1536 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 47, NO.

4, JULY/AUGUST 2011

Shape Measurement of Steel Strips Using a

Laser-Based Three-Dimensional
Reconstruction Technique
Julio Molleda, Rubén Usamentiaga, Daniel F. García, Member, IEEE, Francisco G. Bulnes, and Laura Ema

Abstract—Quality control is of utmost importance in the metal

industry. It requires online measurement and inspection systems
which provide precise feedback to closed-loop controllers in in-
dustrial facilities. In rolled products, shape is one of the main
quality criteria. In this paper, a low-cost real-time 3-D shape
measurement system for long flat (FL)-rolled products based
on laser triangulation is proposed. The system provides online
measurements of two geometrical features of the shape of rolled
products: flatness and width. The proposed system is based on 3-D
surface reconstruction of rolled products with which flatness can
be measured accurately and continuously across the whole width
of the strip. Three-dimensional surface reconstruction provides
highly accurate width measurements not only of FL products, as
do most other systems, but also of non-FL products. The accuracy Fig. 1. Feedback block diagram of a shape control process.
of the laser extraction method used to reconstruct the surface of
the rolled products is evaluated, as well as the online performance
of the system. reconstruct the objects of interest. One major application of
Index Terms—Feature extraction, flatness measurement, image 3-D reconstruction is obtaining metric information of the ob-
processing, measurement by laser beam, metal industry, shape jects of a scene, usually involving 3-D shape measurement.
measurement, width measurement. Three-dimensional shape measurement has evolved rapidly
over the last decade [1] to meet real-time constraints in optical
I. I NTRODUCTION systems. Several optical techniques have been developed to
measure the 3-D shape of objects from one position [2], [3].
T HREE-DIMENSIONAL (3-D) reconstruction is one of
the most outstanding research fields in computer vision.
The wide range of applications of 3-D reconstruction covers au-
Based on these techniques, several machine vision methods
have been developed to inspect industrial processes and, thus,
tonomous navigation and guidance, quality inspection and con- to inspect or control the quality of the manufactured products.
trol, reverse engineering, object recognition, medical imaging, In the metal industry, specifically the production in long flat
and shape digitizing, among others. Three-dimensional recon- (FL)-rolled products, geometrical factors are used to estimate
struction aims to compute a 3-D model of the objects of a scene the quality achieved in the rolling process. An evaluation of
from one or more 2-D images of the scene. Certain applications, these factors allows the manufacturer to determine whether the
such as autonomous airborne navigation and automatic quality production or customer requirements have been fulfilled. This
control of long products, require a combination of the informa- evaluation task usually requires real-time shape measurement
tion gathered from multiple color or range images to accurately during manufacturing.
In this paper, a low-cost real-time 3-D shape measurement
Manuscript received January 3, 2011; accepted March 8, 2011. Date of
system for long FL-rolled products based on laser stripe scan-
publication May 19, 2011; date of current version July 20, 2011. Paper ning is proposed. Based on a 3-D reconstruction of the surface
2011-METC-003, presented at the 2010 Industry Applications Society Annual of the inspected strips, the system aims to compute the two main
Meeting, Houston, TX, October 3–7 and approved for publication in the IEEE
T RANSACTIONS ON I NDUSTRY A PPLICATIONS by the Metals Industry Com-
geometrical factors affecting the final quality of the product:
mittee of the IEEE Industry Applications Society. This work was supported in flatness and width. This system can also be applied to many
part by the Spanish Ministry of Education and Science under Project MICINN- other metallic and nonmetallic rolled products. Fig. 1 shows the
08-TN2008-06045-C001 and in part by the Government of the Principality of
Asturias under Project PCTI-PEST08-19.
block diagram of the shape control process in a strip production
J. Molleda, R. Usamentiaga, D. F. García, and F. G. Bulnes are with the plant, where the component called shape inspection is the
Department of Computer Science, University of Oviedo, 33204 Gijón, Spain system proposed in this paper.
(e-mail: [email protected]; [email protected]; [email protected];
[email protected]). Flatness can be defined mathematically as the degree to
L. Ema is with the R&D Technological Center–Industrial Enhancement, which the surface of an FL product approaches a plane [4].
ArcelorMittal, 33400 Avilés, Spain (e-mail: [email protected]). During the rolling process, a flatness control system drives the
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. mechanical and thermal actuators of the rolling mill to adjust
Digital Object Identifier 10.1109/TIA.2011.2155019 the shape of the strip. The first actuators adapt the roll gap,
0093-9994/$26.00 © 2011 IEEE
MOLLEDA et al.: SHAPE MEASUREMENT OF STEEL STRIPS USING A RECONSTRUCTION TECHNIQUE 1537

the bending, and the tilting of each rolling mill stand [5]. The structured light [9], which simplify the mathematical solution
second actuators compensate for uneven temperature across the of the correspondence problem [10].
strip during rolling [6]. The flatness control system is usually a Active machine vision methods for noncontact shape mea-
closed-loop controller which requires precise feedback. Thus, surement in industry are usually based on the triangulation
flatness measurements must be carried out while the strip is principle [11]. This principle, together with structured light
being rolled. methods, makes the 3-D reconstruction of the surface of rolled
Width is also a vital factor in the quality of an FL-rolled products feasible. The term structured light refers to a method
product. During the rolling process, a width control system which casts simple or encoded light patterns onto the illumi-
drives mechanical actuators to adapt the roll gap of the edging nated scene [12], which improves the definition of the features
mill. Like flatness control systems, width control systems are of the surfaces of objects.
closed-loop controllers which require precise feedback. There- Structured light systems are based on the stereovision tech-
fore, width measurements must also be carried out while the nique, in which one camera is replaced by an active light
strip is being rolled. projector. The projected light pattern can be classified into three
Optical 3-D shape measurement techniques are covered in groups [10], [12]: spot patterns; stripe patterns, which may be
Section II, analyzing their main advantages with regard to static or encoded; and color patterns. Spot- and stripe-based
other techniques, as well as the light patterns that can be systems are slower but more robust than those based on encoded
used. The laser stripe scanner proposed in this paper for shape stripes, which are more prone to error. In real-time applications,
measurement of steel strips is covered in Section III. Both the laser light is the preferred choice for casting patterns onto the
geometry of the proposed system and the laser stripe extraction scene [13].
method are described, and the accuracy achieved is analyzed. Laser stripe scanning, which outperforms the spot-based
The computation of flatness measurement from 3-D surface technique, uses a simple geometric pattern, i.e., a single stripe,
reconstruction is described in Section IV, while Section V projected onto the surface of the object to be measured. A
reports the width measurement approach proposed in this paper. larger set of 3-D measurements can be gathered if several light
The online performance of the proposed system is analyzed in stripes are projected onto the surface of the object; however,
Section VI, and the conclusions are presented in Section VII. the corresponding problems associated with these systems are
more complex. Therefore, the system proposed in this paper for
real-time 3-D shape measurement is based on the projection of
II. T HREE -D IMENSIONAL S HAPE M EASUREMENT
a single stripe.
The ever-changing demands for improved quality of rolled
products make the use of high-speed 3-D measurement systems
III. L ASER S TRIPE S CANNER FOR 3-D S HAPE
essential. These systems are usually required to operate in real
M EASUREMENT OF ROLLED P RODUCTS
time, both to send the measurements as feedback information
to closed-loop controllers and to display the measurements The system presented in this paper is a low-cost real-time
to the human operators of the mill. The deadline for these 3-D shape measurement system of steel strips based on a laser
tasks is imposed by the speed of the product movement during stripe scanning technique, also known as slit scanning. This sys-
manufacturing. tem uses 3-D surface reconstruction of the strips to be inspected
Machine vision systems based on optical 3-D shape measure- to provide online measurements of the two main geometrical
ment outperform mechanical systems, which are mainly based factors of the strips: flatness and width. The proposed system
on contact principles and can cause scratching of the surface of can also be used with many other metallic and nonmetallic
the products during inspection. rolled products.
Traditionally, 3-D machine vision methods for nondestruc-
tive testing are classified into two groups: passive, which only
A. Geometry of the System
requires environmental light to illuminate the scene, and active,
in which light patterns are projected onto the surface of the A laser emitter casts a single beam onto the scene, and its
object to be measured. These methods can also be classified intersection with the surface of the strip produces a visible
based on the spatial or temporal domain in which they locate stripe. The wavelength of the beam is 670 ± 10 nm. Laser light
corresponding features [7]. is used because it is more power efficient and requires less
The most commonly used passive machine vision method maintenance than other kinds of light sources. Also, the high
is shape-from-stereo or stereovision method [8]. This method spectral radiance of laser light reduces the exposure time per
uses two cameras to simultaneously acquire a pair of images measurement. The laser stripe projector is located at the center
of the scene. Three-dimensional information about the scene of the roll path of the production line, as can be seen in Fig. 2.
can be obtained by finding corresponding pairs from the two The visible stripe in the scene is captured by a 1280 × 1024
images. Other passive methods include shape-from-shading, pixel CMOS matrix camera, also centered over the roll path of
shape-from-motion, and shape-from-texture methods. The ef- the production line. The triangulation angle is determined by
fectiveness of these methods is reduced in uncontrolled lighting the dimensions of the region of interest (ROI) of the scene, the
conditions, such as those in industrial facilities. dimensions of the camera sensor, and the vertical fluctuation of
The most widely used systems for shape measurement in the surface of the strip during manufacturing. The most widely
the metal industry are based on active methods, particularly used triangulation angle in slit scanners is 30◦ . In general, as the
1538 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 47, NO. 4, JULY/AUGUST 2011

in the scene in industrial facilities affects general-purpose

methods.
The 3-D shape measurement system proposed in this paper
extracts the laser stripe from each image acquired by the
camera based on the method described in [19]. The laser stripe
extraction method is accomplished in two stages. First, the
laser stripe projected onto the roll path of the production line
is extracted by means of a boundary linking process based on
prior knowledge of the shape of the roll path. Second, the laser
stripe projected onto the surface of the strip is extracted by
means of a global search, where the searching area is delimited,
taking into account the information obtained in the first stage.
In the second stage of the laser stripe extraction method, the
shape of the laser stripe projected onto the surface of the strip
Fig. 2. Geometry of the laser stripe scanner. is fitted using an improved split-and-merge approach based on
the iterative end-point fit procedure [20]. The fitting procedure
adjusts the laser stripe by means of a set of continuous seg-
ments. Several experiments with different lighting conditions in
the production line were carried out to determine the best type
of segments to accurately fit the laser stripe onto the surface of
the steel strips. Linear, quadratic, and Akima splines [21] were
evaluated. The extraction of the laser stripe onto the surface
of steel strips using Akima splines proved to be smoother than
using linear and quadratic segments. Thus, these segments are
the most suitable to fit the laser stripe projected onto the strips.
Furthermore, noise detection and removal are carried out easily
Fig. 3. Image acquired by the camera of the proposed system, including the using Akima splines.
ROI of the scene where the laser stripe is projected on both (center) the surface The laser stripe extraction method provides a set of segments
of the strip and (sides) the surface of the roll path. which describes the shape of the laser stripe projected onto the
triangulation angle increases, the sensor range decreases and surface of the steel strip. The results of the laser stripe extraction
resolution increases. After several simulations and experiments, method applied over three different images are shown in Fig. 4.
the optimal triangulation angle for the proposed system has The first image [first row of Fig. 4(a)] shows a nearly FL steel
been found to be 45◦ , which achieves the highest resolu- strip. The figure shows how the laser stripe projected onto the
tion, eliminating the likelihood of missing data due to sensor surface of the strip is fitted by means of only one segment
occlusions. [second row of Fig. 4(a)]. The second image [Fig. 4(b)] shows
The camera of the system uses an optical filter to increase a strip with a concave down (CD) profile. It can be seen that the
the signal-to-noise ratio of the scene. The passband of the filter extraction method requires two segments to accurately fit the
takes into account the wavelength of the light emitted by the stripe. Finally, the third image [Fig. 4(c)] shows a stripe with a
laser projector, as well as the angle of incidence of the laser concave up (CU) profile. The figure shows the three segments
stripe onto the camera sensor. In the proposed system, a dark red computed by the fitting procedure.
filter is used, where its narrow bandpass is centered at 660 nm. The segments describing the stripe onto the surface of the
The ROI of the scene contains the laser stripe projected on strip provided by the extraction method are expressed in the 2-D
both the surface of the strip and the surface of the roll path coordinate system of the image. Once the system has been cali-
of the production line. The ROI of an image acquired by the brated, the stripe can be expressed in the 3-D coordinate system
camera of the system is shown in Fig. 3. of the scene, making it possible to take metric measurements of
the surface of the strip.
B. Laser Stripe Extraction
C. Accuracy of 3-D Shape Reconstruction
The laser stripe projected onto the scene, together with the
speed of the strip during manufacturing, is used to reconstruct Camera resolution and camera calibration are factors which
the surface of the strip. In order to express the laser stripe on the influence the accuracy of the 3-D shape reconstruction of an
surface of the strip in the 3-D coordinate system of the scene, object. The accuracy of the calibration procedure is evaluated
the camera of the system must be calibrated. Several calibration based on the discrepancy between the real position of the
techniques have been proposed [14]–[17]. In this paper, the reference points of the calibration pattern and the 3-D position
nonlinear method proposed by Heikkilä is favored because of estimated from its 2-D projection in the image. For each 2-D
its balance between accuracy and calibration effort [18]. reference point extracted from the image, the calibrated model
Laser stripe extraction is a straightforward task under con- is used to reproject the point over the laser plane, measuring the
trolled lighting conditions. However, the noise which appears distance between the 3-D reference point and its reprojection.
MOLLEDA et al.: SHAPE MEASUREMENT OF STEEL STRIPS USING A RECONSTRUCTION TECHNIQUE 1539

Fig. 4. Laser stripe extraction. (a) FL. (b) CD. (c) CU.

TABLE I
ACCURACY E VALUATION OF 3-D S HAPE R ECONSTRUCTION

Fig. 5. Pattern used in calibration procedure located at the center of the ROI
of the scene.

Table I shows the results of the accuracy evaluation for both

the calibration procedure and the laser stripe extraction method.
The accuracy of the calibration procedure was achieved us-
ing a calibrated pattern located at three different places of
the ROI, involving over 400 measurements between reference
points. The accuracy achieved in the calibration procedure is
considered excellent since the mean error is less than double
Fig. 6. Calibrated workpiece used in the evaluation of 3-D reconstruction
accuracy. the uncertainty with which the distance between the reference
points is known.
The accuracy evaluation of the laser stripe extraction was
Fig. 5 shows an example of the set of images used in the carried out using three different calibrated workpieces located
calibration procedure. In this image, the reference points used in 21 different places of the ROI, involving over 3000 measure-
for the calibration procedure are the center of mass of each ments of the calibrated workpieces. The accuracy achieved in
rhombus of the pattern. The manufacturing procedure of this the laser strip extraction method multiplies the uncertainty by
pattern guarantees that the distances between adjacent reference up to nine times those of the known physical dimensions of the
points have an error of less than 0.1 mm, which gives an aluminum workpieces. Although this level of accuracy if far
uncertainty in the distance between reference points in the from that achieved in the calibration procedure, it is important
pattern of 0.058 mm. to note that this accuracy evaluation has been carried out taking
The accuracy of the 3-D shape reconstruction of an object into account the worst lighting conditions of the production
is also influenced by the laser stripe extraction method. The line. Therefore, this level of accuracy can be considered very
accuracy of the laser stripe extraction is evaluated by assessing high.
the accuracy of the 3-D reconstruction of objects whose physi-
cal dimensions are known. For this purpose, several aluminum
IV. F LATNESS M EASUREMENT
workpieces calibrated with an error of less than 0.2 mm were
used. This error gives an uncertainty in the dimensions of the The final quality of a rolled product is characterized to a
calibrated workpiece of 0.115 mm. The 3-D shape of each of large degree by its flatness. Rolled products with poor or bad
these workpieces has been reconstructed from several locations flatness are more likely to break, as well as to damage their
of the ROI of the scene using the laser stripe extraction method own surfaces, during subsequent manufacturing processes. To
described in this paper. Fig. 6 shows an example of the set of achieve good flatness, a closed-loop controller which requires
images acquired for reconstructing the shape of a calibrated precise online flatness measurements as feedback is required.
workpiece located at the center of the ROI of the scene with Some mechanical flatness measurement systems are based
adverse lighting conditions. on contact principles, such as those using strength or pressure
1540 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 47, NO. 4, JULY/AUGUST 2011

Fig. 7. Common flatness defects of long FL-rolled products. (a) Bad leveling. (b) Wavy edges. (c) Center buckle.

transducers integrated into a deflector roll to compute flatness

based on radial force measurements [22]. Other mechanical
systems are based on noncontact principles, such as those
involving periodic excitation of the product to be measured to
compute flatness based on the movement amplitudes across the
product width [23].
More recently, noncontact techniques have been used to mea-
sure flatness of rolled products based on stereoscopic principles
using X-ray gauges [24], capacitance principles using capaci- Fig. 8. Approximation of the length of a fiber.
tive sensor arrays [25], and optical principles using structured
light [26], [27]. In this paper, we compute the flatness of steel surface. This computation is based on an integration technique,
strips using the information about their 3-D shape computed approximating the length of the fiber, as shown in Fig. 8, using
previously. (2), where Lj is the approximate length of fiber j; hji is the
The flatness of a steel strip is expressed as deviation indices height of fiber j in the laser plane measured in frame i; ti is the
between the surface of the strip and an FL horizontal surface. time stamp when frame i is acquired; vi is the average of
In order to measure flatness, the surface of a steel strip is con- the speed movements of the strip in the production line between
sidered as multiple adjacent longitudinal fibers. If the lengths frames i and i − 1; and n is the total number of frames in the
of these fibers are the same, the strip is perfectly FL. Flatness fixed longitudinal section of the strip
defects are generated when the working rolls of each rolling  n  2
mill stand do not apply the same pressure over the whole width Lj = hji − hji−1 + vi2 (ti − ti−1 )2 . (2)
of the strip, elongating some fibers more than others. i=1
Common flatness defects include bad leveling, wavy edges,
The flatness provided by the shape measurement system
and center buckle. Bad leveling defects are generated when the
proposed in this paper cannot be validated using real steel strips
length of the fibers of the strip increases progressively from one
during manufacturing because their flatness measurements are
longitudinal edge to the other, as shown in Fig. 7(a). Wavy
not accurately known a priori. Therefore, in order to evaluate
edges are generated when the fibers closer to the edges of
flatness measurement, a strip simulator which replicates the ver-
the strip are longer than the fibers at the center, as shown in
tical fluctuations of the surface of a strip with flatness defects is
Fig. 7(b). Center buckle defects are generated when the fibers
used. The proposed 3-D shape measurement system extracts the
at the center are longer than the fibers at the edges of the strip,
laser stripe projected onto the strip simulator from each frame
as shown in Fig. 7(c).
acquired by the camera in the same way as it was projected onto
The flatness of long FL-rolled products is usually expressed
the surface of a strip. Thereupon, the system reconstructs the
quantitatively in a metric called I-units, calculated using (1),
3-D surface of the strip simulator and computes a flatness
where Lref is the length of a reference fiber and ΔLj /ΔLref
profile for each fixed longitudinal section of the emulated
is the elongation of fiber j with regard to the reference fiber
strip.
ref . Since the relation between the lengths of the fibers is
Table II summarizes the results of 40 experiments replicated
not the same over the entire length of a steel strip, the strip
30 times over an emulated strip considered to be composed of
is divided into fixed longitudinal sections. For each section, a
200 fibers. The experiments are classified into four ranges: low,
flatness index for each fiber is calculated, generating a flatness
medium, high, and very high flatness defects. The mean error
profile. The sections may be partially overlapped, generating
and the error dispersion with 90% confidence for each flatness
a higher density of profiles for a single strip. All flatness
defect range are shown. The measurement error is within the
profiles computed during the manufacturing of a strip can be
uncertainty interval of the magnitude of the flatness of the
represented in a 2-D image, called a flatness map
emulated strip.
ΔLj 5 Lj − Lref 5
I-unit = 10 = 10 . (1)
Lref Lref V. W IDTH M EASUREMENT
In each longitudinal section of the strip, the system computes In the metal industry, there is a strong demand for accurate
the length of each fiber L based on the 3-D reconstruction of its measurements of the dimensional properties of steel strips
MOLLEDA et al.: SHAPE MEASUREMENT OF STEEL STRIPS USING A RECONSTRUCTION TECHNIQUE 1541

TABLE II If the integrand in (3) is called g(x) and the indefinite integral
R ESULTS OF THE F LATNESS M EASUREMENTS IN
S EVERAL D EFECT R ANGES of g(x) is called G(x), the value of L equals G(b) minus G(a).
That is to say the indefinite integral of g(x) evaluated at b minus
the indefinite integral of g(x) evaluated at a

g(x) = 1 + (f  (x))2

G(x) = g(x) dx + C

L = G(b) − G(a). (4)

The laser stripe extraction method considers three types of

functions: linear, quadratic, and Akima splines. In the trivial
case, f (x) is a linear function, and L can be calculated as

L = 1 + a21 (b − a). (5)
Fig. 9. Length approximation using multiple linear segments.
In the case of segments described by quadratic functions, the
during manufacturing. These properties include thickness pro- expressions for g(x) and G(x) are shown in
file, width, and length. Width is considered the most important 
property because any deviation in width requires a large toler- g(x) = 1 + (a1 + 2a2 x)2
ance margin to be trimmed along the full length of the coil [28].
Many commercial products have been released in recent sinh−1 (a1 + 2xa2 )
G(x) =
years for the accurate measurement of the width of steel strips, 4a2
such as [22]. When the strip is perfectly FL, only an accurate 
detection of the position of the edges of the strip is needed to (a1 + 2xa2 ) 1 + (a1 + 2xa2 )2
+ . (6)
calculate the width. The difference between the two positions 4a2
provides the desired information. However, in the case of strips
A spline is a function which is built from piecewise third-
with flatness defects, this method is imprecise because waves
order polynomials. Akima splines are defined for each of these
in the shape of the strip are not taken into account. This is
pieces as follows, where x0 is the first point of the interval:
particularly noticeable in very thin strips, where flatness tends
to be worse. f (x) = a0 + (x − x0 ) (a1 + (x − x0 ) (a2 + a3 (x − x0 ))) .
In this paper, we proposed an accurate method to measure (7)
the width of steel strips using previously computed information
about their 3-D shape. When the shape of the strip across the In this case, g(x) yields as
width is known, width measurement can take flatness defects 
into account, providing a more accurate measurement for non- g(x) = 1+(a1 +2a2 (x−x0 )+3a3 (x − x0 )2 )2 . (8)
FL products.
The result of the laser stripe extraction method carried out However, a serious problem arises when trying to calculate
during the 3-D surface reconstruction of the steel strip is a set the indefinite integral G(x). The operation is so complex that it
of segments which describes the shape of the strip across its cannot be calculated analytically. More complex functions have
width. Given these segments, the width of the steel strip can the same problem. In these cases, numerical integration (NI) is
be calculated by adding the lengths of each of these segments. the only option.
Therefore, the width of the steel strip can be calculated by The Gauss–Kronrod quadrature, which is an efficient variant
computing the length of each individual segment, described by of the Gaussian quadrature, is used to calculate the numerical
a mathematical function. integral. The procedure adapts the integration routine to con-
The length of a segment of curve L can be approximated by centrate new subintervals around integrable singularities. As the
connecting a finite number of points on the curve using linear subintervals decrease in size, the successive approximations to
segments to create a polygonal path, as shown in Fig. 9. the integral converge.
The total length of the approximation can be found by The calculation of length using NI is much slower than
adding the lengths of each linear segment calculated by the applying a single equation, as with linear or quadratic functions.
Pythagorean theorem. As the length Δx of these segments However, modern hardware is able to compute this type of NI
decreases, the approximation improves. Thus, the exact length using an optimized implementation in less than a tenth of a
can be calculated using millisecond with an error lower than 1−15 .
In order to test width measurement, four different methods
b  have been compared. Two of them consider the steel strip
L= 1 + (f  (x))2 dx. (3) to be FL, as do most commercial systems. Under this as-
a sumption, two alternatives are available to approximate the
1542 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 47, NO. 4, JULY/AUGUST 2011

Fig. 10. Laser stripes expressed in spatial coordinates for width measurement. (a) FL. (b) CD. (c) CU.

TABLE III TABLE IV

C OMPARATIVE R ESULTS OF W IDTH M EASUREMENT M ETHODS C OMPUTING T IME OF THE M AIN TASKS OF THE S HAPE
M EASUREMENT S YSTEM

VI. O NLINE S HAPE M EASUREMENT A NALYSIS

width of the strip: approximation by a single horizontal line
and approximation by a single line (SL) which connects the Online shape measurement requires the system to compute
points where the two edges are located. In the SL method, the 3-D shape of the surface of the steel strips within the
the line might not be horizontal. The other two methods are deadline imposed by the image acquisition rate of the camera
approximation of the width using multiple lines (MLs) and NI of the system. Table IV shows the time consumed in the
using the procedure described previously. With the ML method, laser stripe extraction task over several steel strips during
lines are created as the hypotenuses of right triangles lines manufacturing with different lighting conditions of the scene.
where Δx equals 3 mm. The system uses a PC equipped with an Intel Pentium D 930
The methods to measure the width of strips have been applied microprocessor (3.0 GHz) with 4 MiB of cache memory and
to three different images, shown in the first row of Fig. 4. The 2 GiB of RAM memory. As the table shows, using this hard-
laser stripes extracted from these images using Akima spline ware, the acquisition rate of the camera can be as high as 400
functions are shown in Fig. 10. The extracted laser stripes frames per second.
are shown in spatial coordinates (in millimeters), on both the As well as computing the 3-D shape of the surface of the
horizontal and vertical axes. strip, the system must also compute the flatness profile of a
The results of the experiments are shown in Table III. As fixed longitudinal section of the strip and the width of the strip
can be seen, in the FL image, where the strip is not perfectly in each frame. The deadline for carrying out these computations
FL, differences between the four measurement methods are is imposed by the image acquisition rate of the camera of the
negligible. However, when the shape of the strip across the system. Table IV shows the time consumed by both the flatness
width is a curve, such as in CD or CU images, the differences profile and the width computation task over several steel strips
increase. In the case of the CD image, the differences are of during manufacturing with different lighting conditions of the
only 1 mm. However, in the CU image, differences reach nearly scene. Taking into account the maximum time in the laser stripe
5 mm. These differences are vital for many types of applica- extraction task, the acquisition rate of the camera can be as high
tions. Therefore, width measuring methods which consider the as 240 frames per second.
strip to be FL will produce serious errors, depending on the
curvature of the strip across the width.
VII. C ONCLUSION
Differences between ML and NI are very low. The highest
difference occurs in the CU image, but it is only 0.003% High-speed 3-D shape measurement is required by the steel-
of the width of the strip. This indicates that the lines used making industry to meet challenging demands for improve-
approximated the shape of the strips across the width very ments in the quality of rolled products. The proposed shape
accurately. Also, decreasing the value of Δx further reduces measurement system takes advantage of contactless 3-D surface
this difference. reconstruction of rolled products to provide online flatness and
It is important to note that ML and NI methods can only be width measurements. The system is based on structured light
applied accurately because the laser stripe extraction procedure using a laser stripe cast onto the surface of the product to be
produces information about the segments which fit the laser inspected, which has been found to be sufficient to accurately
stripe, i.e., they use the 3-D surface reconstruction. Directly compute the 3-D surface of rolled products in adverse en-
applying these methods to the result of the center of mass would vironmental conditions. Flatness measurements are computed
produce major errors due to quantization. Also, if the segments comparing the elongations of several longitudinal sections of
used to fit the laser stripe are linear or quadratic, NI is not the product. Width measurements are computed using a novel
necessary since analytic calculations can be applied. approach which, unlike most width measurement systems, does
MOLLEDA et al.: SHAPE MEASUREMENT OF STEEL STRIPS USING A RECONSTRUCTION TECHNIQUE 1543

no require perfect flatness of the product to be measured. The [23] G. Spreitzhofer, A. Duemmler, M. Riess, and M. Tomasic, “Si-flat con-
proposed system has been installed in a conditioning line of tactless flatness measurement for cold rolling mills and processing lines,”
Revue de Metallurgie, vol. 102, no. 9, pp. 589–595, 2005.
ArcelorMittal steel company in Asturias, Spain. [24] R. R. Jackson, P. E. Kelly, and D. W. Adams, “A multi-function measure
system for hot-strip mills,” Millenium Steel, vol. 4, pp. 1–4, 2003.
[25] J. M. Lopera, P. J. Villegas, F. F. Linera, F. Hernández-Magadan,
J. Martín-Ramos, J. Díaz, G. Vecino, and J. L. Rendueles, “Flatten it out:
ACKNOWLEDGMENT Designing a low-cost system for flatness monitoring in metal processes,”
IEEE Ind. Appl. Mag., vol. 14, no. 2, pp. 62–66, Mar./Apr. 2008.
The authors would like to thank the technicians and the [26] R. Pirlet, J. Mulder, D. Adriaensen, and J. Boelens, “Rometer, a non-
engineers of ArcelorMittal Asturias for their helpful assistance contact system for measuring hot strip flatness,” Online Inspect. Steel
during tests. Products, pp. 79–94, May 1983.
[27] J. Borchers and A. Gromov, “Topometric measurement of the flatness
of rolled products—The system topplan,” Metallurgist, vol. 52, no. 3/4,
pp. 247–252, 2008.
R EFERENCES [28] W. Hong, P. Kim, and J. Y. Y. H. Moon, “Hot strip width control method
[1] S. Zhang, “Recent progresses on real-time 3D shape measurement using by using looper tension measuring system in finishing mill,” J. Mater.
digital fringe projection techniques,” Opt. Lasers Eng., vol. 48, no. 2, Process. Technol., vol. 111, no. 1–3, pp. 74–78, 2001.
pp. 149–158, Feb. 2010.
[2] F. Chen, G. M. Brown, and M. Song, “Overview of three-dimensional
shape measurement using optical methods,” Opt. Eng., vol. 39, no. 1,
pp. 10–22, Jan. 2000.
[3] F. Blais, “Review of 20 years of range sensor development,” J. Electron.
Imag., vol. 13, no. 1, pp. 231–243, 2004.
[4] V. B. Ginzburg, Geometry of Flat Rolled Products. Pittsburg, PA: United
Eng. Inc., 1992. Julio Molleda received the M.S. degree in computer
[5] C. Nappez, S. Boulot, and R. C. McDermott, “Control of strip flatness in science from the Engineering School, University of
cold rolling: A global approach,” Iron Steel Eng., vol. 74, no. 2, pp. 42–45, Oviedo, Gijón, Spain, in 2005 and the Ph.D. degree
1997. from the University of Oviedo in 2008.
[6] R. Usamentiaga, D. F. García, D. González, and J. Molleda, “Compensa- He is currently an Associate Professor with the
tion for uneven temperature in flatness control systems for steel strips,” in Department of Computer Science and Engineering,
Conf. Rec. IEEE IAS Annu. Meeting, 2006, pp. 521–527. University of Oviedo. In recent years, he has been
[7] J. Davis, D. Nehab, R. Ramamoorthi, and S. Rusinkiewicz, “Spacetime working in real-time imaging research projects in
stereo: A unifying framework for depth from triangulation,” IEEE Trans. the computer engineering area at the University
Pattern Anal. Mach. Intell., vol. 27, no. 2, pp. 296–302, Feb. 2005. of Oviedo. His research interests include real-time
[8] S. T. Barnard and M. A. Fischler, “Computational stereo,” Comput. Surv., imaging systems and range measurement techniques.
vol. 14, no. 4, pp. 553–572, Dec. 1982.
[9] M. Ribo and M. Brandner, “State of the art on vision-based structured
light systems for 3D measurements,” in Proc. IEEE Int. Workshop Robot.
Sens., Robot. Sens. Environ., 2005, pp. 2–6, art. No. 1588327.
[10] J. Salvi, J. Pagès, and J. Batlle, “Pattern codification strategies in struc-
tured light systems,” Pattern Recognit., vol. 37, no. 4, pp. 827–849,
Apr. 2004.
[11] R. I. Hartley and P. Sturm, “Triangulation,” Comput. Vis. Image Underst., Rubén Usamentiaga received the M.S. and Ph.D.
vol. 68, no. 2, pp. 146–157, 1997. degrees in computer science from the Univer-
[12] R. Klette, K. Schluns, and A. Koschan, Computer Vision: Three- sity of Oviedo, Gijón, Spain, in 1999 and 2005,
Dimensional Data From Images. Berlin, Germany: Springer-Verlag, respectively.
1998. He is currently an Associate Professor with the
[13] R. Noll and M. Krauhausen, “Online laser measurement technology for Department of Computer Science and Engineering,
rolled products,” Ironmaking Steelmaking, vol. 35, no. 3, pp. 221–227, University of Oviedo. In recent years, he has been
Apr. 2008. working on several projects related to computer vi-
[14] E. L. Hall, J. B. K. Tio, C. A. McPherson, and F. A. Sadjadi, “Measuring sion and industrial systems. His research interests in-
curved surfaces for robot vision,” Computer, vol. 15, no. 12, pp. 42–54, clude real-time imaging systems and thermographic
Dec. 1982. observers for industrial processes.
[15] R. Y. Tsai, “A versatile camera calibration technique for high accuracy
3D machine vision metrology using off-the-shelf TV cameras and lenses,”
IEEE J. Robot. Autom., vol. RA-3, no. 4, pp. 323–344, Aug. 1987.
[16] Z. Zhang, “A flexible new technique for camera calibration,” IEEE
Trans. Pattern Anal. Mach. Intell., vol. 22, no. 11, pp. 1330–1334,
Nov. 2000.
[17] J. Heikkilä and O. Silven, “A four-step camera calibration procedure with
implicit image correction,” in Proc. IEEE Comput. Soc. Conf. Comput. Daniel F. García received the Ph.D. degree in elec-
Vis. Pattern Recognit., 1997, pp. 1106–1112. trical engineering from the University of Oviedo,
[18] W. Sun and J. R. Cooperstock, “An empirical evaluation of factors influ- Gijón, Spain, in 1988.
encing camera calibration accuracy using three publicly available tech- He is currently a Full Professor with the Depart-
niques,” Mach. Vis. Appl., vol. 17, no. 1, pp. 51–67, 2006. ment of Computer Science and Engineering, Univer-
[19] J. Molleda, R. Usamentiaga, D. F. García, and F. G. Bulnes, “Real-time sity of Oviedo. Since 1994, he has been responsible
flatness inspection of rolled products based on optical laser triangulation for the computer engineering area at the University
and three-dimensional surface reconstruction,” J. Electron. Imag., vol. 19, of Oviedo. His current research interest is in the area
no. 3, pp. 031206-1–031206-14, Jul. 2010. of the development of high-performance real-time
[20] R. O. Duda and P. E. Hart, Pattern Classification and Scene Analysis. and embedded systems applied to quality assurance
Hoboken, NJ: Wiley, 1973. and production inspection in industry, where he has
[21] H. Akima, “A method of bivariate interpolation and smooth surface fitting more than 100 published papers. For the last ten years, he has been conducting
for irregularly distributed data points,” ACM Trans. Math. Softw., vol. 4, research projects in the area of information technologies applied to industry at
no. 2, pp. 148–159, Jun. 1978. national and European levels.
[22] ABB Autom. Technol. AB, Stressometer Flatness Measurement Systems, Dr. García is a member of the Association for Computing Machinery and the
Zurich, Switzerland. IEEE Computer Society.
1544 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 47, NO. 4, JULY/AUGUST 2011

Francisco G. Bulnes received the M.S. degree in Laura Ema received the M.S. degree in industrial
computer science from the Engineering School, Uni- engineering from the Engineering School, University
versity of Oviedo, Gijón, Spain, in 2007, where he of Oviedo, Gijón, Spain, in 2004 and the Certificate
is currently working toward the Ph.D. degree in of Advanced Studies from the University of Oviedo
computer science. in 2009.
He is currently an Associate Professor with the She is currently a Research Engineer with
Department of Computer Science and Engineering, the R&D Technological Center–Industrial Enhance-
University of Oviedo. His current research activity ment, ArcelorMittal, Avilés, Spain. From her profes-
is focused on real-time industrial applications and sional career, her 14-month stay at CERN stands out,
imaging systems. working for the Liquid Argon Calorimeter Electron-
ics Department. During the last five years, she has
been working at ArcelorMittal in different research projects for developing real-
time measurement systems: flatness, temperature, width, etc.