front and rear frame images, as well as 

the time difference between the front and rear frames. 

4. Experimental Analysis and 
Evaluation 

4.1. Simulated Blowout Fluid Velocity 

Measurement Scheme 

In this paper, two high speed industrial cameras are adopted 

as simulated blowout fluid image acquisition devices. The 

specific flow rate measurement process is detailed as follows. 

1) In an outdoor environment, two cameras are set at an angle 

of 90° to the blowout target line; 2) the lens parallel are set to 

the jet fluid axis section; 3) the camera and blowout 

simulation parameters of the device are set; 4) acquisition of 

blowout video; 5) after the video of the blowout fluid is 

collected by the camera, the video is transmitted to the host 



 

23 

computer through the local area network, and the blowout 

video is processed into an image sequence on the host 

computer for subsequent image feature detection operations. 

The blowout fluid feature points on the sequence images 

are extracted using the image extraction algorithm. Afterward, 

the key feature points are matched with the feature-matching 

algorithm. Next, the displacement difference between the 

front and rear feature matching points is calculated following 

the matched key feature points, and the blowout fluid velocity 

calculation is further realized under the time difference. 

Figure 5 exhibits a block diagram of blowout fluid velocity 

measurement. 

During the blowout video collection process, two WP-

GX180 cameras were placed in two orientations at an angle 

of 90° to the blowout target line. The blowout video 

acquisition was performed under the background 

optimization and the natural background while the focal 

length f of the two cameras, the image pixel size, the object 

distance d, and the camera height H remained constant. The 

acquisition scheme and parameter settings are listed in Table 

2. Afterward, the collected blowout video was processed into 

continuous frame images for flow velocity calculation. 
 

Fig 5. Blowout fluid velocity measurement block diagram 

 

Table 2. Blowout video acquisition scheme 

Camera 1: 

Video frame rate:610fps,1600*1100 

Camera 2: 

Video frame rate:610fps,1600*1100 

Object distance: 

d=245cm; 

Camera height: 

H=100.5cm 

Focal distance: 

f1=35mm 

Total height: 

h=165cm 

Speed 1(water 

pressure:4MPa) 

Object distance: 

d=244cm; 

Camera height: 

H=97cm 

Focal distance: 

f1=35mm 

Total height: 

h=137cm 

Speed 1(water 

pressure:4MPa) 

Speed 2(water 

pressure:3MPa) 

Speed 2(water 

pressure:3MPa) 

Speed 3(water 

pressure:2MPa) 

Speed 3(water 

pressure:2MPa) 

Speed 4(water 

pressure:1MPa) 

Speed 4(water 

pressure:1MPa) 

Focal distance: 

f2=75mm 

Total height: 

h=107.5cm 

Speed 1(water 

pressure:4MPa) 

Focal distance: 

f2=75mm 

Total height: 

h=93cm 

Speed 1(water 

pressure:4MPa) 

Speed 2(water 

pressure:3MPa) 

Speed 2(water 

pressure:3MPa) 

Speed 3(water 

pressure:2MPa) 

Speed 3(water 

pressure:2MPa) 

Speed 4(water 

pressure:1MPa) 

Speed 4(water 

pressure:1MPa) 

4.2. Blowout Simulation  

Through the analysis of the actual blowout characteristics, 

the blowout simulation test was conducted using the self-

developed blowout simulation device. Above all, the power 

supply, fluid tank, centrifugal pump, mass flow meter, nozzle, 

host computer, and other equipment were connected. Then, 

the two cameras were placed in two orientations, respectively 

at an angle of 90° to the intersection point of the nozzle 

connection point of the blowout simulation device to collect 

blowout fluid video. Figure 6 and Figure 7 display the 

blowout fluid video collection scheme and camera site layout, 

respectively.  
Fig 6. Flow of video acquisition of simulated blowout fluid 

 

During the experiment, the camera captures video images 

of the blowout fluid, as presented in Figure 8. The video 

image transmits the blowout fluid video data acquired to the 

host computer through wireless transmission, and the video 



 

24 

data is processed into a continuous frame image sequence. 

The processed image sequences are displayed in Fig. 9(a), (b) 

and Fig. 10(a), (b) for the image sequences under background 

optimization and the natural background of camera 1 and 

camera 2, respectively.  

 

 
Fig 7. Camera site layout 

 

   
Fig 8. Blowout fluid image 

 

 
(a) 

 
(b) 

Fig 9. Camera 1 image sequence 

 

 
(a) 

 
(b) 

Fig 10. Camera 2 image sequence 

4.3. Blowout Fluid Velocity Calculation 

The blowout fluid image data obtained by camera 1 and 

camera 2 were selected to simulate the image processing 

model proposed in this paper, so as to verify the reliability of 

the image obtained by the blowout simulation experiment and 

the accuracy of the image processing model in this paper. 

4.3.1. Fluid Image Feature Extraction 

The orb algorithm proposed in this paper is utilized to 

perform feature detection on the blowout fluid pictures of 

camera 1 and camera 2. The results are exhibited in Figure 11. 

(a), (b). 

   
(a) Camera 1 blowout fluid feature detection 

   
(b) Camera 2 blowout fluid feature detection 

Fig 11. Blowout fluid characteristic detection results 

4.3.2. Fluid Image Feature Matching 

After the blowout fluid image feature point detection, the 

blowout fluid image feature matching based on gms and 

ransac is performed with the detected feature point descriptor. 

Figure 12 renders the feature-matching results of the two 

frames before and after the blowout fluid image of camera 1 

and camera 2. 

 

 
(a) 

 
(b) 

Fig 12. Blowout fluid feature matching results 

4.3.3. Calculation of Fluid Velocity 

In this paper, the initial velocity of the blowout fluid of 3 

groups of different initial velocities taken when the focal 

length of the camera is 35mm and 75mm in the blowout 

simulation device is calculated. The performance of the flow 

velocity calculation method proposed in this paper is analyzed 

with respect to the calibrated flow velocity of the mass flow 

meter. The pixel position coordinates of feature points are 

illustrated in Figures 13 and 14. 



 

25 

 
(a) 

 
(b) 

 
(c) 

 
(d) 

Fig 13. Pixel distribution of feature points on the upper boundary 

of well jet flow when the camera focal length is 35 mm 

 
(a) 

 
(b) 

 
(c) 

 
(d) 

Fig 14. Pixel distribution of feature points on the upper boundary 

of well jet flow when the camera focal length is 75 mm 

 

Table 3. Calculation results of blowout velocity 

Experimental 

Samples 

f=35mm f=75mm 

Displacement 

(m*10-2) 

Time 

(s*10-3) 

Velocity 

(m/s) 
Displacement (m*10-2) 

Time 

(s*10-3) 

Velocity 

(m/s) 

Sample 1 3.553 1.7 21.02 3.606 1.8 20.93 

Sample 2 3.026 1.8 16.78 3.584 2.1 18.12 

Sample 3 2.683 1.7 15.63 3.460 2.5 14.63 

Sample 4 2.075 1.8 12.07 3.305 2.8 12.69 

Figures 13 and 14 detail the pixel movement of the feature points at the moment before and after the blowout fluid jet. 



 

26 

The vertical displacement distance of the feature point pixel 

is calculated following the feature-matching pairs of the front 

and back frames. Then, the pinhole imaging method is 

adopted to obtain the actual blowout fluid displacement 

distance. Since the blowout image is obtained by video 

conversion, the time interval of the blowout images of the 

previous and subsequent frames can be obtained. Therefore, 

the blowout velocity can be calculated through equation (14), 

as listed in Table 3. 

4.4. Comparative Analysis of Model Errors 

After the calculation of the blowout fluid flow rate by 

image processing combined with the pinhole imaging method, 

the fluid flow rate recorded by the mass flow meter in the 

blowout simulation device was converted into the nozzle flow 

rate and taken as the calibration for the well blowout fluid 

flow rate to verify the accuracy of calculated blowout fluid 

flow rate. The comparison results of the calculated blowout 

fluid flow rate value and the calibrated flow rate value are 

provided in Table 4. Figure 15 presents a comparison chart of 

the accuracy of the two. 

 

Table 4. Comparison results of flow rate accuracy of blowout fluid at different focal lengths 

Experimental Samples 
Calibrated flow rate 

(m/s) 

f=35mm f=75mm 

Velocity 

(m/s) 

Accuracy 

(%) 

Velocity 

(m/s) 

Accuracy 

(%) 

Sample 1 20.28 21.02 96.48 20.93 96.89 

Sample 2 17.84 16.78 94.06 18.12 98.45 

Sample 3 14.92 15.63 95.46 14.63 98.06 

Sample 4 12.45 12.07 96.95 12.69 98.11 

 
Fig 15. Comparison of blowout fluid flow velocity and calibration 

flow velocity at different focal lengths 

 

From Table 4 and Figure 15, it can be concluded that when 

the camera focal length f=35mm, its flow rate value is 

relatively close to the calibrated flow rate value, and its 

calculation accuracy is the lowest at 94.06% and the highest 

at 96.95%; When the focal length of the camera is f=75mm, 

its flow rate value is closest to the calibrated flow rate value, 

and its calculation accuracy is the lowest at 96.89% and the 

highest at 98.45%. From this, it can be concluded that at a 

focal length of f=35mm, the clarity of the captured fluid 

image is low due to the small focal length, which has a certain 

degree of impact on the recognition and extraction of image 

feature points, and the accuracy of flow velocity measurement 

is low; When the focal length f=75mm, the accuracy of the 

calculated blowout fluid velocity value is relatively high. 

5. Summary 

Considering that it is difficult to collect on-site blowout 

pictures, a self-developed blowout simulation device was 

employed in this study to simulate blowout fluid and collect 

pictures. The characteristics of the blowout fluid were 

analyzed, and the initial velocity of the jet at the blowout 

nozzle was determined as the target for calculation. On this 

basis, the orb-gms-ransac blowout image feature point 

matching method was proposed. With the pinhole imaging 

method, the matching pair of upper edge features of the well 

blowout flow was utilized as the key point to calculate the 

blowout fluid flow rate. Finally, the cause of the error was 

analyzed by comparing it with the flow rate measured by the 

mass flow meter. The results unveiled that the flow rate of 

blowout fluid calculated by this image processing model has 

high accuracy and low error. This provides a reference for the 

well site staff to quickly judge the danger in the early stage of 

blowout accidents and thus is of great significance to the 

emergency rescue of an early blowout. 

References 

[1] Z. F. Liu, J. Q. Liu, Q. Z. Zhu and W. Wu. The weathering of 
oil after the Deepwater Horizon oil spill: insights from the 
chemical composition of the oil from the sea surface, salt 
marshes and sediments[J]. Environmental Research Letters 
2012 Vol. 7 Issue 3. 

[2] R. Islam, F. Khan and R. Venkatesan. Real time risk analysis 
of kick detection: Testing and validation[J]. Reliability 
Engineering & System Safety 2017 Vol. 161 Pages 25-37. 

[3] LIU Z, MA Q, CAI B, LIU Y, ZHENG, C. Risk assessment on 
deepwater drilling well control based on dynamic Bayesian 
network[J]. Process safety and environmental protection, 
2021,149: 643-654. 

[4] MA Q, ZHANG L. CFD simulation study on gas dispersion for 
risk assessment: A case study of sour gas well blowout[J]. 
Safety science, 2011,49(8-9): 1289-1295. 

[5] ZHU Y, CHEN G. Simulation and assessment of SO2 toxic 
environment after ignition of uncontrolled sour gas flow of well 
blowout in hills[J]. Journal of hazardous materials,2010,178(1-
3): 144-151. 

[6] J. Y. Xie, X. Y. Zhang, Y. W. Tang, Y. Wang, Q. Q. Shao and 
B. Yu. Transient simulation of the blowing-out process of the 
air pockets in vertical wellbore[J]. Applied Thermal 
Engineering 2014 Vol. 72 Issue 1 Pages 97-103. 

https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b


 

27 

[7] X. Dong, C. Zhang, C. Gu, J. Yang, M. Wang, Y. Wang. 
Balance analysis of the lowering process of 3000m subsea 
emergency well sealing device[J]. China Offshore Oil and 
Gas,2020,32(05):114-119. 

[8] H. Jia, C. Wang, R. Dang. Influence of fluid velocity on 
ultrasonic flow measurement accuracy and its calibration[J]. 
Journal of Instrumentation,2020,41(07). 

[9] R. Wang, M. Zhou, P. Yan, K. Zhang. Fluid Velocity 
Measurement System Based on LMS Algorithm[J]. Instrument 
Technique and Sensor, 2015(06):63-65. 

[10] S. P. Li, Q. Yao, C. K. Law, W. K. Liang and J. K. Zhuo. Liftoff 
and blowout of the Emmons flame: Analysis of the triple 
flame[J]. Combustion and Flame 2020 Vol. 215 Pages 184-192. 

[11] C. Jux, A. Sciacchitano and F. Scarano. Flow pressure 
evaluation on generic surfaces by robotic volumetric PTV[J]. 
Measurement Science and Technology 2020 Vol. 31 Issue 10. 

[12] S. Rashed, Y. Faraj, M. Wang and S. Wilkinson. Electrical 
resistance tomography-based multi-modality sensor and drift 
flux model for measurement of oil-gas-water flow[J]. 
Measurement Science and Technology 2022 Vol. 33 Issue 9. 

[13] W. Zhou, C. Tropea, B. T. Chen, Y. K. Zhang, X. Luo and X. 
S. Cai.Spray drop measurements using depth from defocus[J]. 
Measurement Science and Technology 2020 Vol. 31 Issue 7. 

[14] RENGEL B, PALACIOS A. Analysis of experimental blowout 
velocities of jet flames[J]. Combustion and flame, 2020,213: 
237-239. 

[15] BANG S, LEE B, CHUNG S. Effect of pressure on the 
characteristics of lifted flames[J]. Proceedings of the 
combustion institute, 2019,37(2): 2013-2020. 

[16] C. Hotz, M. Haas, S. Wachter, S. Fleck and T. Kolb. Two-phase 
free jet model of an atmospheric entrained flow gasifier[J]. 
Fuel 2021 Vol. 304. 

[17] H. T. Peng, W. Nie, S. B. Zhang, W. M. Cheng, Q. Liu, C. Guo, 
et al. Research on negative pressure jet dust-removal water 
curtain technology for coal mine cleaner production[J]. Fuel 
2022 Vol. 310. 

[18] S. Pucci, G. Poletto, A. C. Sterling and M. Romoli. Physical 
parameters of standard and blowout jets[J]. Astrophysical 
Journal. 2013 Vol. 776 Issue 1. 

[19] Z. Liu, Y. Zhang, Y. Zheng, F. Wang, X. Qian, D. Zhang, P. 
Huang, Q. Li, C. Feng. Characteristics of Blowout Flow Field 
in Natural Gas Wells Containing CO2[J]. Petroleum 
Exploration and Development, 2011,38(01):90-96. 

[20] Chuan L, Wei Y, Panling H and Jun Z. Overview of Image 
Matching Based on ORB Algorithm[J]. Journal of Physics: 
Conference Series, Volume 1237, Issue 3. 

[21] E. Rublee, V. Rabaud, K. Konolige, G. Bradski and Ieee. ORB: 
an efficient alternative to SIFT or SURF[C]. IEEE International 
Conference on Computer Vision (ICCV). 2011 Pages: 2564-
2571. 

[22] BIAN J, LIN W, LIU Y, ZHANG L, YEUNG S, CHENG 
M, REID I. GMS: Grid-Based Motion Statistics for Fast, Ultra-
robust Feature Correspondence[J]. international journal of 
computer vision,2020,128(6): 1580-1593. 

[23] G. Li, X. Huang and S. G. Li. A novel circular points-based 
self-calibration method for a camera's intrinsic parameters 
using RANSAC[J]. Measurement Science and Technology 
2019 Vol. 30 Issue 5.  

[24] B. A. Bircher, F. Meli, A. Kung and A. Sofiienko. Traceable x-
ray focal spot reconstruction by circular edge analysis: from 
sub-microfocus to mesofocus[J]. Measurement Science and 
Technology 2022 Vol. 33 Issue 7. 

[25] A. W. Winkler and B. G. Zagar. A curve fitting method for 
extrinsic camera calibration from a single image of a 
cylindrical object[J]. 

[26] B. Li, F. Wang, H. Lv, W. Wu, H. Wei. Microscopic imaging 
quality evaluation method of single-board computer lens based 
on pinhole imaging image plane parameter extraction[J]. 
Applied Optics,2021,42(05):839-847. 

 

 

 

https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://iopscience.iop.org/journal/1742-6596
https://iopscience.iop.org/journal/1742-6596
https://iopscience.iop.org/volume/1742-6596/1237
https://iopscience.iop.org/issue/1742-6596/1237/3
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b
https://www.webofscience.com/wos/alldb/general-summary?queryJson=%5b%7b