International Journal of Scientific and Technical Advancements

ISSN: 2454-1532

Evaluation of the Accuracy of Genetic Algorithms in

Orientation Estimation of Objects in Industrial
Environment
Ankit Sharma1, Raminder Preet Pal Singh2, Parveen Lehana3
1

Electrical Engineering Department, Model Institute of Engineering and Technology, Jammu, Jammu & Kashmir, India-181122
2
Electronics and Communication Engineering Department, Arni University, Kathgarh, Himachal Pradesh, India-176401
3
Department of Physics and Electronics, University of Jammu, Jammu, Jammu & Kashmir, India-180006
E-mail: 1svankit@yahoo.co.in, 2raminder_212003@rediffmail.com, 3pklehana@gmail.com.

AbstractThe world of machine vision and robotic vision revolve around a wide array of techniques and methodologies built around image
processing. While conceiving and developing any such technique, due importance is accorded to the nature of the problem to be addressed
and the ultimate purpose to be realized. For instance, realizing object recognition in an image environment having multiple objects, realizing
object detection and recognition in an image environment characterized by occlusion and clutter and many other unique scenarios like these.
For realizing object detection, localization eventually leading to object recognition in environments characterized by multiple objects and
occlusion and the environments having objects undergoing different rotations in an image plane; a proper estimate of object orientation is
central to an appropriate pose estimation of object which in turn plays a vital role in accurate recognition of the object. For realizing object
recognition in such unique scenarios, orientation and pose estimation will go hand in hand. In the present study, an attempt is made to
accurately estimate the orientation of a single industrial object in an image using genetic algorithm (GA), a nature inspired evolutionary
technique for optimization and thus facilitates in evaluating the potential and accuracy of the GA when used as a standalone GA in providing
a correct orientation estimate. The analysis of the results presents the GA as a reliable, potent and efficient tool for suitably estimating the
orientation of the single object in the image.
Keywords Object detection; object orientation; robotic vision; image segmentation; image thresholding; genetic algorithm; selection;
mutation.

I.

indispensable component of modern day intelligent systems

which have shaped and cut across a broad spectrum of
disciplines in contemporary human life such as security,
health, defence, surveillance, medical diagnostics etc. where
the issue of object detection and recognition need to be
handled quickly and accurately. The research on object
detection and recognition algorithms has spearheaded
significant advancements in factory and office automation,
assembly line industrial inspection systems as well as chip
defect identification systems [6]. It has also resulted in
appreciable and tangible progress in medical imaging, space
exploration and biometrics. Different researchers over the
years have approached the problem of object recognition and
localization by developing methodologies based upon
different techniques which are defined by the nature of the
problem to be addressed. It can be easily observed that object
detection and recognition are significantly influenced by the
confounding parameters of pose, orientation, scale of object
and environment parameters such as intensity, illumination
[5]. The parameters of pose and orientation are observed to
acquire more prominence when object detection and
recognition are to be realized in environments which are
characterized by occlusions, multiple objects in images,
cluttering, overlapped objects in images and where the objects
of a specific category undergo different orientations in the
image plane [7-10]. It is also observed that the orientation
parameter is given more attention than position and scale
parameters when developing techniques centered around the

INTRODUCTION

bject detection and recognition are indispensable

and integral to the domains of machine vision and
robotic vision systems [1]. The functional quality
of robotic vision systems employed in both industrial robots
for industrial automation and service robots for household
applications is greatly influenced not only by the hardware
involved but also by the choice of the software designed and
deeply embedded in such systems to drive them for the
purpose of realizing object detection and recognition [1], [2].
These impart the desired capability to the robotic vision
systems for facilitating inspection, localization, registration,
and manipulation of the objects for automation in different
industries and extending the desired services in household
applications. Object detection and recognition in the real
environment has always presented formidable challenges to
the engineering and scientific community since it first
appeared on the horizon of the machine vision research [3].
Recognizing and detecting the objects of a particular class
such as a human face, a car, a bird, an animal etc. in the static
images make it even more challenging [4]. Notwithstanding
the various complexities and diverse challenges associated
with the domain of machine vision, the research in this domain
has grown rapidly beyond these complexities especially in the
last two decades since it has been both intensively and
extensively researched across the scientific and engineering
fraternities [5]. Object detection and recognition constitute an
7

Ankit Sharma, Raminder Preet Pal Singh, and Parveen Lehana, Evaluation of the accuracy of genetic algorithms in orientation estimation of
objects in industrial environment, International Journal of Scientific and Technical Advancements, Volume 1, Issue 4, pp. 7-14, 2015.

International Journal of Scientific and Technical Advancements

ISSN: 2454-1532

accurate pose estimate of the object to realize object

recognition in such scenarios because even a small difference
in orientation parameter can induce a significant change in the
appearance and shape of the target object for the machine and
robotic vision systems to recognize them appropriately and
this can lead to unpredictable and undesired results [11], [8].
In this research paper, we present our work which is
centered around the implementation of the GA in a standalone
mode and which is upgraded from its original design used
earlier in the detection of industrial objects in our previously
presented work [12] to measure accurately the orientation of
the objects which in the present study are the tools of everyday
use in industrial work environment such as a screw driver, a
plier, adjustable spanner etc. This work is undertaken not as a
comparative study but as an experimental endeavour to
evaluate the accuracy of the GA in estimating the orientation
of the industrial objects when the GA is designed to work
alone on its own, taking complete charge of the problem of
estimating the orientation of the object and eventually leading
to object detection utilising only its inherent evolutionary
attributes such as initialization, reproduction, cross-over,
mutation. In this way, this work implements the GA and
explores its potential, when used in standalone mode, in
accurately estimating the orientation of a single object in nonoccluded and uncluttered static image environment.
The outline of rest of this paper is structured as follows.
Section II highlights some of the relevant research
contributions made in the field of object detection and
recognition centered around the relevance of accurate
orientation and pose estimation in diverse environments which
inspired our research. Section III gives major insights into the
relevant technical constituents which form the foundation of
the presented work. Section IV gives an analysis of the
proposed GA design with regard to its evolutionary attributes
and highlights the flow chart of the proposed GA. Section V
provides the outline of the salient steps in the sequence of flow
of GA. Experimental results and performance evaluation are
covered in Section VI. Conclusion and future scope are
summarized in the last section i.e. Section VII.
II.

A. OrientationA Valued Contributor in Object Recognition

Scale invariant feature transform based method (SIFT) to
successfully recognize the position and the orientation of the
objects for an automated pick and place robotic system used in
industrial and household applications was proposed by Patil
and Chaudhari in their research paper [7]. They emphasized
that recognizing the correct location and orientation of the
given object are central to the ultimate purpose of a robust
robotic vision system. In addition to reviewing different
techniques and algos implemented by different researchers for
the purpose of object detection and localization, the authors in
their experimental results showed that SIFT outperforms the
other methods based on other feature descriptors and emerges
as the best method for object localization and recognition in
pick and place robotic systems.
A simple object recognition method was presented using
the singular value decomposition (SVD) of the object image
matrix and a functional link neural network for a bin picking
vision system to be employed in a bin picking robotic system.
The problem of recognizing objects inside the bin is quite
complex and challenging since the appearances and shapes of
the objects undergo tremendous change owing to their varied
degrees of orientation for the bin picking robots to recognize
them easily. The authors Hema C.R. et al. [8] did not utilize
the visual features of the pixels but used singular value (SV)
features properties of the image matrix for its analysis and
feature extraction. The properties of the SV features played an
important role in the recognition of the objects with different
orientations.
The authors U. Bagci et al. [11] in their research work
presented a model based multi-object recognition method to
assist in recognition of anatomical structures in medical image
segmentation. They observed that the orientation parameter
requires more attention as compared to the scale and the
position parameter in estimating the accurate pose of the
structure in the image as even a small difference in orientation
can lead to an inappropriate recognition. The authors
statistically analyzed and evaluated that the mean Herrmitian
and Cholesky metrics provided more accurate orientation
estimates than other Euclidean and non-Euclidean metrics.
A suite of algorithms was presented after implementation
on an autonomous robot for the purpose of its navigation so as
to determine the most optimal trajectory to reach its
destination while moving through the obstacles on its course
by C. Ilas et al. [9]. The authors implemented two algorithms
for navigation. First one was for the accurate estimation of the
object orientation based upon the canny edge detection and Kmeans clustering and the second one was for the determination
of the collision free paths after comparing the distances
between the obstacles on the path. The output of the first was
used by the second algo and both could work on robots with
medium computational resources.
M. Villamizar et al. [10] presented an approach for the
robust detection of the specific classes of objects that might
appear in the still images under different orientations. Instead
of addressing the problem by using the traditional multiple
classifiers specifically trained at different orientations leading

RELATED RESEARCH AND CONTRIBUTIONS

Informative literature and work documentation are

available in abundance on different techniques and
methodologies adopted to realise object detection and
localization in diversified contexts taking due cognizance of
the relevant object attributes such as orientation, scale and
position. Motivating works by different researchers also
provide creative insights into the versatility of GAs when it
comes to its implementations in variety of real world domains
such as business, sciences, medical diagnostics etc. to extract
reliable and optimal solutions in optimisation problems
integral to such fields. Several such substantial works which
contributed in motivating, guiding and charting the course of
this work may be grouped into three distinct categories:
orientationa valued contributor in object recognition, image
segmentation, and genetic algorithm.

8
Ankit Sharma, Raminder Preet Pal Singh, and Parveen Lehana, Evaluation of the accuracy of genetic algorithms in orientation estimation of
objects in industrial environment, International Journal of Scientific and Technical Advancements, Volume 1, Issue 4, pp. 7-14, 2015.

International Journal of Scientific and Technical Advancements

ISSN: 2454-1532

to high computational costs in both training and test stages, the

authors split the approach in the two stages, the object pose
estimation and the orientation specific classification for
efficient detection of objects in images. The classifiers in both
the stages were based upon the boosted combination of
random ferns evaluated densely over local histograms of
oriented gradients (HOGs). The experimental results showed
the method to be competitive with the state of the art methods
with the benefit of being computationally more efficient.

combined with other good optimization techniques such as

BFO etc.
The authors Karkavitsas and Rangoussi in their research
paper presented GA approach to the problem of object
registration in the medical images of the blood cells [18].
Their work also highlighted underlying basic mechanisms of
work in GA such as parent selection, reproduction etc. The
work showed that the success and efficiency of the GA is
critically dependant on the choice of the evaluation (fitness)
function and an appropriate choice of the parameters.
P. Kanungo et al. [19], in their research paper proposed a
GA based crowding algorithm to determine a suitable
threshold value from the peaks and valleys of the histogram
with bimodal features which can be used for the purpose of
Image segmentation. They emphasized that gray-level
thresholding is an important step in any image analysis
application. This GA based approach worked well for images
with bimodal features but did not perform well for images
with trimodal features.
A method in which GA was used on Fourier descriptors
which were used to represent the shape descriptive features of
an object was presented by the authors Mahmood Ul Hassan et
al. [20] in their research paper. The authors also compared the
GA based method with the PSO technique in their experiments
and observed that GA based technique outperformed the PSO,
since the GAs work in the direction to maintain a population
of potential solutions to a problem at any stage and not just
one point solution as can be seen in techniques like PSO.

B. Image Segmentation
Navneet kaur et al. presented a colour image segmentation
algorithm based upon BFO technique [13]. The basic step in
the proposed algorithm was the quantization of the colours in
the image without degrading the quality of the colours. The
proposed algorithm implemented region growing to obtain the
segmentations in the image and these segmentations generated
the objects of interest in the image.
In their research paper, the authors Amrinder Singh and
Sonika Jindal argued that image segmentation affects the
subsequent processes of image analysis such as object
classification, scene interpretation [14]. Researchers are
continuously trying to improve the quality of image
segmentation by fusing BFO with GA and PSO. Their
technique used the ANFIS edge detector for edge detection on
digital images. It involved a system with the learning
capability of neural network and the advantages of rule based
fuzzy system.
The authors Manjusha Singh et al. in their research paper
conveys that image segmentation occurs as the preprocessing
step before image pattern recognition, image feature extraction
and plays an important role in computer and machine vision
especially in human tracking [15]. In their work, the authors
have voted in favour of methods of image segmentation based
upon the visual principle as these can be employed as
hierarchical approaches that do not require any user input and
are found to perform well especially for the image with single
object in prominence.
Rajeshwar Dass et al. [16] in their research paper showed
that the implementation of the image segmentation techniques
can be usefully utilized in the navigation of the robots,
filtering of the noisy images, medical applications like
location of tumors, cancerous cells, computer-guided surgery,
in locating objects in satellite images e.g. roads, rivers, forests
etc. The authors have clearly deduced that despite several
segmentation techniques available, there is not a single
method which can be considered good and applied for
different types of images equally well.. So there is a great need
to develop a unified approach to image segmentation which
can be applied to all types of images.

III.

TECHNICAL APPROACHKEY COMPONENTS

The major components which constitute the core of the

proposed methodology for orientation estimation of objects in
industrial environment are covered in this section. Although
the proposed methodology is centered around the design of the
GA but for its proper understanding, it can be conveniently
partitioned into two key components, image thresholding, the
Genetic Algorithm--an overview.
A. Image Thresholding
Proper segmentation of an image is the foundation of
object detection and recognition in machine vision [21]. The
purpose of image segmentation is to locate the edges or
boundaries of object in the image which help in feature
extraction and finally object detection and recognition. Edge
detection is a very important step in image analysis and image
segmentation. Object detection and recognition directly
depends upon the quality of the edges detected. Image
segmentation can be realized by adopting different techniques
and approaches. The popular among these are histogram
thresholding, edge based segmentation, region based
segmentation which itself employs different techniques such
as thresholding, splitting and merging, region growing [22].
Gray-level-thresholding is an important first step in any
image analysis application whose purpose is to acquire a
binary image of a gray scale image, i.e., the image which
contains any two pixel values either 0 or 1. In gray-levelthresholding, a gray scale image is divided into distinct

C. Genetic Algorithm
Bajpai and Kumar in their research paper conveyed that
GA is quite reliable in generating optimal solutions to
optimization problems in signal processing, robotic vision
systems, medical imaging, object localization, stock market
and variety of other fields [17]. They also argued that in order
to make GAs more effective and efficient, they should be
9

Ankit Sharma, Raminder Preet Pal Singh, and Parveen Lehana, Evaluation of the accuracy of genetic algorithms in orientation estimation of
objects in industrial environment, International Journal of Scientific and Technical Advancements, Volume 1, Issue 4, pp. 7-14, 2015.

International Journal of Scientific and Technical Advancements

ISSN: 2454-1532

components like foreground objects, where pixel value is 1

and background objects, where the pixel value is 0 by
partitioning the pixels in the gray scale image into foreground
(object) and background classes based upon the relationship
between the gray level value of a pixel and a significant gray
level threshold parameter selected in that gray level image in a
systematic way so as to separate the object from the
background in the image [19].
In the proposed methodology, gray level thresholding has
been incorporated into the GA design taking due cognizance
of this fact that a RGB (True colour) image can actually be
resolved into three distinct gray scale images where each of
the three gray scale images correspond to the R-component
(Red colour), G-component (Green colour) and B-component
(Blue colour) image of the actual RGB image. The
implementation of the gray level thresholding in the proposed
GA results in the binary image of a RGB image shown in
figure 1 with the boundaries or edges of the object in the RGB
image properly detected and clearly shown outlined in white
colour against a black background.

(a)

The GA is a search heuristic that generates a population of

individuals (potential solutions) and allows them to evolve to a
state of maximum fitness by following Darwinian rules of
natural selection and evolution. Thus the GAs promote the
survival of the fittest [23], [24] and this fittest corresponds to
the optimal solution of the search and optimization problem.
Any simple GA incorporates the following five steps in its
functioning which are derived from nature's principles of
natural selection and genetics. It is incorporation of these
principles in it's functioning that imparts GAs its
quintessential genetic character. The five primary operations
employed by GAs in their functioning are initialization,
evaluation and selection, reproduction, cross-over and
mutation.
IV.

THE PROPOSED GA DESIGN ANALYSIS

The GA which is implemented in the present work in order

to evaluate the orientation of the object in industrial
environment is actually an upgradation of its original design as
implemented in our previously presented work [12] and
incorporates the same inherent attributes which constitute the
foundation of any simple, modified and hybrid GAs. This
section presents an overview of the analysis of the GA
designed and upgraded by us for the purpose of orientation
estimation with regard to the essential genetic attributes of
initialization, evaluation, selection, reproduction, cross-over
and mutation.
A. Initialization
Initialization in GA refers to the randomly generated initial
population of the individuals (potential solutions) in a search
problem [17]. In GA parlance, each such individual is a
chromosome which comprises of several genes, i.e., each
chromosome can be held as a string of several genes. The
proposed GA generates an initial population of ten
chromosomes in a search space and each chromosome is
assigned nine genes such as orientation, intensity, scale etc. of
a pixel in the image.

(b)

Fig. 1. Preprocessing of the input RGB image. a) RGB image of an adjustable

spanner, b) Preprocessed binary image with edges shown in white.

B. The Genetic AlgorithmAn Overview

The GAs for the last 20 years have been extensively in use
and gained much popularity in the domains of image
processing, pattern recognition, machine vision and object
detection. GA is a nature inspired search and optimization
technique which derives it's functional character from the
principles of natural selection and genetics [23]. Today the
engineering and scientific community relies on this technique
as a potent answer to wide array of real world optimization
problems in fields like digital signal processing, robotic vision
systems, medical imaging, object detection and recognition
just to mention a few [17].
The use of GAs is highly recommended by the researchers
in all such problems where an optimal solution is to be
extracted from a potentially huge solution space. In such
optimization problems, the GAs have been observed to
outperform the other traditional optimization techniques such
as calculus based optimization, hill climbing, feed forward
ANNs (Artificial Neural Networks) [17], [23]. GAs also score
over other nature inspired evolutionary techniques like
simulated annealing, PSO etc. in the context that GAs always
work in the direction of maintaining a population of potential
solutions to a problem at any stage of their processing and not
just one point solution as can be seen in techniques like PSO
[20].

B. Fitness Evaluation
To select which individuals in the initial population will be
favoured to breed to create the next generation, the fitness of
every individual must be evaluated [24], [25]. The proposed
GA evaluates the fitness (health) of each of the ten
chromosomes in the context of suitable orientation in the
initial population and arranges them in the ascending order of
their fitness through a set of appropriate instructions
incorporated in it.
C. Selection
In this step of the GA, two individuals depending upon
their fitness calculated in the current population are selected as
parents to breed a new generation of individuals [17], [23].
Roulette wheel selection and tournament selection are the two
popular methods in this context but we implement a random
selection method in the GA in which the fittest individual in
the current generation and a randomly selected individual from
the rest of the nine individuals will be selected to breed in
10

Ankit Sharma, Raminder Preet Pal Singh, and Parveen Lehana, Evaluation of the accuracy of genetic algorithms in orientation estimation of
objects in industrial environment, International Journal of Scientific and Technical Advancements, Volume 1, Issue 4, pp. 7-14, 2015.

International Journal of Scientific and Technical Advancements

ISSN: 2454-1532

order to reduce the selection time and adding diversity to the

population in the solution space.

otherwise proceed to the next step to create next

generation.
3.4 Create he next generation of candidates by following
the steps a) to c)
a) Select the best fit candidate from P and randomly
select the other candidate from (P-1) as parents for
the cross-over.
b) Perform the single point cross-over after randomly
selecting the cross-over point between the parents
selected in the step a)
c) Replace the two lowest fit candidates in P with the
two children from cross-over to create the next
generation.
3.5 Perform the mutation on one of the two randomly
chosen child in the new population acquired in step c)
3.6 Proceed to the step 3 above for the repetition of all
the steps from 3.1 to3.5 with the new population

D. Reproduction
In the reproduction phase of the GA, the population of the
next generation is created by implementing the two basic
methods, cross-over and mutation [17], [25]. For every new
child in the next generation, a pair of parents is selected from
which the child inherits it's properties. In the GA proposed, we
incorporate both the methods for reproducing the next
generation.
E. Cross-over
The cross-over operator of the GA selects genes from the
parent chromosomes and creates new offsprings [17], [23],
[24]. The simplest way to realize cross-over is to randomly
select a cross-over point (The locus position of a gene) and
copy everything before this point from the first parent and
after the cross-over point copy from the second parent. This is
known as single point cross-over which results in two
offsprings. Two points, multiple point and other cross-over
techniques can also be used in GAs.
A single point cross-over technique is implemented in the
proposed GA after allowing it to randomly select it's crossover point so as to perform the mating of the two parents and
exchange their genes across this point to create two new
offsprings for the next generation.
F. Mutation
The purpose of the mutation is to enable the GA to explore
new areas of the search space and prevents all the solutions in
the population from falling in the local optima of the solution
space by introducing and preserving genetic diversity [17],
[25]. Mutation is incorporated in our GA for those
chromosomes which are assigned the two lowest positions of
fitness in every generation from the second generation
onwards so as to preserve the best chromosomes having the
highest fitness in every generation. The mutation of the best
chromosomes in every generation is always advised against
[23] and a very low percentage of the population in the
generation should be mutated
V.

SALIENT STEPS IN ALGORITHM FLOW

The salient steps in the sequence of the algorithm flow as

can be observed from the flow chart of the GA shown in figure
2 are the following:
1. Randomly generate an initial population P of candidates,
each having systematically selected an appropriate
orientation value in addition to other genes.
2. Select an appropriate number of generations N.
3. For generations 1 to N do the following. // Main Loop.
3.1 Evaluate the fitness of each candidate in the
population P in the context of the appropriate
orientation.
3.2 Arrange the candidates in the ascending order of their
fitness.
3.3 If the generation limit has reached N then come out
of the main loop and return the best fit candidate,

Fig. 2. Flow chart of the genetic algorithm.

11
Ankit Sharma, Raminder Preet Pal Singh, and Parveen Lehana, Evaluation of the accuracy of genetic algorithms in orientation estimation of
objects in industrial environment, International Journal of Scientific and Technical Advancements, Volume 1, Issue 4, pp. 7-14, 2015.

International Journal of Scientific and Technical Advancements

ISSN: 2454-1532

VI.

EXPERIMENTAL RESULTS AND PERFORMANCE

EVALUATION

original position in the reference image. Due to the orientation

induced, the corners of the chosen reference image are left
fragmented white which are then filled with the colour to
match as closely as possible with the colour of the reference
image background to make the background of the test image
consistently as close as possible with the background of the
chosen reference image for the smooth and proper processing
of the test image by the GA. This is precisely the reason for
the presence of the fragmented corners which can be observed
in all the test images shown in the figure 3. These test images
are then provided as inputs to the GA in several experiments.
It was observed that in most of the experiments, the GA
behaved reliably well and performed as per expectations. It
succeeded in providing a near accurate estimate of the
orientation of the object in the test image. The orientation
estimate of the object in the test image was arrived at by the
GA by working in the counter clockwise direction from the
original position of the object in the reference image. In these
successful experiments, the GA was observed to arrive at the
correct result by evolving less than half of the preset
maximum number of generations.
In few experiments, the GA did not perform as per
expectations and gave undesired results. In these experiments,
it is observed after analysis that sometimes the GA was
observed to get stuck on a test image with orientation already
estimated in one of its last two or three runs. To overcome this
problem, the GA was imparted a little training as a gentle
reminder of its successful runs by making it to process those
test images in which it arrived at near accurate orientation
results so as to facilitate it to rediscover its course of arriving
at correct orientation estimates. With this training, the GA was
really able to turn around its performance in all these
experiments.
In all the experiments conducted, the test images of various
industrial objects captured in different light colours were used.
It was observed that the various light colours did not have any
effect on the performance of the GA in the accurate estimation
of the object orientation.
The visual illustrations and results of the performance of
the GA in six of the several experiments conducted with
various industrial objects having different orientations are
highlighted in figure 3. The test image column in figure 3
shows the test images of different industrial objects at
different orientations captured in different light shades
(colours) and used as input images for the GA in experiments.
The reference image column in figure 3 highlights the
reference orientation of the object of a specific type with
regard to which the orientation of the object of the same type
in the test images is estimated in the counter clockwise
direction. The last two columns in figure 3 gives the value of
the actual orientation induced in the object of the reference
image and the orientation of the object estimated by the
processing of the test image by the GA. The GA arrives at the
appropriate orientation estimate of the object in the test image
by processing the test image in the counter clock wise
direction with regard to the object orientation in the reference
image taken as the zero orientation value.

A. Test Images
Images of the various industrial objects that are frequently
employed in industrial work environment such as screwdriver,
plier, adjustable spanner, open-ended spanner etc. are first
captured with Sony's DSCW320 14MP digital camera under
different light shades (colours) such as red, yellow, blue and
white. Each object image is a JPEG image of 4320 X 3240
pixels resolution which is then resized and converted into
BMP image of 100 X 100 pixels which reduces the cost of
processing in terms of time and memory requirements
multifold, since processing an image with a higher pixel
resolution will consume more time and can prove to be quite a
heavy burden on the memory [26].
Then in each such BMP image, the boundaries of the
object are marked because of the various advantages
associated with the use of markers in image segmentation [27].
It is used primarily to achieve gray-level segmentation using a
single global threshold for the purpose of detecting the edges
of the object and to take out any shadows and noise
accompanying the object in the image and which can be
caused due to the use of light in specific position. These BMP
images are then submitted to the image database which
contains 10 images only comprising of images of various
industrial objects captured in different light shades. These
images in the image database are used in experiments as the
reference images by the genetic algorithm for the accurate
estimation of orientation of the object of the same type in the
test image given as input to the GA. The orientation of the
object of a specific type in the reference image is taken to be
equal to zero by the GA while processing the test image for
accurate estimation of the orientation of the object of the same
type in it. The image database is deliberately confined to a few
images only instead of being made an expansive one owing to
the ultimate objective to be realized in this work which is
centered around the orientation estimation of the object in the
image and not the detection of a particular object from a large
pool of available objects of the different types in the image
database. The purpose behind capturing the images of
industrial objects in different light shades is to determine
whether the GA design shows any sensitivity to the light
colour in the process of estimation of object orientation since
the real industrial environments are also sometimes exposed to
the lights of different colours.
B. Performance and Results
Experiments were conducted using different images of the
various industrial objects as the test images for the GA. The
test images are of the same industrial objects whose images
are kept in the image database but with different object
orientations. These orientations of the specific degrees are
induced in any chosen reference image kept in the image
database by means of any appropriate software such as Corel
photo-paint 10. The software incorporates the orientation of
the desired degree in the chosen reference image thus inducing
orientation of the same degree in the object with regard to its
12

Ankit Sharma, Raminder Preet Pal Singh, and Parveen Lehana, Evaluation of the accuracy of genetic algorithms in orientation estimation of
objects in industrial environment, International Journal of Scientific and Technical Advancements, Volume 1, Issue 4, pp. 7-14, 2015.

International Journal of Scientific and Technical Advancements

ISSN: 2454-1532

Reference Image

*CCW - Counter clockwise

Test Image

Orientation
(CCW*) Induced in Object
in Degrees

Orientation (CCW)
Estimated by GA WRT**
Reference Image

320

328

150

153

330

335

190

197

320

317

50

44

Fig. 3. Visual illustrations and the results of the performance of the genetic algorithm.
**WRT - With regard to

orientation estimates of various industrial objects (tools) of

different shapes at varied degrees of orientations without any
prior training and learning, shows immunity to the effects of
various light colours red, blue, white and yellow used in the
test images, converges to the optimal solution with a high
degree of accuracy without requiring a high number of
generations to evolve, say above 100. In a nutshell, this GA
can be heralded as an adaptive, accurate and efficient in design
and in its approach towards realizing orientation estimation
accurately. But at the same time, it will be pertinent to
mention that it also shows some functional constraints in a few
cases where the need is felt for a little training as a reminder to

VII. CONCLUSION AND FUTURE SCOPE

In this paper, efforts have been made to address the
problem of orientation estimation of objects in industrial work
environment using stand-alone GA and to evaluate its
accuracy for the same. In the proposed methodology, the GA
is designed to work in the stand alone mode taking complete
charge of the problem relying solely on its evolutionary
attributes to realize the objective of orientation estimation of
objects. In its implementation in several experiments
conducted as part of this evaluation, the GA design is
observed to adapt quickly to come up with near accurate
13

Ankit Sharma, Raminder Preet Pal Singh, and Parveen Lehana, Evaluation of the accuracy of genetic algorithms in orientation estimation of
objects in industrial environment, International Journal of Scientific and Technical Advancements, Volume 1, Issue 4, pp. 7-14, 2015.

International Journal of Scientific and Technical Advancements

ISSN: 2454-1532

[10] M. Villamizar, F. M. Nogeur, J. A. Cetto, and A. Sanfeliu, Efficient

rotation invariant object detection using boosted random ferns,
Proceedings of the 2010 IEEE Conference on Computer Vision and
Pattern Recognition, pp. 1038-1045, 2010.
[11] U. Bagci, J. K. Udupa, and X. Chen, Orientation estimation of
anatomical structures in medical images for object recognition, Journal
of Progress in Biomedical Optics and Imaging, vol. 12, no. 31, pp.
79622L.1-79622L.9, 2011.
[12] A. Sharma, R. P. P. Singh, and P. Lehana, Evaluation of the accuracy
of genetic algorithms for object detection in industrial environment,
International Journal of Scientific and Technical Advancements, vol. 1,
issue 3, pp. 105-111, 2015.
[13] N. Kaur, R. Singh, and R. K. Tulli, Bacteria foraging based image
segmentation, International Journal of Engineering Sciences, vol. 6,
pp. 141-150, 2012.
[14] A. Singh and S. Jindal, A systematic way for image segmentation based
on bacteria foraging optimization technique, International Journal of
Computer Science and Information Technologies, vol. 5, issue 1, pp.
130- 133, 2014.
[15] M. Singh and A. Misal, A survey paper on various visual image
segmentation techniques, International Journal of Computer Science
and Management Research, vol. 2, pp. 1282-1288, 2013.
[16] R. Dass, Priyanka, and S. Devi Image segmentation techniques,
International Journal of Electronics and Communication Technology,
vol. 3, issue 1, pp. 66-70, 2012
[17] P. Bajpai and M. Kumar, Genetic algorithm-an approach to solve
global optimization problems, Indian Journal of Computer Science and
Engineering, vol. 1, no. 3, pp. 199-206, 2010.
[18] G. Karkavitsas and M. Rangoussi, Object localization in medical
images using genetic algorithms, International Journal of Signal
Processing, vol. 1, no. 4, pp. 204-207, 2004.
[19] P. Kanungo, P. K. Nanda, and U. C. Samal, Image segmentation using
thresholding and genetic algorithm, Image Processing and Computer
Vision Lab, NIT, Rourkela, 2010.
[20] M. ul Hassan, M. Sarfraz, A. Osman, and M. Alruwaili, Object
recognition using particle swarm optimization and genetic algorithm,
International Journal of Computer Science Issues, vol. 10, pp. 55-64,
2013.
[21] P. Pachunde, V. R. Ingle, Mahindra, and A. Gailwad, Segmentation of
colour images using genetic algorithms-A survey, IOSR Journal of
Electrical and Electronics Engineering, vol. 1, issue 6, pp. 09-12, 2012.
[22] N. Kaur, R. Singh, and R. K. Tulli, Bacteria foraging based image
segmentation," International Journal of Engineering Sciences, vol. 6,
pp. 141-150, 2012.
[23] R. L. Haupt and S. E. Haupt, Practical Genetic Algorithms, River Street,
Hoboken, New Jersey: John Wiley & Sons, Inc., 2004.
[24] K .S. Tang, K. F. Man, S. Kwong, and Q.He," Genetic algorithms and
their applications, IEEE Signal Processing Magazine, vol. 13, no. 6, pp.
22-37,1996.
[25] A. Masum, M. Shahjalal, M. Faruque, and M. Sarker, Solving the
vehicle routing problem using genetic algorithm, International Journal
of Advanced Computer Science and Applications, vol. 2, no. 7, pp. 126131, 2011.
[26] D. Kumar, Implementation of bacteria foraging optimization for colour
image quantization and its evaluation for various file formats,
International Journal of Computer Science and Communication
Engineering, vol. 2, issue 1, pp. 36-42, 2013.
[27] R. C. Gonzalez and R. E. Woods, Digital Image Processing, Upper
Saddle River, New Jersey: Prentice Hall, 2004.

the GA in order to enable it to rediscover the course welltreaded by it in its earlier runs towards correct orientation
estimation.
The future work will be focussed on taking this design of
GA and modifying it not only to iron out the little weaknesses
it exhibited in a few cases but also for the accurate estimation
of object localization and scale estimation based object
detection. Attempts will also be made to address the problems
of recognition of multiple objects in images based upon
correct orientation and pose estimations, to detect and
recognize overlapped objects in images centered around
accurate orientation estimates of objects, to detect and
recognize occluded objects in images. The GA will also be
modified to address the problems of detection and recognition
based upon accurate pose and orientation estimates in cluttered
environments. The GA design in its current form can be taken
to fuse with some other core technique which has a proven
track record of reliability and efficiency to address the core
problems in the aforementioned areas and allied areas as
hybrid GAs have been observed to perform better and reliably
well in most of these areas
REFERENCES
[1]

[2]

[3]

[4]

[5]
[6]

[7]

[8]

[9]

G. Sziebig, B. Solvang, and P. Koronadi, Image Processing for Next

Generation Robots, Book: Computer Vision (Edited by Xiong Zhihui),
Vienna: I-Tech Education and Publishing, pp.429-440, 2008.
D. A. Mitzias and B. G. Mertzios, A robot vision system for object
recognition and manipulation based on a neural multiclassifier system,
ResearchGate, pp. 1-8, 2015.
J. J. Yokono and T. Poggio, Rotation invariant object recognition from
one training example, Massachusetts Institute of Technology-Computer
Science and Artificial Intelligence Laboratory, Cambridge, 2004.
A. Mohan, C. Papageorgiou, and T. Poggio, Example based object
detection in images by components, IEEE Transactions on Pattern
Analysis and Machine Intelligence, vol. 23, no. 4, pp. 349-361, 2001.
K. Grauman and B. Leibe, Visual Object Recognition, San Rafael,
California (USA): Morgan and Claypool Publishers, 2011.
A. Andreopoulos and J. K. Tsotsos, 50 years of object recognition:
Directions forward, Journal of Computer Vision and Image
Understanding, vol. 117, issue 8, pp. 827-891, 2013.
G. Patil and D. Chaudhari, SIFT based approach: Object recognition
and localization for pick-and-place system, International Journal of
Advanced Research in Computer Science and Software Engineering,
vol. 3, issue 3, pp. 196-201, 2013.
Hema C. R., Paulraj M., R. Nagarajan, and Sazali Yaacob, Bin object
recognition using image matrix decomposition and neural networks,
Iraq Journal Electrical and Electronic Engineering, vol. 3, no.1, pp. 6064, 2007.
C. Ilas, D. Novischi, S. Paturca, and M. Ilas, Real-time image
processing algorithms for object and distances identification in mobile
robot trajectory planning, Journal of Control Engineering and Applied
Informatics, vol. 13, no. 2, pp. 32-37, 2011.

14
Ankit Sharma, Raminder Preet Pal Singh, and Parveen Lehana, Evaluation of the accuracy of genetic algorithms in orientation estimation of
objects in industrial environment, International Journal of Scientific and Technical Advancements, Volume 1, Issue 4, pp. 7-14, 2015.