2010 International Conference on Measuring Technology and Mechatronics Automation

Size Detection of Firebricks Based on Machine Vision Technology

He Junji, Shi Li, Xiao Jianli, Cheng Jun, Zhu Ying
Shanghai Maritime University, Shanghai, 201306, China
[email protected]
limited, otherwise the detecting precision would be
degraded[3].
This paper proposed a method and system for fast and
automatic size detection of firebricks which is based on
monocular machine vision technology. The system takes a
picture of a firebrick and obtains the precise size of the brick
by image processing technique. It can improve the efficiency
and stability of the detection greatly, which can bring the
manufactories profit and improve the competitiveness of
their products.
The principle and structure of the system are introduced
in the first section. In the next section the image process
methods of the system are presented. The experiments and
results of the system calibration and detection are given in
the third section. In the last section, conclusion is given.

AbstractAt the present time, quality of firebricks is detected

manually one by one under a tough working environment. It is
urgent to invent an automatic device to do this job. In response
to this situation, this paper proposed a fast firebricks size
detecting method and device which was based on monocular
machine vision technology. The method is based on the
homographic principle between imaging plane and object
plane during monocular imaging process. Processing the
firebricks surface image from a high resolution camera by
image processing technique such as edge extracting, linking,
segmenting, fitting, etc, it is able to calculate the length and
width of a firebrick precisely. The test results proved that it
has obtained the precision of manual method at present time
and is very robust. The structure of the detecting system is
simple, efficient and practical.
KeywordsFirebricks;Size
Machine Vision;Image Processing

I.

Measurement;Monocular

II.

OF FIREBRICKS

Machine vision inspection technique is detecting

technology which is a simulation of human visual function.
It obtains required information by means of analysis and
calculation of the scene image. This technique is excellent in
extracting information of two-dimensional or threedimensional shape, contour, size, features and others of an
object. The technique has the merits of high speed, high
precision, abundant information, noncontact detection, etc.
which has a lot of successful applications of a variety of
objects and a number of industries [4,5]. The detecting
system proposed in this paper is a monocular machine vision
system whose structure belongs to the simplest one of
machine vision systems. Since the length, width and height
of firebricks to be detected is scalar parameters and they are
constants in respect to standard production of firebricks, the
monocular machine vision system is the most appropriate to
be adopted. Not only its structure is simple but also its
mathematical model, calculation method, calibration process
is simple and easy to be implemented. Therefore it is easy to
ensure the reliability of the system.
The basic structure of monocular machine vision system
as shown in figure 1 is composed of a camera and a
computer. The camera faces to one of the planar surfaces of
a detected object and takes a picture, then the image is
transferred to the computer where it is analyzed and
calculated to get the desired information. The principle it
based on is the homography of two planes in space. In other
words a point in a plane of the detected object and its
projection point in image plane is one to one mapping as
illustrated in figure 2. Point C in the figure is the center of
projection. The mapping is linear when represent in

INTRODUCTION

Firebricks are special materials which are used to lay a

blast furnace or a cement kiln. The special usage and high
temperature environment enforce special requirements of the
quality of firebricks. The error of their size, the defection of
their outside surface and inside structure and etc must to
some extant meet the quality requirements. Therefore it is an
important step to detect the quality of firebricks during their
production. Today manufactories detect their products one
by one manually and the detecting tools are simple steel
rulers. The outdated operation mode makes the speed of
detecting slow and the precision of measurement unstable.
At the same time it is a tough work for workers because of
the remaining heat and high weight of firebricks. It is urgent
for manufactories to adopt automatic detecting technology
instead of manual one with the ever increasing demands of
users.
At present, it is rare to find automatic detecting method
and devices around the world. Most of the existing detecting
devices are optical or photoelectric mechanical devices. One
method is invented to use optical fiber sensor to measure the
thickness of firebricks, which is proved to be high accurate.
However each sensor can only measure the thickness of a
firebrick[1]. Another method measures the thickness and the
flatness of tiles with CCD sensor[2]. It can measure the
thickness and flatness of a tile but requires fixed position of
a tile being detected on a convey belt. There is another
method which detects the size of a brick though the
projection of it from a fixed light source. It is a smart device,
however the position and angle to lay a brick are strictly

978-0-7695-3962-1/10 $26.00 2010 IEEE

DOI 10.1109/ICMTMA.2010.798

STRUCTURE AND PRINCIPLE OF DETECTING SYSTEM

394

homogeneous coordinates, which can be expressed as

follows[6]:
u h1
s v = h 4
1 h 7

h2
h5
h8

h3
h 6
h 9

X
X
Y = H Y
1
1

(1)

Where (u,v) is the coordinate of an image point in image

reference coordinate frames O1-UV, (X, Y) is the coordinates
of a point in object plane referenced in object plane
coordinate frames O-XY, s is a scale factor. H is the
homographic matrix of this mapping, which is in fact a
matrix composed of internal and external parameters of the
camera.

Figure 3. The system structure

III.

BASIC ALGORITHM OF FIREBRICKS IMAGE

PROCESSING

Most of the shapes of firebricks products are rectangle,

so this paper focuses on size measurement of rectangular
firebricks. As a firebrick is usually compressed out of a
mold, the surfaces of it are relatively rough. The edge and
corner of it are often worn out slightly within the extant of
tolerance, which makes the image process more complex
than expected. In addition to adoption of suitable lighting
mode to get clear images, a reasonable and stable image
processing algorithm must be invented. After analyzing the
characteristics of the edge of firebricks in detail, a four-step
method was proposed which contains edge extracting, edge
linking, edge segmenting and edge fitting.
A. Edge extraction of firebricks image
After reading the image and filtering out noises, we
extract edge from firebricks image by canny edge detecting
algorithm. An edge image of single pixel width is obtained
as shown in figure 4. Most of the actual edges are extracted
from the firebrick image as shown in the figure. In addition,
the edges of some texture which belongs to disturbing
variable on the surface of the brick are also extracted. The
edges of miniature worn part of corners are extracted too,
which are usually irregular curve. It is a disturbing variable
which destroys the edge fitting of the whole contour of a
brick. This kind of disturbance can be reduced by adjusting
the projecting angle and brightness of the light as well as
increasing the color contrast between the conveyor belt and
bricks.

Figure 1. Principle of monocular machine vision system

Figure 2. Mapping between the two planes

In actual application, an additional lighting source is

always added to the system in order to get a clear image. The
actual structure of the system is illustrated in figure 3.
Figure 4. Image after edge extraction

395

Figure 5. Image after edge linking

Figure 7. Image of calibration board

B. Edge linking of firebrick image

The edges gotten from the above steps are actually
discrete edge points so that edge linking must be performed
before edge fitting. In the image of discrete edge points
eight-neighborhood searching can be done and adjacent edge
points are considered to be on the same edge line. Several
sets of edge points which represent several corresponding
edge lines with different length are given after the searching.
There is many searching method and we used deep-first
algorithm. Since the textures on a firebricks surface always
have granular profile, the characteristics of their edges are
large curvature and short length. Therefore most of the edges
of textures on surface can be eliminated from the searching
results according to the length and curvature of edge lines.
Figure 5 shows the image after linking and elimination. The
thick lines of different gray in the figure represent the
qualified edge lines being searched.

IV.

EXPERIMENT AND RESULTS

A. Calibration experiment and results of monocular

machine vision system
The calibration task of the vision system is to determine
the values of each element in the homographic matrix H.
The common practice is to take the picture of planar
calibration target with known size, select 4 or more feature
points with image coordinates (ui, vi) and real space
coordinates (Xi, Yi), substitute them into (1), establish a set of
equations of unknown variable hj(j=1,,8h9=1) , solve
the unknown parameters from the equation[6]. The style of
our target is Checkerboard pattern format as illustrated in
figure 7. We actually select 12 points to perform the
calculation and the result is shown as follows:
4.2498
H = 0.1110
0 . 0001

C. Edge segmenting of firebrick image

As the two adjacent edges of the brick are connected to
one edge line as shown in figure 5, it is unpractical to do line
fitting directly on this set of edge line. The correct way is to
split them first. The splitting point is the one with the largest
curvature. So an edge segmenting step is needed after edge
linking step. Each set of edge lines is checked and the
curvature of each point on a line is calculated. When the
curvature of a point is unusually large, the line is broken into
two lines at this point. The segmented lines are illustrated in
figure 6 by thick lines of different gray. As can be seen, the
segmented lines are very suitable for fitting the edge lines.

0.1206
4.3137
0

79.8009
62.2959
1

B. Experiment and Results of Measuring Length and

Width of Firebrick
Once accurate values of elements in matrix H are
determined, the length and width measurement of firebricks
can be carried out. The coordinates (X, Y) in object plane can
be gotten by transforming (1) as follows:
( h 2 h8 u )( v h 6 ) ( h5 h8 v )( u h3 )

X = ( h h u )( h h v ) ( h h v )( h h u )
2
8
4
7
5
8
1
7

( h1 h7 u )( v h6 ) ( h 4 h7 v )(u h3 )
Y =
( h1 h7 u )( h5 h8 v ) ( h 4 h7 v )( h 2 h8 u )

D. Edge fitting
It is an easy task for edge fitting on the base of good
edge segmentation. Line fitting is done to each edge line,
then the parameters such as the coordinates of intersection of
edges, the distance between the parallel edges, the length of
edges and etc can be calculated from the equations of
straight fitting lines. The unit of these parameters is pixel.

(2)

According to all the edge point coordinates and corner

point coordinates of bricks gotten from the previous image
processing, we can use formula (2) to calculate their
corresponding points in the O-XY coordinate system, then
the information such as the actual length, width, parallelism,
perpendicularity and more of the firebricks can be
calculated.
We carried out the calculation on the photos of the same
bricks from different view angle, and pictures are shown in
figure 8. The measurement results obtained are shown in
table I. Two values for each length and width represent the
corresponding values of the two parallel edges. The standard
size of the brick should be 300mm150mm, however the

Figure 6. Edge lines after segmentation

396

actual length and width measured by a slide caliper change

slightly with different measuring point, which is in the range
(3000.5)mm(1500.5)mm.

accuracy and stable performance. It is a good start for

machine vision technology to be used for firebricks quality
inspection. We will continue further development on this
application.
ACKNOWLEDGEMENTS
The ongoing research described herein is supported by
Shanghai
Maritime
University
Research
Fund
(20090169) Shanghai Emphasis Subject Project
(J50602) Shanghai Maritime University Graduate Fund
Shanghai Engineering Research Center Construction
Project(09DZ2250400).

(a)

REFERENCES
[1]

[2]

(b)

[3]

[4]

[5]

(c)

[6]

(d)
Figure 8. Pictures of a firebrick from different view

TABLE I.

LENGTH AND WIDTH OF A FIREBRICK MEASURED BY THE

MACHINE VISION SYSTEM

No.

Length(mm)

Width(mm)

300.5931

300.0476

150.0905

150.3227

301.0740

300.0994

150.5486

150.1196

300.2688

298.9699

149.8153

150.5654

300.8835

299.7102

150.0559

150.4404

The data in table I shows that errors between the

measured size and the actual size are small, which meet with
quality inspection requirements. This experiment verifies the
algorithm is stable.
V.

CONCLUSIONS

A new measuring approach of firebricks size is

proposed in this paper. A monocular machine vision system
has been established and a four-step image processing
method is introduced. It is characterized with simple system
configuration, simple algorithm, as well as appropriate

397

Yang YU, Wang Rui, He Zhong, Liu Fei, Zhang YaPing, Methods
and system research of improving dynamical detective precision of
firebricks thickness, Refractories, 1997,31(6):348-350(in Chinese)
HU Hong-hao, LI Bing-lin, Cai Zhi-rui, Li Shi-hong, Research for
the On-Line Thickness Detection of Ceramic Tile Based on CCD,
Shandong Ceramics, 2006, 29(3):14-16(in Chinese)
La Verne Thornton, Gerald L. Stuart, Apparatus For Sensing And
Ejecting SBricks Of Improper Size, United States: Appl. No.
64,740, Aug. 18, 1970.
Wei Zhenzhong, Flexible On-line 3D Coordinates Measurement
System Based on Machine Vision, PHD thesis of Beijing University
of Aeronautics and Astronautics, 2003. (in Chinese)
Zhang Guangjun, Machine Vision, Beijing, Science Press, 2005. (in
Chinese)
Criminisi, Antonio. Accurate Visual Metrology from Single and
Multiple Uncalibrated Images. University of Oxford, June 2001.