High Speed Vision based automatic Inspection and Path Planning for Processing

conveyed Objects
M. Weyrich1, Y. Wang1, J. Winkel1, M. Laurowski1
1
University Siegen, Chair of Automated Manufacturing and Assembly
Paul-Bonatz-Str. 9-11, 57068 Siegen, Germany

Abstract
Under the pressure of cost reduction and productivity improvement, this paper presents a new methodology
which provides a fast inspection of defective objects and generates a real time motion trajectory for
processing objects being conveyed with high speed in an industrial large-scale production. The image data
obtained by a multispectral imaging system is analyzed within image processing algorithms using
classification methods based on support vector machine. These data provide a basis for a path planning
algorithm which considers location, orientation and arrangement of defects on the conveyed objects. Selective
processing tool guided by the planed path is motion controlled.
Keywords:
Design method, Visual inspection, Adaptive control

1 INTRODUCTION
Machine vision has been researched and applied in many
branches in science and engineering. Their applications
allow automating many technological processes, e.g.
industrial quality control.
In the industrial quality control, instead of manual
inspection, conventional vision based automatic
inspection has been applied. The approach aims to sort
objects or products according to vision features, which
are predefined, into variant categories indicating different
quality levels. Quality grading and automatically sorting of
oranges is a typical example [1]. However, the increasing
interests are not only aimed at simply sorting the
defective objects from flawless objects, but also
processing the defective objects directly after defect
detection. This will help to reduce waste during production
under the pressure of cost reduction and productivity
improvement. Therefore, vision based path planning
which can be applied for guiding the manipulator to
process defective objects, is becoming an important issue
in vision based automatic inspection system.
2

OBJECTIVES, PROBLEM FORMULATION AND

REQUIREMENTS

2.1 Objectives
This paper presents an approach for an automated
system consisting of a vision unit, path planning and
motion control as well as object processing. The system
provides a fast inspection of objects being transported on
a conveyor belt with high speed. An object can be a 2D or
3D object. According to the inspection result, the target
objects are classified in three categories: (1) flawless, (2)
defective and processable and (3) defective and nonprocessable (see Figure 1). The defective and
processable target objects are further classified according
to user-defined defect types. Meanwhile, a real time
motion trajectory is generated for processing the defective
and processable objects. Figure 1 display the schematic
representation of work flow concept of our proposed
processing system.
Comparing to conventional vision based inspection
systems, our approach allows detecting and classifying

defects on objects and apply a suitable processing

strategy according to detection results; it provides a
competitive alternative for handling assembly products
and easy-processing product, especially agricultural
goods (cucumber, apple, etc). The high speed processing
of agricultural good has special requirements on process
tolerances and machine characteristics which are varying
a lot from those systems for the discrete manufacturing
e.g. of metal.
2.2 Problem formulation
High speed vision based inspection and path planning are
the two main issues in this paper. They demand a real
time image acquisition (for 2D or 3D objects) and
analysis, path planning, motion control and a suitable
machine processing within a sequential working system.
Especially, in context of the processing of agricultural
products several constrains for instance irregular object
shape, varying color, high productivity with very fast
treatment as well as hygienic standards have to be taken
into account. Furthermore, due to the variant defect types
of conveyed objects, individual processing on conveyed
objects (which are determined to be defective and
processable) are often required, which may enhance the
flexibility of large-scale production system.
2.3 Requirements for an application
The requirements for this approach depend on the
desired inspection and processing task. Only reduced
special accuracy is economically viable because of the
following machining implemented in the industrial
process. Therefore, a depth resolution in the range of +/1.5 mm is sufficient for the detection of the surface
geometry. The size of the testing objects is in the range of
a few centimeters in diameter. A prismatic measurement
volume of width x length x height: 150mm x 150mm x 100
mm is defined. These dimensions are associated with the
size of the inspected objects and constrains of the
mechanical integration space. The shape of the objects
does not need to be completely captured, as long a priori
knowledge on the typical object structure is available. The
geometry of the objects varies in a certain range within
the specification. Missing areas of geometry can be
reconstructed, based on a statistically model.

Up to five objects per second may be processed

(depends on the velocity of conveyor belt). This leads to a
cycle time of about 0.2 s. In addition, a robust mode in
rough industrial environments should be possible with this
system to be developed. Robust mode in this context
means that the method works under adverse conditions
delivering expected accuracy and high reliability. Varying
color gradation of the object may not affect the geometry
of acquisition. In addition, possible rocking and swinging
motions of the conveyor system have to be compensated.

content based threshold adaption for image segmentation

was developed to find the optimal threshold value.
Multispectral image processing is discussed in [17] [18].
Kker [19] developed a vision based system for industrial
robotic manipulator to classify objects on a moving
conveyor. For classification objects, a neural network is
designed and applied. Moment invariants are used as
feature vector set. Machine learning as a branch of
artificial intelligence is always employed in modern vision
system; Examples of current development in this field is
described in [20], [21] and [22].
4 APPROACH
In this section a system concept adapted to the scenario
described in Section 2.1 is developed. Then the concept
solution to each problem described in Section 2.2 with
respect to our system concept is proposed.
4.1 System concept
Figure 2 illustrates the concept of a vision based
automated inspection and path planning for the
processing system. Continuously fed objects are placed
on a conveyor belt and transported with a quasi-constant
velocity. The imaging system is located above the
conveyor and provides image acquisition positioned. The
obtained image data is analyzed by a computer system.
The analysis results provide a basis for a path planning
algorithm which considers the location, orientation and
arrangement of defects on the conveyed objects. Forward
the image acquisition system is a processing unit (e.g. a
robot system) which is equipped with selective processing
tool. This processing unit is motion controlled according to
the result path planning for removing defects or another
desired treatment.
Camera bridge
Objects
feeding

Figure 1: Schematic representation of work flow concept.

3 STATE OF THE ART
Firstly, a survey on the recent development and
application of vision based automatic inspection in
industry was implemented. In the field of quality
inspection of agricultural products, one can refer to [2] [3]
[4] [5]. In [2], an inspection system and processing for
herb flowers was developed. In [3], a vision-based
automatic inspection system was introduced for sorting
fruits into variant classes. Moreover, in [6], an industrial
vision based automatic inspection system was developed
for inspecting welded nuts on the support hinge. In [1], a
vision-based methodology was developed for processing
3D objects on a conveyor belt. In [7], a vision system for
fast moving objects was introduced.
The acquisition of geometric information of a target
conveyed objects has been discussed in [8] [9] [10] [11].
In addition, the technique camera-space manipulation
(CSM) is always used in vision based path planning [12]
[13]. An approach to methodology for selection of a casebased appropriate imaging system is proposed in [14].
Additionally, segmentation as the vital image process
technique
always
attracts
attention.
General
segmentation algorithm can be found in [15]. In [16],

Processing tool
Conveying
direction

Further
processing
Conveyor belt
Rotary encoder signal

Figure 2: Concept of the inspection and processing

system.
4.2 Methodology of system development
Electromagnetic band for detecting surface defects
In the field of optical surface inspection, improvement of
contrast between the defective and flawless area of an
object plays a major role for reliable and stable automated
detection processes [1]. In order to obtain the optimal
contrast, a systematic experiment with variant illumination
and corresponding multispectral sensors (which possess
different bands of the electromagnetic radiation) has to be
made. In Figure 3 the experiments with agricultural
products with general defects are displayed, wherein (a)
the experiments with a monochrome camera with optical
band-pass filters in order to examine the spectral
response of defects is shown; (b) an image of thermal
camera after a thermal excitation; (c) the experiment with
a computed tomography system; (d) the experiment with
magnetic resonance imaging system. For a specific
defects type, the contrast may reach maxima in a
particular electromagnetic band.

Image acquisition and illumination

Respect to the determined electromagnetic band, an
image acquisition and illumination should be developed
for surface defects detection and localization. Firstly, the
target objects are divided into two categories: 2D objects
and 3D objects. 2D objects are considered to be the
objects of slice form, whereas 3D objects tuber shape.
For the case of 2D objects, a multispectral vision system
is proposed to facilitate the segmentation of objects and
defect detection without extra expense of acquisition time.
In Figure 4 the imaging setup for defects detection of 2D
objects is illustrated. Multispectral cameras (denoted by
sensor A and sensor B) are perpendicularly installed
above the conveyor belt. Two fronts lighting (denoted by
Spectral lighting A and Spectral lighting B) are installed
between camera and conveyor belt. A back lighting
(denoted by spectral lighting C) is installed below the
conveyor belt. The benefit of this arrangement is that: the
contour of the target objects can be segmented from a
structured conveyor belt. Object segmentation is achieved
by the combination of reflected-light illumination (spectral
lighting A and B) and transmitted light illumination
(spectral lighting C). This step reduces the redundant
image data by focusing the region of interest direct on the
target object. In Figure 10 the segmentation of potato
slices from conveyor belt is shown.

(c)

(d)
Figure 3: Experiment with different bands of
electromagnetic radiation.

SensorA
Camerasensor

(a)

SensorB
Twoobjectives

Spectral
lightingA

Spectral
lightingB

Spectral
lightingC

(b)

Figure 4: Imaging setup for segmentation and defects

detection on 2D objects.
For the case of 3D objects, the condition is more
complex. For 3D weak texture object, the defective area
is mostly determined by the inhomogeneous geometric
shape. Therefore instead of texture inspection, 3D shape
of object should be estimated and analyzed. For 3D
texture object, defective area is determined by texture
analysis of object surface. Therefore in this case, surface
inspection and geometry reconstruction are always
combined to detect and locate the position of defects. In
[1], the prevalent technologies for the shape estimation of
3D objects have been compared. It has been concluded

that, the stereo vision with auxiliary pattern projection

(see Figure 5) is a competitive alternative for both shape
estimation and surface inspection. As shown in Figure 5,
the stereo vision consists of two stereo modules, each of
which consists of two monochrome cameras with
entocentric lenses and laser pattern generator. The
optical axes are oriented toward the object under
examination and are approximately antiparallel aligned
[1]. For inspection of a rotationally symmetric part one can
refer to [14].

Figure 5: Imaging setup for shape estimation of 3D

objects [1].
Camera selection
For the selection of camera, various criterions of the
inspection system must be taken into account. For
instance, line scan camera is predominant to inspect
moving objects. The following crucial criterions: (1) image
acquisition, (2) adaptiveness to variant conveyor velocity,
(3) accuracy, (4) object feature extraction ability and (5)
reciprocal economical factor are taken into account. In
Figure 6 a comparison of four preselected camera
systems which are used to image a moving object is
illustrated, and leads to a suggestion of a line scan color
camera.

Figure 6: Evaluation criterions and their rating for camera

selection.
Image processing algorithm
Image processing is the subsequent step to image
acquisition. It can be divided into the following steps: Preprocessing, object segmentation, defect detection,
features extraction and defect classification (review
Figure 1).
Texture analysis is the measure to detect surface defects.
As an important approach to region description, texture
analysis is used to detect surface defects by quantifying
texture content. Although no formal definition of texture

exists, intuitively this descriptor provides measures of

properties such as smoothness, coarseness, and
regularity [15]. The three principle approaches used in
image processing to describe the texture of a region are
statistical, structural, and spectral. For an image as
shown in Figure 10 neither appears obvious arrangement
of image primitives (e.g. regularly spaced parallel lines)
nor periodical spectrum (e.g. noise), the first principle
statistical approaches (standard deviation, normalized
variance, third moment, uniformity, entropy) are propose
as features.
Classification (employing features) of defects is quite
demanding in general. The algorithms have to meet the
expectations concerning the computational time, flexibility
and adaptability. These requirements can be reached by
application of support vector machine (SVM), which was
developed by Cortes and Vapnik [23] and is one of
supervised learning algorithms. Due to the increasing
number of features, an approach cross-validation and
Grid-search is introduced in [24] for finding the optimal
parameters, if RBF kernel was chosen for SVM.
Path planning and motion control
Path planning is the next issue of our vision system after
location of defective objects and their defects. Aljarboua
[25] introduced an interesting geometrical path planning
for general robot manipulator. However in our work, more
than obstacle avoidance and seeking shortest moving
path, path planning in our work is associated with path
achievability and maximization of regain value from
defective objects.
To achieve the planned path of processing tool
orientation, a special motion control program algorithm
has to be developed. This control program must be able
to process the transferred image data in such a way, that
a straight path can be generated. Thereby the movement
of conveyer belt is embedded into the control program as
a virtual encoder axes by using the current rotary encoder
information. Based on this, the movement of the
processing tool can be synchronized directly on the
movement of our conveyer belt. As both the image
processing and the control system are using the same
encoder information, a harmonized data communication is
eliminated. For the analysis of path achievability, current
state of manipulator is always up to date and taken into
account. And then within the achievable area, an optimal
path is chosen to obtain the maximal regain value. The
detected failure information including the path planning
information is reduced on sparse relevant factors, which
are written directly in a shift register of the motion control
program in real time. Inside the control program, the
particular values are executed consecutively to approach
the desired coordinates in time and cut out the failure
objects.
Conveyor belt
Conveyor belt for material feeding should be carefully
designed according to the following factors: shape of
conveyed objects, further treatment with respect to the
inspection result, desired productivity, hygienic standards
(for food industry), etc.. The following general conveyor
means and their variation have been compared in our
work.
a. Conveyor belts with nests
b. Rotary indexing table
c. V-Band
Moreover, the material of conveyor belt which is a crucial
influence factor to the treatment step on target objects
should be considered.

Treatment
The selection of a suitable processing tool is a major
challenge in order to the given problem formulation. For
the materiel processing, an efficient, fast and food
compatible system have to be found. Within an extensive
application analysis, the systems in Figure 7 have been
compared with the consideration of desired processing
task.
As a result of the high processing speed, the demand of a
linear section orientation per processing unit and the
massive operating costs of using compressed air brought
the selection to a one dimensional linear axis using a
water cutting system. In combination with the moving
conveyer belt, fast and effective cutting path can be
realized.

industrial computer PC-VU (vision unit), which is

equipped with a frame grabber that captures the image. In
PC-VU, the location of a target object and defects, as well
as information of defect types will be obtained by a
developed image processing algorithm. Meanwhile a
motion trajectory, which is regarding to selective
processing method and comprised of a list of coordinates,
is generated and sent to the second industrial computer
PC-PU (processing unit). For each coordinate, motion
command according to the current status of processing
tool (e.g. position, velocity and acceleration) and
conveyor belt are sent to motion control system of
processing tool. Data transmission between PC-VU and
PC-PU is synchronized via a field bus system. Figure 9(a)
displays the experimental setup of water jet as the
processing tool, which is motion controlled by a linear axis
manipulator; Figure 9(b) displays the water beam with a
diameter of 0.2 mm.

Figure 7: Processing possibilities for the give task

5 IMPLEMENTATION AND EXPERIMENTAL RESULT
Based on the theoretical research in previous section, the
feasibility and advantages of our approach will be proved
by the implementation with an example of agricultural
product quality control. The production of quality and
safety food is quite demanding in modern food industry. In
the production of e.g. potato chips, there is a variety of
requirements. For instance, some defective fraction,
which is harmless to health (e.g. peel), may be remained
or removed (according to the customer requirement);
some defective fraction, which is harmful to health (e.g.
green, seed), have to be removed; even the whole chip
should be given up. According to the system concept
described in Section 4.1, an experimental system of chips
inspection and processing has been developed and built
up.
line scan camera (color)
processing tool

bbz

conveyor belt

x
y
time to process
d

encoder

signal
distributor

PC-VU

data
communication

PC-PU

Figure 8: Configuration of inspection and processing

system.
Figure 8 represents the configuration of the inspection
and processing system. A line scan camera, which is
positioned above the conveyor belt and triggered by an
encoder signal, acquires the image of conveyed objects.
The image information is sent from the camera to the

(a)
(b)
Figure 9: Water jet motion controlled by a linear axis
manipulator.
Rotary encoder signal from conveyor motor is the
footstone of our control system. Benefits of using encoder
signal are: on the one hand, synchronization of unstable
conveying velocity and camera (allows to capture
conveyed objects precisely without consideration of
unstable velocity), on the other hand, synchronization of
vision system and processing tool (allows guiding
processing tool to reach target position).
The proposed multispectral vision system in Section 4.2 is
used for inspection of potato slice. The image processing
algorithm is implemented with the development platform
LabVIEW. The image processing is performed in the color
space HSL. Figure 10 displays some experimental
results; wherein, (a) displays the defects of peel, (b)
green and (c) decayed spot.
Path planning and processing strategy
Through variant test series, the development of an
overarching processing strategy was focused. These
come to the result, that the direct processing strategy is a
solution for our problem formulation. The planned path for
an online processing has already been taken place in
LabVIEW-based image processing algorithm. Figure 11
shows an example of a direct processing strategy using a
LabVIEW-based path planning. The equivalent ellipse
according to the geometry shape of a defect will be firstly
analyzed. Cutting path is planned to be parallel (with an
offset of s) to the major axis of the equivalent ellipse. If a
potato slice has a big defective fraction, after cutting only
a small non-defective fraction will be left. In this case, this
potato slice will be considered as a non-processable
object. Cutting path will be planned to go around the
potato slice with the avoidance of wrong treatment (e.g.
cutting a non-processable potato is a wrong treatment).
The weight of the cutting head-sets limits the maximum
acceleration of linear axis. Due to this system behavior,
the minimal accessible cutting angle is a key factor for
the processing strategy.

The edited geometrical points are pre-calculated to save

the time of motion control of the linear axis manipulator.

center point of the water jet. With the maximum speed of

the conveyer belt of 0.4 m/s the system is able to process
approximately 2-3 defective objects per second.

peel

(a)
Figure 11: Path planning based on the result of image
data analysis.

green

(b)

decayed

(c)
Figure 10: Result of image analysis.
5.1 Remove of defects by using water jet
The developed processing tool consists of a water jet
cutting head based on a high speed linear axis with a
maximum acceleration of 15 g. For the holding fixture of
the components and to absorb the processing forces, an
individual food compatible stainless steel frame was
developed and embedded in the system concept. The
linear axis is placed, as illustrated in Figure 9,
perpendicular to the moving direction of the conveyer belt,
so that a two dimensional cutting movement can be
realized.
Thereby the tool center point of the high pressure water
jet is adjusted on the coordinate zero point line (process
center line) of the image processing system to reach
cutting path directly and accurately.
The system behavior of the water jet cutting application
was tested in different experiments, using paperboards to
map the expected cutting path. Furthermore the properly
functioning of the developed data communication was
surveyed by entering the committed point with the tool

The removed defective fractions of potato slices may be

eliminated by pick and place robots. To approach this, the
center point of the removed defective fraction should be
given in LabVIEW-based image processing algorithm.
Due to the constraint of working speed of pick and place
robots, 3 cycle per second is overstrained.
6 CONCLUSION AND OUTLOOK
This paper presents a systematic development approach,
which is applied for the development of a vision-based
automatic inspection and path planning system for
processing conveyed objects. This approach is based on
existing
technologies
and
application-oriented
combination of required components. The novelty of the
proposed approach lies on the application-oriented
combination of technologies and control technology. By
using the described system, significant savings of
inspected products, such as the example in food industry,
can be achieved. The approach has been implemented
and tested on an example with the field of agricultural
production. The prototype system is able to automatically
inspect the surface of objects and remove the defects
from objects in the subsequent processing step. The
setup consists essentially of a conveyor belt, vision unit,
and control unit. For data processing and system control,
industrial PCs are used. They work in a synchronized way
based on the encoder signal, and communicate through
an industrial bus system. In the vision unit, the image
acquisition and analysis, classification and location of
defects are performed. The following processing system
allows the removal of the detected defects. The
experiment
demonstrates
practicability
of
the
development concept and functionality of the application,
which can be transferred to similar applications.
A further development of the system may be
accomplished by increasing the performance of the
automatic evaluation system as well as by optimizing the
processing unit. These both steps could achieve an
improvement in the system productivity.
7 ACKNOWLEDGMENTS
This work was funded by the European Fond for Regional
Development and the government of North Rhine
Westphalia under the reference number NA035D.
8 REFERENCES
[1] Weyrich, M., Laurowski, M., Klein, P., Wang, Y.-H.,
2011, A Real-time and Vision-based Methodology

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]
[16]

[17]

for Processing 3D Objects on a Conveyor Belt, Int.

J. of Sys. Appl. Eng. & Dev., Issue 4, Vol. 5, pp.
561-569.
Gracia, L., Perez-Vidal, C., Gracia, C., 2011,
Computer Vision Applied to Flower, Fruit and
Vegetable Processing, proceedings of ICPRCV,
year 7, issue 78, Amsterdam, Netherlands.
Wilson, A., 2011, Machine-vision-based inspector
sorts oranges and mandarins in : Vision Systems
Design, Volume 16, Issue 10.
Narendra V G, Hareesh K S, 2010, Quality
Inspection and Grading of Agricultural and Food
Products by Computer Vison A Review,
International Journal of Computer Application (0975
- 8887), Volume 2, No. 1.
Cubero S., Aleixos, N., Molto, E., Gomez-Sanchis,
J., Blasco, J., 2011, Advances in Machine Vision
Applications for Automatic Inspection and Quality
Evaluation of Fruits and Vegetables, Journal of Food
and Bioprocess technology, vol. 4, no. 4, pp. 487504.
Kim, S.-M., Lee, Y.-C., Lee, S.-C., 2006, Vision
Based Automatic Inspection System for Nuts
Welded on the Support Hinge, SICE-ICASE, pp.
1508-1512.
Kopacek, P., Han, M.-W., Putz, B., Schierer, E.,
Wurzl, M., 2005, An industrial image processing
system for fast moving objects, IEEE, pp. 278-282.
Laurowski, M., Kerstein, T., Klein, P., Weyrich, M.,
Roth, H., Wahrbug, J., 2011, One Shot Optical 3D
Surface Reconstruction of Weak Textured Objects
for an Agricultural Application, proceedings of 56th
International Scientific Colloquium IWK, Ilmenau.
Kerstein, T., Laurowski, M., Klein, P., Weyrich, M.,
Roth, H., Wahrburg, J., 2011, Optical 3D-Surface
Reconstruction of Weak Textured Objects Based on
an Approach of Disparity Stereo Inspection,
proceedings of ICPRCV, year 7, issue 78, p. 581586.
Gao, G., Chen, X., Zhang, H.-M., 2010, A Fast
Structured Light Matching Method in Robot Stereo
Vision, in Journal Applied Mechanics and Materials,
vol. 29-32, pp. 1981-1984.
Ambrosch, K., Kubinger, W., 2010, Accurate
hardware-based stereo vision, in Journal of comput.
Vis. Image Understanding, vol. 114, pp. 1303-1316.
Bonilla, I., Galvan, E.-J., Olivares, C.-C., Mendoza,
M., Loredo-Flores, A., Reyes, F., Zhang, B., 2010, A
Vision-based Impedance Control Strategy for
Industrial Robot Manipulators, 6th annual IEEE on
CASE, pp. 216-221.
Zhang, B., Wang, J.-J., Rossano, G., Martinez, C.,
2011, Vision-guided Robotic Assembly Using
Uncalibrated Vision, ICMA, Issue 7-10 Aug., pp.
1384-1389.
Laurowski, M., Klein, P., Weyrich, M., Scharf, P.,
2011, Use-Appropriate Design of Automated Optical
Inspection Systems for Rotationally Symmetric
Parts, 56th IWK.
Gonzalez, R.-C., Woods, R.-E., 2008, Digital Image
Processing, Pearson Education, Int., third edition.
Aiteanu, D., Ristic, D., Graser, A., 2005, Content
Based Threshold Adaptation for Image Processing
in Industrial Application, ICCA2005, Vol. 2, pp.
1022-1027.
Lu, R., Park, B., 2008, Hyperspectral and
multispectral imaging for food quality and safety,

[18]

[19]

[20]

[21]

[22]
[23]
[24]

[25]

Sensing and Instrumentation for Food Quality and

Safety, Jg.2, pp. 131-132.
Brosnan, T., Sun, D.-W., 2002, Inspection and
grading of agricultural and food products by
computer vision systems a review, in Computers
and Electronics in Agriculture, Jg. 36, H. 2-3, pp.
193-213.
Kker, R., Oz, C., Ferikoglu, A., 2001, Development
of a Vision based Object Classification System for
an Industrial Robotic Manipulator, IEEE, Vol. 3, pp.
1281-1284.
Marsland, S., 2009, Machine Learning An
algorithmic Perspective, CRC Press, Taylor &
Francis Group.
Sumathi S., Surekha, P., 2009, Computational
Intelligence Paradigms, CRC Press, Taylor &
Francis Group.
Lughofer, E., 2011, Evolving Fuzzy Systems,
Springer-Verlag Berlin Heidelberg, 1st Edition.
Cortes, C., Vapnik, V., 1995, Support-vector
network, Machine Learning, 20, 1-25.
Hsu, C.-W., Chang, C.-C., Lin, C.-J., 2010, A
Practical Guide to Support Vector Classification,
Department of Computer Science, National Taiwan
University, Taipei 106, Taiwan.
Aljarboua Z., 2009, Geometric Path Path Planning
for General Robot Manipulators, proceeding of
WCECS.