acoustic signature recognition of moving vehicles … · 183 acoustic signature recognition of...

183

Acoustic Signature Recognition of Moving Vehicles Using Elman Neural Network

1School of Mechatronic Engineering, Universiti

Malaysia Perlis, Perlis, Malaysia.

2Faculty of Engineering, Karpagam University,

Coimbatore, India

Acoustic Signature Recognition of Moving Vehicles Using ElmanNeural Network

Paulraj M P1, Abdul Hamid Adom1, Hema C R2, Sathishkumar Sundararaj1

ABSTRACT

Hearing impaired people cannot distinguish the sound

of moving vehicles approaching them from behind. Since,

it is difficult for hearing impaired to hear and judge sound

information of vehicles, they often encounter risky situ-

ations while they are outdoors. In this paper, a simple

algorithm is proposed to classify the type and distance

of the moving vehicles based on the sound signature. A

simple experimental protocol is designed to record the

vehicle sound under different environment conditions

and also for different speed of the vehicles. The noise

emanated from the moving vehicles along the roadside

is recorded along with the type and distance of the ve-

hicle. Autoregressive modeling algorithm is used to ex-

tract features from the recorded sound signal. Elman

neural network models are developed and trained using

backpropagation algorithm to classify the vehicle type

and its distance. The effectiveness of the network is vali-

dated through simulation.

Keywords: Hearing impaired, Sound signature,

Autoregressive model, Elman neural network.

I. INTRODUCTION

People with hearing impairment are exposed to many

environmental hazards, one such danger is when they

walk on roads and are unaware of the vehicles

approaching them from behind. Acoustic sensors could

provide rehabilitation to such individuals to tackle such

situation. Statistics compiled by the Social Welfare

Department of Malaysia (SWDM), shows that the

number of registered deaf people in Malaysia as of the

year 2006 is 29,522 the actual numbers can be possibly

higher [1]. According to the 2005 estimates of World

Health Organization (WHO), 278 million people have

disability of hearing in both the ears [2].

Several research studies have developed devices to

provide acoustic information through senses of touch

or vision for hearing impaired. In 1973, Frank Saunders,

proposed an electro tactile sound detector using two

microphones and converted the sound into electrical

pulses. The source of sound is localized based on the

difference in the intensity of the pulses [3]. Saunders

et.al developed a tactile aid for the profoundly hearing

impaired children to help understand speech [4]. In 1986,

A. Boothroyd et.al, have developed a wearable tactile

sensory aid which presents a vibratory signal

representative of voice pitch and intonation patterns to

the skin [5].

Acoustic noise signature emanated from moving vehicle

is mainly influenced by the engine vibration and the

friction between the tires and the road. Vehicles of similar

types are known to generate similar noise signature [6].

Nooralahiyan et.al, proposed a vehicle classification

184

Karpagam Jcs Vol. 6 Issue 4 May. - June. 2012

method based on acoustic signature analysis using linear

predictive coding [7]. Eigen method to recognize the

vehicle sound signature based on the frequency vector

principal component analysis has been proposed by Wu

et.al [8]. A review on the state of the art vehicle

classification analysis was done by Xiao et al [9]. Okada,

et.al have computed the direction of the vehicle by

comparing the amplitude of the sound signals captured

by two microphones [10].

In recent years wavelets have been used to identifying

and classifying the vehicle type. Lopez et al proposed

a wavelet-based feature extraction for target

identification [11]. Averbuch et.al, have also used

wavelet packet algorithm for classification of vehicles

and detection of moving vehicles [12, 13]. Using noise

signature to identify moving vehicles has been

investigated by Maciejewski et. al.[14].

Literature reveals that most of the studies on vehicle

noise signature have been restricted to identification of

vehicles only. However the distance of the vehicle also

plays an important role in designing devices for the

hearing impaired. In this study the type of vehicle as

well as the distance of the vehicle is taken into

consideration.

II METHODS

Data Collection

According to the Doppler principle the pitch and the

original frequency of sound changes when a vehicle

passes from the source to the observer. The pitch and

the frequency increases when the sound emitting source

approaches the observer and decreases when it moves

away from the observer [15]. Based on this an experiment

is developed to capture the sound emanated from the

moving vehicle. For this work, a particular section of

the road is considered and the various sound affecting

parameters such as road condition, speed limit,

background noise, and weather condition and wind

direction are also studied. The road selected is a two way

lane with a speed limit of 60 km per hour. The sound

signal from the vehicles moving from X to Y is measured

separately. Two locations A and B with a distance of

separation 100 meters are chosen along the road as shown

in Figure 1. A sound recorder ICD-SX700 is placed at

the location B. which is used to record the noise from

the vehicles. The sound is recorded between A and B.

The corresponding time taken by the vehicle to travel

from A to B is also observed. The number of samples

collected along with the type of vehicle is shown in

Table 1.

Fig. 1 Experimental Recording Setup

TABLE. I Type and number of vehicles

Vehicle type Number of Vehicles

Car 35

Bike 35

Lorry 35

Truck 35

Total 140

Preprocessing and Feature Extraction

The sound signal is recorded at a sampling frequency of

44100 Hz. The normal human auditory system responds

to the frequency ranges from 20 Hz to 20 kHz [16]. The

signal is down sampled to 22050 Hz for further

processing. The time taken by the vehicle to travel the

185


distance of 100 meters is also calculated. The distance

between A and B is divided into four equal distance zones

namely zone 1, zone 2, zone 3 and zone 4 such that each

zone distance is fixed as 25 meters. The four zone

representation is shown in Figure 2. The signal

corresponding to each zone is further segmented into

frames of size 1024. The number of frames

corresponding to a particular zone varies as the speed

of the vehicles is not same. The number of frames in

each zone ranges between 10 to 50.

Fig. 2 Typical representation of zone segmentationof a signal

Autoregressive model is used to extract the data from

the sound signal. A pthorder AR model shown in equation

(1) is used to derive the AR coeffecients [17].

∑=

− +=p

ininin xax

1

ε (1)

where the nx is the thn value which can be predicted

by its previous p successive values:

1−nx , 2−nx , …, pxn − .

ia

(i = 1,2,3, …, p. nε is the ûtting error for nx )

The goal of an AR model is to estimate the AR

coefficients that can ût the original data as much as

possible through an optimization process.

In order to determine the optimal number of consecutive

frames that will produce high classification accuracy, a

simple analysis proposed using the AR features extracted

from the consecutive frames. In order to determine the

type and the position of vehicle, AR coefficients are

extracted as features from the segmented frame signals.

The AR features extracted from the frames are then

associated to the vehicle type and zone position and a

database is created. Further, the features from two

consecutive two frames are combined together and

associated to the vehicle type and associated to the zone

position and the second database is created. In a similar

manner the AR features obtained from 3, 4, 5, 6 and 7

consecutive frames are combined together and five more

databases are created. Thus seven different feature

databases are created by combining consecutive framefeatures.

Classification Results

A dynamic Elman neural network is trained using a back

propagation (BP) training algorithm [18]. The BP training

algorithm involves three stages, the feed forward of the

input training pattern, the calculation and back

propagation of the associated weight error and the weight

adjustments. Using the seven data bases, seven network

models are developed to classify the vehicle type. The

features data are normalized using binary normalization

method so as to rescale the values into a definite range

(0.1 – 0.9). The normalized data are further randomized

and for each data set three different sets of training

samples namely 60 %, 70 % and 80 % of the total samples

are selected. Using the guideline proposed in Master [19],

the numbers of hidden neurons are chosen, for all the

network models, the number of output neurons is fixed

as three. The maximum epoch is fixed as 5000. The

training tolerance was set to 0.05. Hyperbolic tangent

186


sigmoidal activation function was used as an activation

function for the hidden layer activation and logistic

sigmoidal activation function was used for the output

layer neurons.

The network is tested using the testing data samples and

its performance is validated by measuring the

classification accuracy. The network model is trained for

25 times and considered as one trail. Five such trails are

carried out and the average classification accuracy, for

all the network models is shown in Table. II. From Table.

II it is observed that the features obtained from the four

consecutive frames has the highest mean classification

accuracies of 89.48%, 92.01% and 93.52% respectively

for the 60%, 70% and 80% training data sets.

Seven Elman neural network models are developed to

classify the vehicle position. The network models are

trained using gradient descent backpropagation

algorithm. For all the network models, the numbers of

output neurons are fixed as three. The maximum iteration

number was fixed as 5000. The training tolerance was

set to 0.05. Hyperbolic tangent sigmoidal activation

function was used as an activation function for the

hidden layer activation and logistic sigmoidal activation

function was used for the output layer neurons. The

network was tested using the testing data samples and

its performance is validated by measuring the

classification accuracy. The network model is trained

for 25 times and considered as one trail. Five such trails

are carried out and the average classification accuracy,

for all the network models is shown in Table. III, from

Table. III it can be observed that the features obtained

four frames has the highest mean classification accuracy

of 87.34%, 89.955% and 92.871% respectively for the

60%, 70% and 80% training data sets.

TABLE. II Neural Network training results for theclassification of the type of vehicle

Classific ation accuracy

Fra

me

No

. Elman Network

60% 70% 80%

Min Max Mean Min

Max Mean Min

Max Mean

1 83.4 87.6 86.5 86.4 87.8 87.1 86.2 89. 1 87.6

2 83.9 87.9 86.9 85 89.1 88.5 87.8 89. 9 88.8

3 84.3 90.4 87.9 87.9 90.1 90.9 89.4 93. 9 91.7

4 87.8 91.2 89.4 89.9 93.8 92 91.7 95. 4 93.5

5 83.3 89.3 88.3 87.2 90.5 90.9 89.3 92. 7 91.5

6 85.7 89.4 88.9 86.1 91.1 90.6 90.6 92. 9 92.2

7 87.0 90.7 89 86.7 90.9 91.3 90.3 93. 5 92.4

TABLE. III Neural Network training results for theclassification of vehicle position

Classification accuracy

Fra

me

No

Elman Network

60% 70% 80%

Min Max Me an Min Max Me a

n Min Max Mean

1 81 83.8 82.4 83 87.4 85.2 85.7 88.9 87.2

2 81.7 83.6 82.7 83.2 88.6 85.9 86.1 89.9 88

3 84.8 86.4 85.3 86.9 89.9 88.4 89.8 92.5 91.2

4 86 88.6 87.3 89.9 92.9 89.9 91.9 94.8 92.8

5 84.5 86.6 85 86.8 90.6 88.2 90.6 92.5 91.6

6 83.1 85.3 84.2 85.5 89.2 87.4 90.4 93.7 91

7 83.4 85.6 84.5 86.4 89.9 88.1 91.8 93.9 92.4

187


III. CONCLUSION

This paper presents an experimental procedure to capture

the sound signals emanated from the moving vehicle.

Autoregressive model features are extracted from the

recorded signal. Simple neural network models are

developed and trained using backpropagation algorithm

to classify the type of vehicle and its distance from the

subject. The training results show that the use of

autoregressive modeling features differs from the frames

contributes towards a better classification of the vehicle

type and its distance from the subject.

ACKNOWLEDGEMENT

The authors would like to acknowledge the support and

encouragement by the Vice Chancellor of Universiti

Malaysia Perlis, Y.Bhg. Kol.Prof Dato’ Dr. Khamarudin

b. Hussin. This work is financially assisted by the

Fundamental Research Grant Scheme (FRGS) (9003-

00186): by the Ministry of Higher Education, Malaysia

REFERENCES

[1]. O.S.Hock, Salimah Mokhtar, and Roziati

Zainuddin, (2007) “A Review on the Teaching and

Learning Resources for the Deaf Community in

Malaysia”. Chiang Mai University Journal of

Natural Sciences, 1 (1). pp. 1-12. 2007.

[2]. S.Garg, S. Chadha, S. Malhotra and A.K. Agarwal,

“Deafness: Burden, prevention and control in

India” in The National Medical Journal of

India, Vol. 22, No. 2, pp. 79-81, 2009.

[3]. F.A. Saunders, “An Electrotactile sound detector

for the deaf”, IEEE Transactions on Audio and

Electroacoustics, Vol, AU-21, No 3, June, 1973.

[4]. F. A. Saunders “A Wearable Tactile Sensory Aid

for Profoundly Deaf Children”, Journal of Medical

Systems, Vol 5., No 4., 1981.

[5]. Arthur Boothroyd, “Wearable Tactile Sensory Aid

Providing Information on voice Pitch and

Intonation Patterns” Research Corporation, New

York. 1986.

[6]. R. B. Randall, “Frequency Analysis,” Brüel &

Kjær, 1987.

[7]. Amir Y.N., Mark D., Denis McKeown and Howard

R.K, “A field trial of Acoustic Signature Analysis

for Vehicle Classification”, Vol 5, No 3/4, pp 165-

177, Elsevier Science Ltd. 1997.

[8]. W. Huadong, M. Siegel, and P. Khosla, “Vehicle

sound signature recognition by frequency vector

principal component analysis” in Instrumentation

and Measurement Technology Conference, 1998.

IMTC/98. Conference Proceedings. IEEE, 1998,

pp. 429-434 vol.1.

[9]. Hanguang Xiao, Congzhong Cai, Qianfei Yuan,

Xinghua Liu, Yufeng Wen, “A Comparative Study

of feature Extraction and Classification Methods

for Military Vehicle Type Recognition Using

Acoustic and Seismic Signals”, Department of

Applies Physics, Chongqing University and

Chongqing Institute of Technology, Chongqing,

China.

[10]. Kazuhide Okada, Gwan Kim and Pyong Sik Pak,

“Sound Information Notification System by Two-

Channel Electrotactile Stimulation for Hearing

Impaired Persons”, Proceedings of the 29th Annual

International, Conference of the IEEE EMBS, Cite

Internationale Lyon, France.

[11]. Joes E.Lopez, Hung Han Chen and Jennifer

Saulnier, “Target Identitfication Using Wavelet-

based Feature Extraction and Neural Network

Classifiers” Cytel Systems,inc. Hudson.

188


[12]. Averbuch, E. Hulata, V. Zheludev and I. Kozlov,

“A Wavelet Packet Algorithm for Classification

and Detection of Moving Vehicles”,

Multidimensional Systems and Signal Processing,

vol. 12, pp. 9-31, 2001

[13]. Averbuch, V. A. Zheludev, N. Rabin, and A.

Schclar, “Wavelet-based acoustic detection of

moving vehicles”, Multidimensional Systems

and Signal Processing, vol. 20, pp. 55-80, 2009.

[14]. H. Maciejewski, J. Mazurkiewicz, K. Skowron,

and T. Walkowiak, “Neural Networks for Vehicle

Recognition,” in Proceeding of the 6th International

Conference on Microelectronics for Neural

Networks, Evolutionary and Fuzzy Systems, 1997,

p. 5.

[15]. Dev Maulik, “Doppler Ultrasound in obstetrics and

Gynecology”, pp 13, ISBN 3-540-23088-2,

Springer Verlag Berlin Heidelberg Newyork.

[16]. Weixiang zhao, Cristina E.Davis, “Autoregressive

model based feature extraction method for time

shifted chromatography data” Elsevier, chemo

metrics and intelligent laboratory systems96 (2009)

252 – 257.

[17]. Bhanu Prasad, S.R.M. Prasanna, “Speech, Audio,

Image and Biomedical Signal Processing using

Neural Networks”, Studies in Computational

Intelligence, vol. 83, pp. 169-188, ISBN 978-3-540-

75397-1, Springer 2008.

[18]. S.N Sivanandam and Paulraj M, “An Introduction

to Artificial Neural Networks”, Vikhas Publication,

India, 2003.

[19]. T. Masters, “Practical Neural Network Recipes in

C++”, Academic Press, New York 1993.

Author’s Biography

Paulraj MP received his BE in Electrical

and Electronics Engineering from

Madras University (1983), Master of

Engineering in Computer Science and

Engineering (1991) as well as Ph.D. in Computer

Science from Bharathiyar University (2001), India. He

is currently working as an Associate Professor in the

School of Mechatronic Engineering, University Malaysia

Perlis, and Malaysia. His research interests include

Principle, Analysis and Design of Intelligent Learning

Algorithms, Brain Machine Interfacing, Dynamic Human

Movement Analysis, Fuzzy Systems, and Acoustic

Applications. He has co-authored a book on neural

networks and 290 contributions in international journals

and conference papers. He is a member of IEEE, member

of the Institute of Engineers (India), member of Computer

Society of India and a life member in the System Society

of India.

Abdul Hamid Bin Adom is currently

the Dean of School of Mechatronic

Engineering at University Malaysia

Perlis, Malaysia. He received his B.E,

MSc and PhD from LJMU, UK. His

research interests include Neural Networks, System

Modeling and Control, System Identification, Electronic

Nose/ Tongue, Mobile Robots. He holds various research

grants and published several research papers. Currently

his research interests have ventured into Mobile Robot

development and applications, as well as Human

Mimicking Electronic Sensor Systems for agricultural

and environmental applications.

189


Hema C R obtained her BE and MS in

EEE from Madurai Kamaraj

University, India and University

Malaysia Sabah, Malaysia in 1989 and

2005 respectively. She obtained her

PhD in Mechatronic Engineering at University Malaysia

Perlis, Malaysia in 2010. She is currently the Dean

Engineering Research at Karpagam University, India. Her

research interests include EEG signal processing, Neural

Networks and Machine Vision. She holds many research

grants and has published 8 books and 5 book chapters

and around 108 papers in referred Journals and

International Conferences.. She has received gold and

Bronze medals in National and International exhibitions

for her research products on vision and brain machine

interfaces .She is cited in WHO IS WHO in the world

2009 to 2011. She is a member the IEEE, IEEE EMB

Society and IEEE WIE Society.

Sathishkumar Sundararaj obtained his

B.Tech in Information Technology from

Anna University (2008). He is currently

pursuing Masters Degree in Mechatronic

engineering at University Malaysia Perlis. His research

interest includes signal processing and Neural Networks.

He published some papers in conferences

190


Fuzzy MPPT Based Voltage Regulation on Photovoltaic PowerSupply System For Continuously Varying Illumination Condition

A. Durgadevi1, S. Arulselvi2

ABSTRACT

Due to scarcity of fossil fuel and increasing demand of

power supply we are forced to utilize the renewable

energy resources with the shortage of energy and ever

increasing of the oil price. Researches on the renewable

and green energy sources, especially solar arrays and

fuel cells become more and more important. Considering

easy availability and vast potential, world has turned to

solar photovoltaic energy to meet its ever increasing

energy demand. Due to high initial investment on PV

systems and non linearity of PV cell output

characteristics counteract its wide commercialization.

The PV array has an optimum operating point to generate

maximum power at some particular point called

maximum power point (MPP). To track this maximum

power point and to draw maximum power from PV

arrays, MPPT controller is required in a stand-alone PV

system. Due to the nonlinearity in the output

characteristics of PV array, it is very much essential to

track the MPPT of the PV array for varying maximum

power point due to the insolation variation. In order to

track the MPPT intelligent controller like fuzzy logic

controller is proposed and simulated. The output of the

controller, pulse generated from PWM can switch

MOSFET to change the duty cycle of buck DC-DC

converter. The result reveals that the maximum power

point is tracked satisfactorily for varying insolation

condition.

Key words:

Photovoltaic, Pulse Width Modulation, Fuzzy Logic

Controller.

I. INTRODUCTION

Today photovoltaic (PV) systems are becoming more

and more popular with increase of energy demand and

there is also a great environmental pollution around the

world due to fossils and oxides. Solar energy which is

free and abundant in most parts of world has proven to

be economical source of energy in many applications

[1]. The energy that the earth receives from the sun is so

enormous and so lasting that the total energy consumed

annually by the entire world is supplied in as short a

time as half an hour. The sun is a clean and renewable

energy source, which produces neither green house effect

gas nor toxic waste through its utilization. It can

withstand severe weather conditions, including cloudy

weather. The watt peak price is decreased since the

seventies, this leads to large scale promising areas. It

does not have any moving parts and no materials

consumed or emitted. Unfortunately, this system has two

major disadvantages, which the low conversion

efficiency of electric power generation (9 to 16%),

especially under low irradiation conditions and the

amount of electric power generated by solar array

changes continuously with the weather conditions like

irradiation and temperature. To overcome this problem,

maximum power point tracking (MPPT) technique will

be used.

Department of Electronics and Instrumentation

Engineering Annamalai University, Chidambaram,

Tamil Nadu, India.

Email: [email protected] and

[email protected]

191

Fuzzy MPPT Based Voltage Regulation on Photovoltaic

Power Supply System For Continuously Varying Illumination Condition

In order to regulate the converter due to non linearity

artificial intelligent controller like fuzzy logic controller

is proposed and simulated. The tracking algorithm

integrated with a solar PV system has been simulated

with buck DC-DC converter for the application of

battery charging in stand - alone PV system. The proposed

MPPT system with buck DC-DC converter is shown in

Fig.1.

Fig.1. Photovoltaic module with DC-DC buck converter.

II. PHOTOVOLTAIC CELL MODELING

The proposed MPPT is based on the behaviour of the

photovoltaic array by means of temperature and

irradiation variation [2]. This the mathematical model of

PV array is implemented in the form of current source

controlled by voltage, sensible to two impact parameters,

that is, temperature (°C) and solar irradiation power (w/

m2).

An equivalent simplified electric circuit of a photovoltaic

cell presented in Fig.2.

Fig.2. Equivalent circuit of photovoltaic cell.

The expressions obtained from fig.2. are given below.

The load current IL is obtained is given in equation (1)

as,

(1)

IL is the photo electric current related to the given

irradiation condition given by equation (2),

(2)

The diode saturation current (Io) is given by the equation

(3),

(3)

where - ID is the diode current; I-

L- is the photoelectric

current related to a given condition of radiation and

temperature; V is the output voltage [V]; Io is the

saturation diode current [A]; is the form factor which

represents an index of the cell failing; Rs is the series

resistance of the cell [Ù]; q is the electric charge

(1.602*10-19C); k is the Boltzmann’s constant

(1.381*10-23K); Tc is the module temperature [K]. E

g is

the energy gap of the material with which the cell is made

(for the silicon it is 1.12 eV); G is the radiation [W/m2];

GREF is the radiation under standard conditions [W/m2]

IL,REF

is the photoelectric current under standard

conditions [A]; TC,REF

is the module temperature under

standard conditions [K]; is the temperature

coefficient of the short circuit current [A/K], given by

the manufacturer according to CEI EN 60891 standard

[3-4].

Fig. 3. shows the simulated P-V characteristics for

varying irradiation and temperature in MATLAB/

SIMULINK environment.

(a) (b)

Fig.3. Simulated waveforms showing the effect of (a) radiation and

(b) temperature on P-V characteristics.

192


It can be observed from simulated results as shown in

Fig. 3(a), the photo current is directly proportional to

irradiation. It is noted from Fig. 3(b) that the terminal

voltage increases with decreasing temperature.

The manufacturers data at standard conditions are

given as Pmax

= 80W, Imax

= 4.515 A and Vmax

= 21.6V.

The simulation results obtained were: Pmax

= 78.51W,

Imax

= 4.515 A and Vmax

= 21.65V. It is seen that the

simulation model showed excellent correspondence to

manufacturer’s data and therefore this model was

considered sufficient for the purpose of further study

[4-8].

Simulated I-V, P-V characteristics for the maximum

power point tracking (MPPT) is shown in figure.4.

At this Maximum Power Point (MPP), the solar array is

matched to its load and when operated at this point the

array will yield the maximum power output. From Fig. 4

(a) & (b), it is observed that the power output has an

almost linear relationship with array voltage unit, hence

the MPP is attained. Any further increase in voltage

results in power reduction [5].

(a) (b)

Fig.4. PV array simulated curves (a) I-V curve (25°C) and (b) P-Vcurve (1000w/m2).

III. DESIGNING OF BUCK CONVERTER

A. Circuit diagram of buck converter.

Fig. (5) shows the schematic diagram of buck

converter with varying irradiation, which consists of DC

supply voltage Vs, as PV generator controlled switch S,

diode D, buck inductor L, filter capacitor C and load

resistance R. The current and voltage waveforms of the

converter in CCM are presented in fig.6.

Fig.5. Circuit diagram of buck converter with PV

module.

It can be seen from the circuit that when the switch S is

commanded to the on state, the Diode D is reverse biased.

When the switch S is off, the diode conducts to support

an uninterrupted current in the inductor through the

output RC circuit using faradays law for the buck

inductor as given in (4)

Fig.6. Theoretical voltage and current waveforms of buck converter.

(4)

The DC voltage transfer function turns out to be,

193



(5)

The buck converter operates in the CCM for L>Lb. The

calculated value of inductance L=15µH. To limit the

ripples in the output side, larger filter capacitor is

required. The filter capacitor must provide the output dc

current to the load when diode D is off. The minimum

value of filter capacitance calculated that results in the

voltage ripple Vr is given by C

min =12.675uF.

Thus the buck converter is designed in the open loop for

the supply voltage of 21.7V DC, which is generated by

the Photovoltaic panel for 1000w/m2 and 25°C. Fig.7

shows the simulated voltage and current waveforms of

buck converter. It is seen that these waveforms are agreed

closely with theoretical waveforms as shown in fig.6.

Fig.7. Simulated waveforms showing the voltage and current ofbuck converter.

IV. MPPT WITH INTELLIGENT FUZZY METHOD

The block diagram of closed loop simulation with fuzzy

logic controller is shown in fig.8. To regulate the output

voltage Vo, the switching frequency of the PWM pulses

are varied depends on error.

Fig. 8. Block diagram of the proposed fuzzy based MPPT scheme.

A. MPPT with the Intelligent Fuzzy Method

To track and extract maximum power from the PV

arrays for a varying insolation level and at a given cell

temperature, a novel fuzzy logic controller is proposed

based on the intelligent fuzzy logic method.

It consists of three parts: fuzzification, inference

engine and defuzzification [9-11].

An error function (E) and change of error (ÄE) are

calculated from (6) and (7) and created during

fuzzificaion. These variables are then compared to a set

of pre-designed values during inference method in order

to determine the appropriate response.

Defuzzification is for converting the fuzzy subset of

control from inference back to crisp values [12-15].

(6)

(7)

The E and ÄE function is compared to the graph of Fig.

9. to obtain a variable NB or ZE, then this parameter

will be used to locate the respective output function (u)

from the fuzzy rule table I.

The inference mechanism is implemented with mamdani

algorithm and the centre of gravity method is used for

defuzzification.

194


Fig No. 9, Fuzzy Logic membership function

TABLE IFUZZY RULE TABLE

E

NB NM NS Z PS PM PB

NB Z NM NS NB NS PB PB

NM NM Z Z NB PM Z NB NS PM NB Z NS Z PM NS

Z NB NB Z Z Z Z PS PS NM PM NS Z Z NB Z

PM PM NS NS NB Z Z NB PB NB PB PM Z NS NS Z

The insolation variation from 100 w/m2 to 1000 w/m2,input voltage (V-

in) and output voltage (V

o) are shown

in Fig.10.

Fig.10. Simulation results depicting the change of insolation from100 to 1000w/m2, input voltage (V

in) and output voltage (V

o).

V. CONCLUSION

An intelligent fuzzy method for MPPT of Photovoltaic

array is presented in this paper on the base of fuzzy logic

control algorithm. The simulation results show that the

fuzzy controller has the merits such as simplicity fast

response, low over-tuning, high control, precision and

easy implementation.

REFERENCES

[1] Bimal K. Bose, “Global Warming Energy:

Environmental Pollution and the Impact of Power

Electronics”, IEEE Industrial Electronics

Magazine, pp.1-17, March 2010.

[2] Adel Mellit, “Artificial Intelligence technique for

modelling and forecasting of solar radiation data:

a review,” Int. J. Artificial Intelligence and Soft

Computing Vol. 1. No. 1, 2008.

[3] Akihiro Oi, “Design and Simulation of Photovoltaic

Water Pumping System”, Master of Science Thesis,

California Polytechnic State University, 2005.

195



[4] Adedamola Omole, “Analysis, Modelling and

Simulation of Optimal Power Tracking of Multiple-

Modules of Paralleled Solar Cell Systems”, Master

of Science Thesis, The Florida State University

College of Engineering, 2006.

[5] V. Di Dio, D. La Cascia, R. Miceli, “A Mathematical

Model to Determine the Electrical Energy

Production in Photovoltaic Fields under Mismatch

Effect”, Proceedings of the 978(1) IEEE, pp.46-51,

August 2009.

[6] Renji V Chacko, Z.V. Lakaprrampil, K.A. Fathima,

“Solar Inverter with MPP Tracking for Pumps and

UPS Application”, ERDC, pp.1-12, 2000.

[7] Christopher A. Otieno, George N. Nyakoe, Cyrus

W. Wekesa, “A Neural Fuzzy Based Maximum

Power Point Tracker for a Photovoltaic System”,

IEEE AFRICON, September 23-25, pp.1-6, 2009.

[8] A. Mellit, “An ANFIS-Based Prediction for Monthly

Clearness Index and Daily Solar Radiation:

Application for sizing of a Stand-Alone Photovoltaic

System”, Journal of Physical Science, Vol. 18(2),

pp.15–35, 2007.

[9] N. Patcharaprakiti and S. Premrudeepreechacharn,

“Maximum PowerPoint Tracking Using Adaptive

Fuzzy Logic Control for Gridconnected

Photovoltaic System”, PESW2002, volume 1,

PP:372-377, 002.

[10]Marcelo Gradella Villalva, Jonas Rafael Gazoli, and

Ernesto Ruppert Filho , “Comprehensive Approach

to Modeling and Simulation of Photovoltaic Arrays”,

IEEE Transactions on Power Electronics, vol 24,

no. 5, 2009, pp 1198-1208.

[11] A. Durgadevi, “Design of intelligent controller for

quasi resonant converters”, M.E. dissertation,

Departmne of Electronics and Instrumentation

Engineering, Annamalai Univ., Chidambaram, 2008.

[12]C. Y. Won, D. H. Kim, et al, “A New Maximum

Power Point Tracker of Photovoltaic Arrays Using

Fuzzy Controller”, PESC’1994 Record, PP:396-403,

1994.

[13]H. Yamashita and K. Tamahashi, et al, “A Novel

Simulation Technique of the PV Generation System

Using Real Weather Conditions,’’ in Proceedings of

the PCC,2002, volume2, PP:839-844, 2002

[14]Cung-Yuen Won, Duk-Heon Kim, Sei-chan Kim,

Won-Sam Kim, hack_Sung Kim, “A New Maximum

Power Point Tracker of Photovoltaic Arrays Using

Fuzzy Controller”, PESC’94.

[15]Guohui Zeng, Qizhong Liu, “An Intelligent Fuzzy

Method for MPPT of Photovoltaic arrays”, 2009

Second International Symposium on Computational

Intelligence and Design, pp. 356-359.


A. Durgadevi received her B.E. in Elec-

trical and Electronics Engineering from

I.FE.T engineering college, villupuram

Madras University in 1999 and M.E. in

Process Control and Instrumentation En-

gineering from Annamalai University in 2008. Currently

she is a Doctorate student at department of Electronics

and Instrumentation Engineering in Annamalai Univer-

sity. Her research interests are in Solar Energy Systems

and Power Electronics.

196


Design and Development of a Gripper System for an IndoorService Robot

Hema C. R.1 Vivian Tang Sui Lot2 Paulraj M.P.2

ABSTRACT

Service robot are designed to provide assistance in home

environments, floor cleaning, material transporting and

security purpose. This paper focuses on the design and

development of a gripper system for a housekeeping

robot ‘ROOMBOT’ to perform pick and place operations.

A gripper with three degree of freedom is designed and

developed and tested for pick and place of several objects.

Keywords : Roliot Geipprens, Robot Coutiol, Dinner

Degree of Freedom.

I. INTRODUCTION

Robotic grippers find wide applications in industrial,

medical and domestic use. Research in this area has been

continuously evolving to provide more dexterous

grippers which can mimic the human arm. Despite a

number of prototypes of robotic gripper that have been

developed in more than 20 years of research mainly in

academic institutions among many others [1], effort is

being devoted to seek and evaluate alternative solutions

with sufficient dexterity to perform in any case non trivial

operations on wide range of objects.

A gripper is a device which enables the holding of an

object to be manipulated. The easier way to describe a

gripper is to think of the human hand. Just like a hand, a

1Faculty of Engineering, Karpagam University,Coimbatore, INDIA2School of Mechatronic Engineering, UniversityMalaysia Perlis, Perlis, MALAYSIAhemacr@ karpagam.ac.in

gripper enables holding, tightening, handling and

releasing of an object. Gripper is just one component of

an automated system. A gripper can be attached to a robot

or it can be part of a fixed automation system.

The basic operating principal of the gripper consists of

three primary motions which is parallel, angular and

toggle [2]. These operating principals refer to the motion

of the gripper jaws in relation to the gripper body (refer

Table 1). Rotary actuator is a device use to alternate the

rotated position of an object. Just like the human wrist

the actuator enables the rotation of an object, except that

rotary actuators are available in a wide variety of models

with different sizes, torques, and rotation angles.

The robot gripper can be divided into two sections, each

with different function:

1. Arm and body – the arm and body of a robot are

used to move and position parts or tools within a

work envelope.

2. Wrist or end effectors – these parts are used to orient

the parts or tools at the work location.

II. DESIGN OF THE GRIPPER

The ROOMBOT robot is a housekeeping robot which

was required to collect crushed paper in a controlled

environment.

Identification of the object to be picked is accomplished

by a hybrid vision system. Co-ordinates of the object to

197

Design and Development of a Gripper System for an Indoor Service Robot

be picked will be provided by this system to the gripper.

The mechanical design of the gripper is required to fulfill

the interaction between the robot and the object in order

to pick and place i.e. the object from the ground to the

dustbin at the back of the robot. The initial target of object

is identified as crushed paper. The movement mechanism

of the gripper is simple and straightforward. It will react

once the input is received from the vision system from

the main controller.

Table 1 – Gripper Operating Principal

Operating Principal

Illustration

Parallel

Angular

Toggle

Design Specification

The mechanical design of the gripper is inspired by the

following issues [3-5, 7-13]:

a. The gripper should be able to manipulate object in

round shape. The primary object to be grasped is a

crash paper. The surface of a crash paper is rough

and its shape being almost like round shape. The

size targeted for the crash paper is 3cm to 8cm in

diameter.

b. Grasping area is 25cm in front of the robot. The

gripper should be able to extend to 25cm distance

from the robot front body.

c. Grasping precision and position repeatability of the

gripper is a requirement as memory of location is

preset in controller.

d. Power consumption by the actuator should be small

considering larger supply in need will increase the

burden of the robot weight by installing larger battery

cell. The strength of grasping should be enough to

lift up a crash paper and hold it when transferring.

Mechanical Design

From the required design specification, the gripper is

constructed first by using the Solid Works software to

draft out the architecture of the whole gripper system.

There are three parts in construction which is the base,

the arm and the end effectors.

The base of the gripper system is constructed using L-

shape aluminum bar with dimension of 9x8cm. One

motor is use as the base driving actuator. The height of

the gripper base is extended 3cm from ground by using

cylinder screw. Figure 1 shows a drawing illustration of

the base by using the Solid Works software.

Figure 1 - Gripper base design drawing inSolid Works

198


The arm have two joints linking the arm and the wrist to

the end effectors, labeled A and B in figure 2. Two motor

is use to drive the first arm joint A, while the wrist only

uses one of the standard servo motor for its drive. Two

motor is use in for supporting and driving the first arm

linked to the base.

Figure 2 - Gripper arm and wrist design drawing inSolid Works

The design of the end effectors is using the pivoting or

rotating jaws mechanism. This design is simple and only

requires one actuator to drive the mechanism (refer figure

3). This design of mechanism requires sufficient accuracy

in center of grasping of the object. Otherwise, the object

may slip off when the gripper is closing the end effectors

or the object may drop off from the end effectors in the

middle of transferring the object to the dustbin.

Figure 3 - End effectors design drawing in SolidWorks

Gripper Degrees of Freedom

To locate an object in space, one need to specify the

location of the selected object, thus it requires three

pieces of information to be located as desired. These three

pieces of information are the X, Y, and Z coordinates

value [4]. With the three pieces of the information, the

object location is specified. Hence, there is need for a

total of six pieces of information to fully specify the

location and orientation of the gripper.

The designed gripper only consists of 3 degrees of

freedom, where it can only move along the X, and Y

axes (refer figure 4). In this case, no orientation can be

specified. All the robot can do is to pick up the object

and move it to the dustbin. The orientation is always

remains the same.

The end effectors are not considered as one of the degrees

of freedom. All the gripper has this additional capability,

which may appear to be similar to a degree of freedom.

However, none of the movements in the end effectors

are counted towards the gripper’s degrees of freedom.

The weight limitation was important in construction of

the gripper. The material used should have low density

and good yield in strength properties and are easy to

machine. Aluminum was choose as it criteria fits in.

Aluminum is soft, durable, lightweight. It is ductile, and

easily machined, cast and extruded. Hence, aluminum

provides easy machinery and shaping with the tools, and

allowing a high accuracy in part dimensions to be

obtained. The aluminum sheet use has thickness of 2mm.

This concludes that the whole gripper weight after the

construction is 0.5kg.

199


Gripper Actuation

Parallax standard servo motor is selected as the gripper

actuator because of its precision in positioning, and its

weight is 45grams. The servo motor can produce torque

up to 3.4kg-cm and it have small size casing with low

price compare other similar servo motor that offer

equivalent strength. One standard motor is use to drive

the rotation of the base. Standard servo motor has

limitation in rotation up to 180 degrees [6]. In that case,

the base of the gripper only can turn in one side of area

at 0 to 180 degrees. This rotation is sufficient to cover

the area of object that could be detected in front of the

robot.

Two motors are used to drive the first arm joint with the

base. The weight induced over the end effectors are

insufficient to be lifted up by only one motor torque. To

avoid over current draw by the motor in producing

sufficient torque, two motors are applied. The two motor

sequences should be synchronized.

Gripper end effectors are driven by only one standard

motor (figure 3). Two gears is used in series, and one of

the gears is attached to the motor. When the motor rotate,

the gear attach to the motor will rotate and drive the

second gear in series with it to rotate in opposite direction.

This effect will operate the end effectors to open wide

up reaching to 180 degrees.

Figure 4 - Gripper design drawing – Isometric view

� ��

��

��

Drive and Control System

The gripper control program is written in assembly

language, and running in Basic Stamp controller. The

gripper program is executed together with mobile

platform and ultrasonic sensor program in one processor.

Another controller used was the servo controller. Servo

controller can control up to 16 motor at one time. Servo

controller communicates with Basic Stamp controller

through serial cable. Signal is sent from Basic Stamp

controller to the servo controller in pulse form.

The main features of the flow for the gripper program

are shown in Figure 5. When the robot is powered up,

the gripper will move to its standby position which is

heading toward the back of the robot. When the main

program is executed, the robot will move around with

ultrasonic sensor as primary sensor to detect any object

appears. When an object is been encounter, the vision

system will activate to capture and recognize the

encountered object. If an object is identified, the vision

system will send input through the serial communication

between the mobile platform main controllers and initiate

the gripper system.

When the gripper is initiated, the base motor will rotate

the body arm of the gripper towards the front side of the

robot. Followed, the controller will instruct the first arm

joint two motor to lift down the end effectors to a set

height. At the set height the end effectors will open to

180 degrees wide. Again the gripper wrist will lower

down bringing the end effectors closer to the object and

perform the picking operation.

When the object is secure in the end effectors, the gripper

is instructed by the controller to returns and place the

object into the bin at the back of the robot before it

reposition back to its standby position. After the operation

of pick and place is completed, the robot will continue

to move around in the environment.

200


Figure 5 - Program Flow Chart

III. EXPERIMENTS AND RESULTS

There were two types of experiments is done towards

the designed gripper system to test it reliability and

efficiency, the offline and online experiments. Offline

testing is conducted on the gripper simulation without

installation it onto the robot mobile platform. This is to

test on the gripper functionality and define the coverage

area of the gripper in picking object. A model of mobile

platform is build to install the gripper system to perform

the testing.

Online testing is conducted when the gripper system

is assembled onto the mobile platform together with

ultrasonic system and vision system. This testing

conducted to test on the serial communication between

the controller of PC and Basic Stamp which sending

signal to trigger the gripper system when an object is

been identified. Another testing is tested on the relocation

of the Y-position when there is an input of Y-value from

the vision system.

Discussion

From the testing, the result shows the gripper system is

fully operate able. The average time for one cycle of pick

and place operation is 12 seconds (Table 2). Once the

gripper is triggered by the main controller, the gripper

can operate to complete the task in that particular short

of time period. This result shows the efficiency of the

gripper in decreasing the operating time as less delay is

created due to the pick and place action.

The gripper system can pick able to vary object. 5 type

of objects is giving a testing which are the plastic bag,

crash paper, empty bottle (500ml), carbonate drink can

(325ml) and a small boxes. 4 of the item show pick able

by the gripper. Empty bottle is unable to be picked

because of its plastic body surfaces that give frictionless

to the grasping force from the end effectors.

The gripper has a limitation in picking object with weight

greater than 85grams. Object that weight over 85grams

make the gripper to draw more current to produce more

torque in lifting up the object and transferring it to the

dustbin.

The communication between the Basic Stamp and the

servo controller is via RS 232 serial protocol. The

firmware number which replies by the servo controller

when communication is established shows the version

of the servo controller. After the confirmation of the

firmware version, the baud rate is elevated to 38K4 baud.

Now the servo motor can be control by entering the

position command. Each position command is comprised

of a heater, three parameters: C, R, PW, and a command

terminator. The C parameter is a binary number 0-31

corresponding to the servo channel number. The R

parameter is a binary number 0-63 that controls the ramp

function for each channel. The PW parameter is the 16

bit Word that corresponds to the desired servo position.

201


When the robot detects an object, the distance for the

robot to stop is 25cm away from the object regardless of

which location of the object maybe laying in the Y-axis.

The possible area of object fall in is coverage by the

wide open angle of the gripper end effectors.

Coordination of Y-position happen when the object fall

in the camera coverage area where servoing is unable to

perform to adjust the robot position to change the location

of the object to the center in front of the robot.

Referring to Figure 6, the location Y1, Y2, and Y3 is

pre-taught to the gripper. These three locations are set

on the gripper base servo motor. Once there is an input

from the vision system signaling the location of the object

through the serial com, the gripper is activated and based

on the data received from the vision system, the program

command line will call to the desired preset location of

the Y1, Y2, or Y3 and turn the gripper towards the

location and perform pick and place.

IV. CONCLUSION

The gripper is designed with 3 degrees of freedom, build

using light weight aluminum with total weight of 0.5kg,

and drive by 5 standard servo motor. The gripper

operation time is 12 seconds with pick able of vary object

of plastic bag, small boxes, carbonate drink can and

crashed paper.

The gripper also can manipulate in Y-position with the

input from the vision system.

In conclusion to this project, the primary objective is

achieved successfully. The crash paper selected as initial

target of object is successfully picked able by the gripper

in the range of 25cm distance apart from the front of the

robot.

There still have a lot of improvement area in the gripper

system that can be done to make it more reliable and

effective. Any further improvement and upgrade on the

version can still be carry on to perfected the system and

commercialize for public usage.

�

��

��

��

��

��

��

��

��

��

Figure 6 - Y-positioning of the gripper relative to thelocation of the objects.

Table 2 – Time taken for one completed cycle of pickand place

Trials Time Taken (s)

1 12.64

2 12.76

3 12.60

Average 12.66

There are many design improvements that can be done

in the proposed gripper system. The standard motor can

be replaced by more better and powerful actuator to

increase the torque and more degrees of freedom. A

stepper motor can provide higher torque and 360 degrees

of angle rotation compare with servo motor. So as, the

area coverage by the gripper will extended more than

just 25cm in radius.

The end effectors function can be improved by adding a

degree of subsystem flexibility like variable jaw geometry

202


featuring wide range gripper end effectors encompassing

a board spectrum of object dimensions. Additional force

sensor can be implemented on the end effectors to control

the grasping force applied. It also can prevent the two

finger of the gripper to over strain the object picked.

REFERENCES

[1] L. Biagiotti, C. Melchiorri, G. Vassura 2003. A

Dexterous Robotic Gripper for Autonomous

Grasping. Research Paper Volume 30, University

of Bologna, Bologna, Italy.

[2] Applied Robotics: Solutions in reach, April 1,

2008. Tutorials on Grippers and Rotary Actuators

http://www.arobotics.com/technical/tutorials.aspx.

[3] Zaytran, Inc. 1996. Robotic Gripper Sizing: The

Science, Technology and Lore. Elyria, Ohio USA.

http://www.grippers.com/size.htm

[4] Saeed B. Niku, 2001. Introduction to Robotics

Analysis, Systems, Applications. 1st Edition,

Prentice Hall, Upper Saddle River, New Jersey.

[5] B. S. Baker, S. Fortune and E. Groose, 1995. Stable

Prehension with a Multi-Fingered Hand. IEEE

International Conference on Robotics and

Automation, pp. 570-575.

[6] Seattle Robotics Society, March 16, 2009. Whats

a Servo: A Quick Tutorial, available online at http:/

/www.seattlerobotics.org/guide/servos.html.

[7] Gareth J. Monkman, Stefan Hesse, Ralf

Steinmann, Henrik Schumk, 2007. Robot

Grippers. 1st Edition, WILEY-VCH Verlag GmbH

& Co,KGaA, Weinheim.

[8] Thomas R. Kurfess, 2005. Robotic and

Automation Handbook. 1st Edition, CRC PRESS,

USA.

[9] Seguna, C.M., Saliba, M.A. The Mechanical and

Control System Design of a Dexterous Robotic

Gripper. IEEE, pp. 1195-1201.

[10] Andres Herrera, Andres Bernal, David Isaza,

Malek Adjouadi. Design of an Electrical Prosthetic

Gripper using EMG and Linear Motion Approach.

Center for Advance Technology and Education,

Department of Electrical and Computer

Engineering Florida International University,

Florida.

http://fcrar.ucf.edu/papers/tp2_ahabdima_fiu.pdf

[11] Chen Guoliang, Huang Xinhan, 2004. Research

on Vacuum Micro-Gripper of Intelligent

Micromanipulation Robots. Proceeding of the

2004 IEEE International Conference on Robotics

and Biomimetics, pp 279-283.

[12] Dominik Henrich and Heinz Wörn. 2000. Robot

Manupulation of Deformable Objects. 1st Edition,

Springer-Verlag London Limited, Great Britain.

[13] K. Venugopal Varma and U. Tasch, 1991.

Prehension Mode Constraints for a Robot Gripper.

Proceedings of the IEEE International Conference

on Systems, Man and Cybernetics, vol.3, pp. 1441-

1446.

203



Hema C R obtained her BE and MS

in EEE from Madurai Kamaraj Uni-

versity, India and University Malay-

sia Sabah, Malaysia in 1989 and 2005

respectively. She obtained her PhD

in Mechatronic Engineering at Uni-

versity Malaysia Perlis, Malaysia in

2010. She is currently the Dean Engineering Research

at Karpagam University, India. Her research interests

include EEG signal processing, Neural Networks and

Machine Vision. She holds many research grants and has

published 8 books and 5 book chapters and around 108

papers in referred Journals and International Confer-

ences.. She has received gold and Bronze medals in

National and International exhibitions for her research

products on vision and brain machine interfaces .She is

cited in WHO IS WHO in the world 2009 to 2011. She

is a member the IEEE, IEEE EMB Society and IEEE

WIE Society.

Vivian Tang Sui Lot is an

undergraduate student of the School

of Mechatronics Engineering,

University Malaysia Perlis, Malaysia.

Paulraj MP received his BE in

Electrical and Electronics

Engineering from Madras University (1983), Master of

Engineering in Computer Science and Engineering

(1991) as well as Ph.D. in Computer Science from

Bharathiyar University (2001), India. He is currently

working as an Associate Professor in the School of

Mechatronic Engineering, University Malaysia Perlis,

and Malaysia. His research interests include Principle,

Analysis and Design of Intelligent Learning Algorithms,

Brain Machine Interfacing, Dynamic Human Movement

Analysis, Fuzzy Systems, and Acoustic Applications. He

has co-authored a book on neural networks and 290

contributions in international journals and conference

papers. He is a member of IEEE, member of the Institute

of Engineers (India), member of Computer Society of

India and a life member in the System Society of India.

204


ABSTRACT

One of the most important and challenging issues in

real-time applications of resource-constrained wireless

sensor networks (WSNs) is providing end-to-end delay

requirement. Many wireless sensor network (WSN)

applications require real-time communication. In order

to address this challenge, we propose the Power Aware

Routing Protocol, which attains application-specified

communication delays at low energy cost by dynamically

adapting transmission power and routing decisions.

Extensive simulation results prove that the proposed

Protocol attains better QoS and reduced power

consumption.

Keywords: WSN, Robust nodes, link Quality

I. INTRODUCTION

Smart environments represent the next evolutionary

development step in building, utilities, industrial, home,

shipboard, and transportation systems automation.

Like any sentient organism, the smart environment relies

first and foremost on sensory data from the real world.

Sensory data comes from multiple sensors of different

modalities in distributed locations [1,2,3] . The smart

environment needs information about its surroundings

as well as about its internal workings; this is captured

in biological systems by the distinction between

exteroceptors and proprioceptors.

Power Adaption Routing Protocol For Realtime ApplicationsIn Wireless Sensor Networks

R.Prema1, R.Rangarajan2

1. Research Scholar Bharathiar University Tamilnadu,India E-mail ID:[email protected]

2. VSB Engineering College, Tamilnadu, India. E-mailID:[email protected]

Wireless sensor networks (WSN) represent a new

generation of embedded systems for routing sensory

data from the originator sensor node to the control station

[4]. Recent technological advances have enabled the

development of tiny battery-operated sensors [5,6,7].

Although energy efficiency is usually the primary concern

in WSNs, the requirement of low latency

communication is getting more and more important in

new applications. For example, a surveillance system

needs to alert authorities of an intruder within a few

seconds of detection.

Supporting real-time communication in WSNs is very

challenging. First, WSNs have lossy links that are greatly

affected by environmental factors [8][9]. As a result,

communication delays are highly unpredictable. Second,

many WSN applications (e.g., border surveillance) must

operate for months without wired power supplies.

Therefore, WSNs must meet the delay requirements at

minimum energy cost. Third, different packets may have

different delay requirements.

Literature Review

Power-aware algorithms for routing in WSNs have

received considerable attention over the past few years.

A distributed position-based algorithm to form topologies

containing a minimum total energy route between any

pair of connected nodes is proposed in [10]. Based on

this initial work, a computationally simpler protocol with

better performance is described in [11,12]. Similar

topology control algorithms based on discretization of

the coverage region of a node into cones are proposed in

[13,14]. The idea is to select appropriate transmitter

205

Power Adaption Routing Protocol For Realtime Applications In Wireless Sensor Networks

power levels to guarantee network connectivity while at

the same time transmission energy is saved.

Putting a node into sleep mode whenever its active

collaboration in the current network task is not required

is another way to save energy. The geographical adaptive

fidelity (GAF) algorithm [15] conserves energy by

turning off nodes that are equivalent from a routing

perspective, thereby keeping a constant level of routing

fidelity. An improvement of GAF based on a relationship

between optimal transmission range and traffic is

described [16]. In Span [17], the decision whether a node

should be awake or sleep is made depending on how

many of its neighbors will get benefit and how much

remaining energy it has. The sparse topology and energy

management (STEM) protocol [18] puts nodes

aggressively into sleep mode and only wakes them up

when they are needed to forward data. Data fusion is a

technique that can be used to reduce the amount of

redundant information prevalent in dense sensor

networks. By combining data with equal semantics,

unnecessary power consumption due to transmission and

processing of duplicate data is prevented. Two prominent

routing protocols that use upper layer information for

data fusion as well as making routing decisions are

Directed Diffusion [19] and SPIN [20]. Application-

specific fusion enables even more sophisticated data and

node management functionalities insideWSNs [21,22].

Both sleep scheduling and data fusion are desirable

functionalities which may complement energy-efficient

MAC androuting protocols.

Proposed Work

Among all the sensor nodes in the network, there are

some robust nodes. These robust nodes serve as the

backbone for the routing in wireless sensor networks.

The remaining sensor nodes are common sensor nodes.

Each robust node maintains a table of sensor node power

at other robust nodes. So in the route, each robust node

will compute the end-to-end power from itself to any

other robust nodes. The sensor node power is estimated

and updated periodically by each robust node. The robust

node which is nearest to the source node finds the robust

nodes which are along the route towards destination

sensor node. Then packets will be forwarded through

these robust nodes to the destination node. Since robust

nodes have better communication capability than

common nodes, most of the time the power is less than

the maximum power. This protocol is compared with

AODV protocol. This protocol shows better power

adaption than AODV protocol.

Estimation of Link Quality

The communication in mobile ad-hoc network

is based on electronic signals. In mobile ad-hoc networks

it is possible that a communication path (route) will break.

This will happen primarily because of the nodes present

in the network are moving around the region. The fig.1,

depicts the scenario when the link is active. In the fig.1,

three nodes are present namely a, b and c. The node-b is

within the range of the node-a and node-c. But, the node-

a is not within the range of node-c and node-c is not

within the range of node-a. Hence for transmission of

data from node-a to node-c, the node-b acts as an

intermediate node. After certain duration, due to the

mobility of sensor nodes, the link gets break and the data

communication between the nodes becomes unreliable.

Fig.1 Before the link breaks

206


Fig.2 After the link breaks

to the mobility of nodes present in wireless sensor

network it becomes mandatory to consider the quality of

the link.

To be able to see that when a node in the wireless

sensor network is moving and hence a route is about to

break as shown in Fig.2. So that factor, it is probable to

measure the quality of the signal and based upon that

presumption, when the link is going to break. This

information which is identified by the physical layer is

send to the upper layer when packets are received from

a node, and then indicate that node is in pre-emptive zone.

Pre-emptive zone is the region where the signal strength

is weaker which leads to the link failure. Pre-emptive

zone uses the pre-emptive threshold value to fix the pr-

emptive zone’s location. Thus, using the received signal

strength from physical layer, the quality of the link is

predicted and then the links which are having low signal

strength will be discarded from the route selection.

When a sending node broadcasts RTS packet,

it piggybacks its transmission power. While receiving

the RTS packet, the projected node quantifies the strength

of the signal received.

Lq = P

R

Where,

PR refers Power of the Receiving node,

PT stands for Power of the Transmitting node,

ë stands for wavelength carrier,

d is the distance between the sending and the

receiving node,

UGR stands for unity gain of receiving omni-

directional antenna

UGT stands for unity gain of transmitting omni-

directional antenna.

T POW

= max (Lq &

RPOW

)

Where,

CV = Cost Value,

Lq = Link quality

R POW

= Residual Power of the sensor node

Election of robust node

At the start, one robust node is set in each grid.

We need an election mechanism to produce new Robust

nodes because robust nodes also move around. When a

Robust node leaves its current grid or due to any other

reason there is no robust node in the grid. Suppose, there

are more Robust nodes in the current grid of the network,

then, the next node with maximum weighted value from

the sorted list will be chosen as the new Robust node for

the grid. In the proposed routing algorithm, we need to

compute the minimum delay between two robust nodes,

and find the path with the minimum delay.

For each valid path Pi,

For every node nk in Pi

t_power = t_power + power (nL, nk) + power (nk)

If t_power >= max_power, delete this path, break.

If t_power >= min_power, delete this path, break.

If nk is the destination D, and t_power < min_power,

min_power = t_power;

best_path = Pi + {nk};

Else add node nk to the end of the path,

End For

End For

Pseudo code for Robust Sensor node election

Simulation Settings & Graphs

207


No. of Nodes 50, 75, 100, 125 and 150

Area Size 1000 X 1000

Mac 802.11

Radio Range 250m

Simulation Time

50 sec

Traffic Source CBR

Packet Size 512 KB

Mobility Model

Random Way Point

Speed 5 m/s

Pause time 100 Seconds

�

02468

101214

50 75 100 125 150

Po

wer

(W

atts

)

No. of Nodes

TOTAL POW ER CONSUMPTION OF SENSOR NODES (AODV Vs PARP)

AODV

PARP

Table 1. Simulation Settings

Fig.3 Pause time Vs Paper Consumption

It is proved that from Fig. 3 the Power adaption routing

protocol produces better power consumption than AODV

protocol.

II. Conclusion

This paper addresses the issue of Power adaption and

QoS effective routing by design and development of

Power Adaption Routing Protocol (PARP). Also,

scalability issue is kept in mind when the number of

nodes in the network is increased from the range of 50

to 150 nodes. The total power consumption, delivery ratio

and delay are taken as the performance metrics and the

simulation results proved PARP is better than AODV

protocol.

REFERENCES

[1] V. Rodoplu and T. H. Meng. Minimum energy

mobile wireless networks. IEEE Journal on Selected

Areas in Communications, 17(8):1333–1344, Aug.

1999.

[2] L. Li and J. Y. Halpern. Minimum-energy mobile

wireless networks revisited. In Proc. IEEE

International Conference on Communications

(ICC), pages 278–283, June 2001.

[3] L. Li, J. Y. Halpern, P. Bahl, Y.-M. Wang, and R.

Wattenhofer. A cone-based distributed topology-

control algorithm for wireless multi-hop networks.

IEEE/ACM Transactions on Networking,

13(1):147–159, Feb. 2005.

[4] R. Wattenhofer, L. Li, P. Bahl, and Y.-M. Wang.

Distributed topology control for power efficient

operation in multihop wireless ad hoc networks. In

Proc. IEEE INFOCOM, pages 1388–1397, Apr.

2001.

[5] Y. Xu, J. Heidemann, and D. Estrin. Geography-

informed energy conservation for ad hoc routing.

In Proc. ACM/IEEE International Conference on

Mobile Computing and Networking (MobiCom),

pages 70–84, July 2001.

[6] Q. Gao, K. J. Blow, D. J. Holding, I. W. Marshall,

and X. Peng. Routing analysis and energy efficiency

in wireless sensor networks. In Proc. IEEE

International Symposium on Circuits and Systems

(ISCAS 2004), pages 533–536, June 2004.

[7] B. Chen, K. Jamieson, H. Balakrishnan, and R.

Morris. Span: An energy-efficient coordination

208


algorithm for topology maintenance in ad hoc

wireless networks. In Proc. ACM/IEEE

International Conference on Mobile Computing and

Networking (MobiCom), pages 85–96, July 2001.

[8] C. Schurgers, V. Tsiatsis, S. Ganeriwal, and M.

Srivastava. Topology management for sensor

networks: Exploiting latency and density. In Proc.

ACM International Symposium on Mobile Ad Hoc

Networking and Computing (MobiHoc), pages 135–

145, June 2002.

[9] C. Intanagonwiwat, R. Govindan, and D. Estrin. Directed

diffusion: A scalable and robust communication paradigm

for sensor networks. In Proc. ACM/IEEE International

Conference on Mobile Computing and Networking

(MobiCom), pages 56–67, Aug. 2000.

[10] J. Kulik, W. Heinzelman, and H. Balakrishnan.

Negotiation-based protocols for disseminating

information in wireless sensor networks. Wireless

Networks, 8(2/3):169–185, Mar. 2002.

[11] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan.

Energy-efficient communication protocol for wireless

microsensor networks. In Proc. Hawaiian International

Conference on Systems Science, pages 1–10, Jan. 2000.

[12] D. Estrin, R. Govindan, J. Heidemann, and S. Kumar.

Next century challenges: Scalable coordination in sensor

networks. In Proc. ACM/IEEE International Conference

on Mobile Computing and Networking (MobiCom),

pages 263–270, Aug. 1999.

[13] E. J. Duarte-Melo and M. Liu. Analysis of energy

consumption and lifetime of heterogeneous wireless

sensor networks. In Proc. IEEE GLOBCOM, pages 21–

25, Nov. 2002.

[14] J. Pan, L. Cai, Y. T. Hou, Y. Shi, and S. X. Shen. Optimal

base-station locations in two-tiered wireless sensor

networks. IEEE Transactions on Mobile Computing,

4(5):458–473, Sept. 2005.

[15] J. Pan, Y. T. Hou, L. Cai, Y. Shi, and S. X. Shen. Topology

control for wireless sensor networks. In Proc. ACM

International Conference on Mobile Computing and

Networking (MobiCom), pages 286–299, Sept. 2003.

[16] T. S. Rappaport. Wireless Communications: Principles

& Practice. Prentice-Hall, New Jersey, 1996.

[17] G. Gupta and M. Younis. Fault-tolerant clustering of

wireless sensor networks. In Proc. IEEE Wireless

Communications and Networking Conference (WCNC),

pages 1579–1584, Mar. 2003.

[18] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan.

Energy-efficient communication protocol for wireless

microsensor networks. In Proc. Hawaiian International

Conference on Systems Science, pages 1–10, Jan. 2000.

[19] O. Younis and S. Fahmy. HEED: A hybrid, energy-

efficient, distributed clustering approach for ad hoc sensor

networks. IEEE Transactions on Mobile Computing,

3(4):366–379, Oct. 2004.

[20] Y.-B. Ko and N. H. Vaidya. Location-aided routing (LAR)

in mobile ad hoc networks. Wireless Networks, 6(4):307–

321, July 2000.

[21] I. Stojmenovic and X. Lin. Power-aware localized routing

in wireless networks. IEEE Transactions on Parallel and

Distributed Systems, 12(11):1122–1133, Nov. 2001.

[22] Y. Xu, J. Heidemann, and D. Estrin. Geography-informed

energy conservation for ad hoc routing. In Proc. ACM/

IEEE International Conference on Mobile Computing and

Networking (MobiCom), pages 70–84, July 2001.

209



Mrs. R. Prema working as Assistant

Professor in the Department of

Electroincs in Karpagam University.

She is pursuing her PhD in Bharathiar

University. She has presented about 7

papers in the National Conference and 1 in International

Conference. She has published a paper in an International

Journal of Wireless Sensor Networks. Her research area

is Wireless Sensor Networks.

Dr. R. Rangarajan is the Director in

V.S.B Engineering College, Karur. He

has a distinguished career in teaching

and research for more than 35 years.

Dr.Rangarajan has to his credit 20 technical papers in

National and International Journals. He has presented

about 35 papers in national and international

Conferences.

210


Path Planning in AI (Artificial Intelligence)

T.C. Manjunath 1 G.V. Jayaramaiah 2

ABSTRACT

This paper presents a technique of designing a wide path

motion heuristic path in robots using artificial intelli-

gence in 2D scenario, i.e., using simulations. The simu-

lation results done in C++ show the effectiveness of the

method used.

Keywords:

Motion heuristics, Wide path motion heuristics, Path

Planning, Artificial Intelligence.

1.INTRODUCTION

In the recent days, path planning using Artificial

Intelligence has gained a lot of importance, similar to

those searching the path in the real world by humans.

These concepts are being used widely in the field of

autonomous agents such as the robots. Wong in [1]

presented a heuristic-based motion estimation technique

which reduced both the number of search locations (via

sub-sampling) as well as the number of operations to

perform at each search location (via simplified signature),

thus achieving a high speedup factor [1]. Test results

showed that quality of the motion vectors found by this

heuristic-based technique was close to that of full-search

motion vectors [1].

1. Principal, HKBK College of Engineering, S.No. 22/1,Nagawara, Arabic College Post, Bangalore-45, Karnataka.Email : [email protected]

2. Professor & HOD, Dept. of ECE, Dr. Ambedkar Inst. ofTech., Bangalore, Karnataka.Email : [email protected];[email protected] ;gv_ [email protected]

The issue of motion planning for closed-loop

mechanisms, such as parallel manipulators or robots, is

still an open question. Houssem Abdellatif & Bodo

Heimann proposed a novel approach for motion planning

of spatial parallel robots in [2]. In this paper, the

framework for the geometric modeling was based on the

visibility graph methodology. It was opted for a multiple-

heuristics approach, where different influences were

integrated in a multiplicative way within the heuristic

cost function.

Since the issue of singularities was a fundamental

one for parallel robots, it was emphasized on the

avoidance of such configurations. To include singularity-

free planning within the heuristic approach, two heuristic

functions were proposed by them, the inverse local

dexterity as well as a novel defined “next-singularity”

function, in such a way, well conditioned motions can

be provided by a single planning procedure [2]. Paolo

Fiorini & Zvi SIIillCr in [3]

Since the issue of singularities was a fundamental

one for parallel robots, it was emphasized on the

avoidance of such configurations. To include singularity-

free planning within the heuristic approach, two heuristic

functions were proposed by them, the inverse local

dexterity as well as a novel defined “next-singularity”

function, in such a way, well conditioned motions can

be provided by a single planning procedure [2]. Paolo

Fiorini & Zvi SIIillCr in [3] presented heuristic methods

for motion planning in dynamic environments, based on

the concept of Velocity Obstacle (VO). Two heuristic

strategies, viz., selecting the maximum velocity along


211

the line to the target, and selecting the maximum feasible

velocity within a specified angle from the straight line to

the target were presented [3].

In this paper, we present a method of designing a

wide path motion based on motion heuristics techniques.

The path uses the concept of graph theory & going in

mid-way between the obstacles so that the obstacles are

avoided. The paper is organized in the following

sequence. Firstly, a brief introduction to the related work

was presented in the previous paragraphs. Section 2 gives

a in-depth concept on the motion heuristics approach

used in the AI techniques. In section 3, the mathematical

modeling of the wide path motion heuristic approach is

dealt with. Section 4 shows the simulation results. The

paper concludes with the conclusions followed by the

references.

I.CONCEPT OF MOTION HEURISTICS

Motion Heuristics is defined as the method of

searching an obstacle collision free path in the free work

space of the robot from the source to the destination by

making use of search techniques such as the graph theory

( AND / OR graphs ), chain coding techniques and the

state space search techniques (best first search, breadth

first search) used in Artificial Intelligence [4]. The search

techniques used in AI to find the path from the source S

to the goal G are called as motion heuristics or the robot

problem solving techniques. The word ‘heuristic’ means

to search, what to search ? an obstacle collision free path

to search [5].

Holding the object by the gripper ⊥r to its width and

searching for an obstacle collision free path in the work

space of the robot from S to G is known as wide path

motion heuristics [6]. In this type of motion heuristics

method of searching a path, the mobile object / part which

is to be held by the gripper is rectangular in shape with a

length of L units and width of W units [7]. The object is

held by the gripper ⊥r to its width and the point p of the

gripper should match with the center of gravity of the

object, like how we hold a duster while rubbing the board.

II.MATHEMATICAL CONCEPTS

Consider the moving / mobile part ABCD to be rectangular

in shape as shown in Fig. 1. Let W and L be the width and

length of the rectangular mobile part or object and let W ≤

L. Let P1 and P

2 be the two reference points marked on the

major axis of the mobile rectangle at a distance of 4

L units

from each ends and at a distance of 2

W units from the parallel

side as shown in the Fig. 1 [8].

These two points will always be moving along the

GVD path. G is the centre of gravity of the rectangular

mobile object. Two reference points are used in order to

simplify the wide path motion heuristics. The two

reference points are like the front and back wheels of a

two wheeler automobile. The front point P2 is advanced

incrementally along the GVD path. The back point P1 is

found by drawing a GVD circle of radius 2

L centered at

P2. This circle cuts the major axis of the mobile rectangle

ABCD at the point P1. Thus, both the points on the GVD

path are obtained [9].

�

Figure 1. Wide path motion heuristics

A mobile part ( Rectangle – ABCD ) ;L → Length ; W → Width ;P

1, P

2 → Reference points ; G → Centroid ;

R → Radius of GVD circle

212


Let us determine how the wide path motion

heuristics is applicable. Consider two circles ; one with

P1 as center

222

42

+

== LWRRadius and another circle

with P2 as center [10].

The entire mobile rectangular part is contained

within the two overlapping circles. The wide path motion

heuristic is therefore attempted when the minimum radius

of the chosen GVD path is greater than the radius of

either of the circles. i.e., when the radius of the path in

between the obstacles is greater than the radius R of the

mobile part (of either of the circles on the mobile part),

then, there will be sufficient space for the mobile part to

move in between the obstacles and as a result of which

there will be no collision of the part with the edges of

the obstacle [11].

The minimum radii constraint of the chosen path is

given by the formula which is derived as follows [13].

Consider any of the circle with either P1 or P

2 as center.

Consider the triangle DP2E [12].

DE = EC = W2 ;

P2E =

14 L

=

L4

∴, minimum radius [14],

P2 D = P

2 C = P

1A = P

1B >

�

22

42

+

LW

R > �

22

42

+

LW

i.e., R > � 2

122

42

+

LW (1)

As long as the space between the obstacles is greater

than this minimum radii constraint R as given by the

equation (1), then the part will not collide with the ob-

stacle, since it is at [16] a safer distance from the ob-

stacle and clearance exists between the walls of the ob-

stacles and the part [15].

III. WIDE PATH MOTION HEURISTICS

In this section, we present an example of a motion

planned with wide path motion heuristics. An example

of a motion planned with wide path motion heuristics is

shown in the Fig. 2 [17]. The workspace consists of three

obstacles ; viz., a rhombus, a triangle, a rectangle inside

the robot workspace boundary which is a rectangle. The

mobile part is considered as a small rectangle. The mobile

part is at source position, S. It has to reach the point G

[18].

A shortest path is found out using the GVD

techniques and the motion heuristics. The mobile part

starts at S, moves along the [19] shortest GVD path as

shown in the Fig. 2 in which the center point of the mobile

part lies always on the GVD path. Note that translations

of the mobile part is performed along the freeways and

rotations of the part has to performed [20] at the junction

the freeways in order to obtain a smooth motion.

Sophisticated Wide path motion heuristics

Many a times, ordinary wide path motion heuristics is

not sufficient to determine an obstacle collision free path

from the source to the destination [21]. In this case,

sophisticated wide path motion heuristics comes into the

picture as the work space is fully cluttered with obstacles

[22]. When the work space is cluttered with obstacles,

then sophisticated wide path motion heuristics has to be

used. More sophisticated wide path motion heuristic

techniques can be applied to the original path if the only

available path satisfies R > 2

W; but, fails to satisfy the

[31] minimum radius constraint as given by Eqn (1) ;

i.e., the space in between the obstacles is greater than

only half the width of the mobile part [33]. If this is the

condition [30]; then, a well designed path has to be


213

planned ; since, there is only minimum space left between

the mobile part and the obstacle [23].

�

S

B

A

C

G

Source

Goal

GVD path

Figure 2. : Wide path motion heuristicsA, B, C - Obstacles ; S - Source ; G – Destination /

Goal

An example of a path planned in a cluttered

workspace using more sophisticated motion heuristics

is shown in the Fig. 2. In a cluttered / densely packed

obstacle zone, the GVD graph will be a complicated

graph and the shortest path consists of a huge number of

nodes, arcs and segments [24].

IV. SIMULATION EXAMPLE

The best example of an sophisticated wide path motion

heuristics can be a real time application, i.e., when a

human being is driving his vehicle in a densely packed

traffic jam, then he searches a path [32] in the 2D space

( i.e., on the road ) by moving in between the obstacles,

avoiding all the collisions between his vehicular system

and the other vehicles which are acting as the obstacles

[25].

�

S

G

A

D

C

B

E

Figure 3.: Sophisticated wide path motion heuristicsA, B, C, D, E - Obstacles ;

S, G - Source and goal scene of the mobile part

The human brain which is acting as the computer tries

to search for a obstacle collision free GVD path on the

road from the source to the destination [28]. Normally,

when we are driving our vehicle from the source to the

destination, we make use of the GVD technique to move

along the desired path [26].

A GUI was developed in C/C++ to simulate the same

for an obstacle collision free path in the 2D space with

the workspace is cluttered with simple 2D obstacles, say

a square [29]. The C/C++ code was run the & the

following simulation results were obtained as shown in

the Figs. 4 to 6 respectively [27].

Simulation Results

Figure 4. Inputting the parameters

214


Figure 5. Robot navigation

Figure 6. Source

Figure 7. Destination by avoiding the obstacles bymoving around them

V. CONCLUSIONS

A brief conceptual design of the wide path

motion heuristics was presented in this paper

along with the mathematical modeling concepts

followed by the simulations. The paper can be

further extended to 3 D objects & obstacles in

the work space. In this paper, the authors have

tried to explore the existing literature available

into programming form so that any generalized

path planning algorithm can be implemented

using the software.

REFERENCES

[1]. Y. Wong, “An efficient heuristic-based motion

estimation algorithm”, Proc. ICIP, Col. 2, pp. 22 -

25, 1995 International Conference on Image

Processing (ICIP’95) - Volume 2, 1995.

[2]. Houssem Abdellatif, Bodo Heimann, “A novel

multiple-heuristic approach for singularity-free

motion planning of spatial parallel manipulators”,

Robotica, Vol. 26, Issue 5, pp. 679-689, Sep. 2008.

[3]. Paolo Fiorini & Zvi SIIillCr, “Heuristic Motion

Planning in Dynamic Environments using Velocity

Obstacles”, Int. Conf. Paper.

[4]. P. Fiorini and Z. Shillcr, “Motiou planning in

dynamic environments using the relative velocity

paradigm”, IEEE Int. Conf. Robotics and

Automation, Atlanta, GA, May 7-10, 1993.

[5]. Craig J, Introduction to Robotics : Mechanics,

Dynamics & Control, Addison Wessely, USA,

1986.

[6]. Robert, J. Schilling, Fundamentals of Robotics -

Analysis and Control, PHI, New Delhi.

[7]. Klafter, Thomas and Negin, Robotic Engineering,

PHI, New Delhi.

[8]. Fu, Gonzalez and Lee, Robotics: Control, Sensing,

Vision and Intelligence, McGraw Hill.

[9]. Groover, Weiss, Nagel and Odrey, Industrial

Robotics, McGraw Hill.

[10]. Ranky, P. G., C. Y. Ho, Robot Modeling, Control &

Applications, IFS Publishers, Springer, UK.

[11]. Crane, Joseph Duffy, Kinematic Analysis of Robotic

Manipulators, Cambridge Press, UK.

[12]. Manjunath, T.C., (2005), Fundamentals of

Robotics, Fourth edn., Nandu Publishers, Mumbai.


215

[13]. Manjunath, T.C., (2005), Fast Track to Robotics,

Second edn., Nandu Publishers, Mumbai.

[14]. William Burns and Janet Evans, (2000), Practical

Robotics - Systems, Interfacing, Applications,

Reston Publishing Co.

[15]. Fundamentals of Robotics : Analysis and Control

– Robert J Schilling ; PHI, New Delhi.

[16]. Robotics Series ( Volume I to VIII ) – Phillip

Coiffette ; Kogan Page, London.

[17]. Robotic Engineering – Klafter, Thomas, Negin ;

PHI, New Delhi.

[18]. A Robotic Engineering Text Book – Mohsen

Shahinpoor ; Harper & Row Publishers.

[19]. Robotics and Image Processing – Janakiraman ;

Tata McGraw Hill.

[20]. Robotic Manipulators – Richard A Paul ; MIT

press, Cambridge.

[21]. Computer Vision for Robotic Systems – Fairhunt

; New Delhi.

[22]. Robotics for Engineers – Yoram Koren ; McGraw

Hill.

[23]. Industrial Robotics – Groover, Weiss, Nagel, Odrey;

McGraw Hill.

[24]. Robotics : Control, Sensing, Vision & Intelligence

– Fu, Gonzalez and Lee ; McGraw Hill.

[25]. Industrial Robotics – Bernard Hodges ; Jaico

Publishing House.

[26]. Foundations of Robotics : Analysis and Control –

Tsuneo Yoshikawa ; PHI.

[27]. Robotics : Principles & Practice – Dr. Jain and Dr.

Aggarwal ; Khanna Publishers, Delhi.

[28]. Modeling and Control of Robotic Manipulators –

Lorenzo and Siciliano, McGraw Hill.

[29]. Mechanotronics of Robotics Systems – Dr.

Amitabha Bhattacharya.

[30]. Industrial Robotics - S.R. Deb, Tata MacGraw Hill.

[31]. Robot Modelling, control and applications – P G

Ranky and C Y Ho,IFS publishers, Springler,UK.

[32]. Industrial Robots and Robotics – Edward Kafrissen and Mark

Stephans, Prentice Hall Inc, Virginia.

[33]. Fundamentals of Industrial Robots and Robotics – Rex Miller,

PWS Kent Pub Co., Boston.


Dr. T.C. Manjunath was born

in Bangalore, Karnataka, India

on Feb. 6, 1966 & received the

B.E. Degree (Bachelor of Engg.)

in Electrical Engg. from R.V.

College of Engg. (Bangalore

University) in the year 1989,

M.Tech. degree in Electrical Engg. with specialization

in Automation, Control & Robotics from the Govt.’s L.D.

College of Engg. (Gujarat University) in the year 1995

and Ph.D. in Systems & Control Engg. from the

prestigious Indian Institute of Technology Bombay (IIT

Bombay) in the year 2007 respectively. Engineering (IIT

Bombay, India) and worked on control of space launch

vehicles using FOS feedback technique in IITB. He has

published a number of publications in the various

National, International journals and Conferences and

published 3 textbooks on Robotics,He has also published

a research monograph in the International level from the

Springer-Verlag publishers based on my Ph.D. thesis

topic titled, “Modeling, Control and Implementation of

Smart Structures”, Vol. 350, LNCIS, costing 79.95 Euros.

He is a member of IEEE for the past 12 years, He was

awarded with the “Best research scholar award in

engineering discipline” for the academic year 2006-07

from the Research Scholars Forum (RSF) from Indian

Institute of Technology Bombay (IITB). His current areas

216


of interest are Control System Engineering, Robotics,

Signals & systems (continuous time and discrete time),

Digital Signal Processing, Digital Image Processing,

Electrical Circuits & Networks - Analysis & Synthesis,

Elements of Electrical Engineering, Mechatronics,

Artificial Intelligence, Matlab Programming.

Dr. Jayaramaiah GV was born in

Tumkur district, Karnataka, India &

received the B.E. Degree (Bachelor

of Engg.) from Siddaganga Institute

of Tech., Tumkur (Bangalore

University) in the year 1990, M.Tech.

degree with specialization in Power Electronics from the

BMS College of Engg., Bangalore (Bangalore

University) in the year 1992 & Ph.D. in Electrical Engg.

from the prestigious Indian Institute of Technology

Bombay (IIT Bombay) in the year 2008 respectively.

He is currently working as Professor & Head of the

Department of Electronics & Communications in Dr. BR

Ambedkar Institute of Technology (Dr. AIT), Bangalore,

Karnataka, India. He has published more than 1 dozen

publications in the various National, International

journals and Conferences.

Efficient Object Detection and Classification using Hybrid ELM

with Analytic Hierarchy Process and Bayesian Network

217


with Analytic Hierarchy Process and Bayesian NetworkN.V. Balaji1, M. Punithavalli2,

ABSTRACT

Detecting object is one of the typical difficulties in computer

technology which has its usage to surveillance, robotics, mul-

timedia processing, and HCI. The multi-resolution framework

is utilized by the proposed technique for object detection. In

this efficient object detection, the lower resolution features

are first used to discard the majority of negative windows at

comparatively small cost, leaving a relatively small amount of

windows to be processed in higher resolutions and this helps

to attain better detection accuracy. Then the frameworks on

Histograms of Oriented Gradient (HOG) features are used to

detect the objects. For training and detection, the classifier

used previously is Support Vector Machine (SVM) and Ex-

treme Learning Machine (ELM). Hybrid ELM is used in the

proposed technique to reduce the time for detection and im-

prove the accuracy of classification. The input weights and

hidden biases are created with the help of integrated Analytic

Hierarchy Process (AHP) and Bayesian Network (BN) model.

The experimental result shows that the proposed technique

achieves better detection rate when compared to the existing

techniques.

Keywords:

Histograms of Oriented Gradient (HOG), Modified Extreme

Learning Machine, Multi-Resolution Framework, Analytic

Hierarchy Process (AHP), Bayesian Network (BN).

I.INTRODUCTION

Object detection and recognition in noisy and cluttered

images is challenging problem in computer vision. The

goal of this research is to identify objects in an image

accurately. Today, there is an increased need for object

detection. There are several problems in detecting and

recognizing the objects in an image. It is an important

part in many applications such as image search, image

auto-annotation and scene understanding; however it is

still an open difficulty due to the complexity of object

classes and images. [3] The robot usage will reduce the

work load for the housekeepers. Some of the works

performed by home robots are cleaning, tidying, fetching

objects, etc. These tasks need some degree of semantic

knowledge about human surroundings so that they can

find the way, search for objects and communicate with

humans successfully. For example, for a robot to execute

the command “bring me an apple”, a robot must recognize

the term “apple” and to know which places in the

surroundings are expected to contain it (e.g., the kitchen).

This paper provides the better object detection technique.

Initially, multi-resolution framework is used to eliminate

the unnecessary features from the image which will help

in speedy processing.[4] The learning machine used in

this work is Hybrid ELM. Finally, Histograms of Oriented

Gradient features are applied for accurate detection of

object in the image.

II. RELATED WORKS

Broussard et al., [5] put forth the physiologically

motivated image fusion for object detection using a pulse

1 Assistant Professor Department of Computer Science,Karpagam University, Coimbatore, India.2 Director Department of Computer Applications, Sri RamakrishnaEngineering College,Coimbatore, India.

218


coupled neural network. PCNN are implemented to

combine the results of several object detection techniques

to improve object detection accuracy. The object

detection assets of the obtained image fusion network

are demonstrated on mammograms and forward-looking

infrared radar (FLIR) images. This technique exceeded

the accuracy obtained by any individual filtering methods

or by logical ANDing the individual objects detection

technique results.

Kerekes et al., [6] had a look on spectral imaging

technique analytical model for sub-pixel object detection

[16]. To learn the system parameters in the context of

land cover classification, an end-to-end remote sensing

system modeling approach was previously developed.

In this work, the author extends this technique to sub-

pixel object detection applications by including a linear

mixing model for an unresolved object in a background

and using object detection algorithms and probability of

detection (PD) versus false alarm (P

FA) curves to

characterize performance.

Kubo et al., [7] developed an image processing system

for direction detection of an object using neural network.

The proposed direction detection system was considered

to consist of two modules of neural networks; one is an

edge detection module and the other is a direction

detection module. The edge detection technique could

detect edges when differences of gray level between the

object and the background were larger than thirteen at

any gray level of the object.

Wen-jie Wang et al., [8] illustrate the Object-oriented

multi-feature fusion change detection method for high

resolution remote sensing image. The key steps of this

technique are segment the image, choose optimized

features from spectral features, texture features, shape

features of the segment objects, use optimized features

to do change detection, make fusion of the preliminary

change detection results of the different optimized

features to get the final result.

III. METHODOLOGY

The initial process for the proposed object detection

technique is the multi-resolution framework which is

applied in different resolution space. The next process

is the application of training and detection technique with

the help of Hybrid ELM. Finally, Histogram of Oriented

Gradients (HOG) features are applied for detection of

objects.

Multiple Resolution Framework

The common approach of object detection using multi-

resolution is shown in Fig. 1. The framework encodes

resolution together with scale spaces in 2D coordinate

scheme.

Along the vertical axis in scale space, the image gets

down sampled in order to use a fixed size detection

window to locate objects of larger scales. Along the

horizontal axis in resolution space, any detection window

is classified hierarchically from its lowest resolution to

full resolution. A window rejected in lower resolution

will not be passed to any higher resolutions.

Figure 2 shows why multi-resolution is useful for object

detection. A Canny edge detector was applied to images

from the VOC 2006 at different resolutions. Images in

lower resolutions were created by downsampling and

then upsampling to the original size for easier

comparison. For the motorbikes and cows categories, two

samples are picked from each and computed the shape

context matching distances shown in the figure.



219

Figure 1. Multiple resolution framework for object detection.

Figure 2. Sample images from VOC 2006 database, with edge detected images at different resolutions

Training/detection algorithms

For a typical object detection system, the training set is

composed of normalized image patches. Negative patches

were scanned from images that do not contain any target

object instances. Suppose R resolutions are used, where

r=1,...,R from the lowest resolution to the full resolution.

Assume the step of downsampling ratio is á. For each

resolution r, the training set is defined as Tr = {(Ir(i),

lr(i)), i=1, ...,Nr}, where Ir(i) is the image patch, lr(i) “

{1,”1} is the associated label and Nr is the number of

patches. The training

starts from the lowest resolution, and then loops between

bootstrap and training towards higher resolutions until

the full resolution. The output of the training is a

hierarchical classifier with components from each

resolution. Suppose that the image is downsampled with

a ratio of â for each scale up, and that it have S scales

s=1,...,S, for larger to smaller objects. Suppose the

220


window size at each resolution is (wr,hr)=(w,h)/áR”r,

and that the size is fixed across all scales. For each

resolution r and scale s, a list of status of detection

windows ãr, s(k)”{1,”1} is kept. The status is initialized

to be 1, and set to “1 if the window is rejected by a

classifier.

The training and detection is further improved by Hybrid

Extreme Learning Machine (ELM). Previous approaches

have used SVM and ELM for the purpose of training

and detection.

SVM Algorithm

SVM is used to find an optimal separating hyper plane

(OSH) which generates a maximum margin between the

categories of data. To build an SVM classifier, a kernel

function and its parameters need to be chosen. So far, no

analytical or empirical studies have established the

superiority of one kernel over another conclusively. The

following three kernel functions have been applied to

build SVM classifiers:

1) Linear kernel function, K(x, z) =⟨x, z⟩ ;

2) Polynomial kernel function K(x, z) =(⟨x, z ⟩+1) d is the

degree of polynomial;

3) Radial basis function K(x,z)=exp is the

width of the function.

The usage of SVM will reduce the time requirement for

classification and also the accuracy of classification but

it can be improved using ELM.

Extreme Learning Machine

Extreme Learning Machine (ELM) meant for Single

Hidden Layer Feed-Forward Neural Networks (SLFNs)

that will randomly select the input weights and

analytically determines the output weights of SLFNs. The

ELM has several interesting and significant features

different from traditional popular learning algorithms for

feed forward neural networks.

Extreme Learning Machine Training Algorithm

If there are N samples (xi, t

i), where x

i = [x

i1, x

i2… x

in] T

∈Rn and ti = [t

i1, t

i2, … , t

im]T∈Rn, then the standard SLFN

with N hidden neurons and activation function g(x) is

defined as:

where wi = [w

i1, w

i2, … , w

in]T represents the weight vec-

tor that links the ith hidden neuron and the input neu-

rons, ßi = [ß

i1, ßi2, … , ßim]T represents weight vector that

links the ith neuron and the output neurons, and bi repre-

sents the threshold of the ith hidden neuron. The “.” in

wi . x

j indicates the inner product of w

i and x

j. The SLFN

try to reduce the difference between oj and t

j.

More in a matrix format as H ß = T, where

�

and



221

The result reduces the norm of this least squares equation

is

where H† is known as Moore-Penrose generalized

inverse.

The usage of Hybrid ELM will reduce the time

requirement for classification and also the accuracy of

classification can be improved over the previous

techniques.

Modified Extreme Learning Machine

A Hybrid ELM technique which uses ELM and LM

technique can be described as below:

Initially, the input weights and hidden biases are created

with the help of integrated Analytic Hierarchy Process

(AHP) and Bayesian Network (BN) model.

By integrating AHP and Bayesian Network in such a way,

that the weights, the output of AHP will be given as input

to BN. Bayesian Network can obtain inputs from several

points, but in the combination BN adopts the weights, as

inputs in leafs of the graph; and these weights are

structured in vector format. The serial of comparative

weights of computed by AHP method can be written in a

vector format, this vector will be given as input to BN.

Next, the corresponding output weights are analytically

determined with the help of ELM algorithm during the

first iteration and randomly generate the output hidden

biases. Then, the parameters (all weights and biases) are

restructured with the help of LM algorithm.

The process for the Hybrid Extreme Learning Machine

is described below:

Provided a training set

activation functions , and hidden nodes

namely of first and second hidden layer.

Step 1: Randomly choose the starting values of input

weight vectors and bias vector with the help of

AHP-BN technique and bias vector without using

the AHP-BN technique.

Step2: Determine the first hidden layer output matrix

. With the help of ELM algorithm, determine the

output weight

Step3: Determine the second hidden layer output

matrix , errors

Determine the elements of the Jacobian matrix with the

equations

and

Application to HOG

In Histogram of oriented Gradients (HOG) feature for

pedestrian detection, pixels are first grouped into smaller

spatial units called cells. For each cell, a histogram

feature on gradients orientations is extracted. The

magnitude of the gradient is applied as the weight for

voting into the histogram. The descriptor of each block

is the concatenation of all cell features. Inside each

detection window, densely sampled and overlapping

blocks produce redundant descriptors, which is important

222


for better performance. Finally, a linear Support Vector

Machine (SVM) is used to classify individual detection

windows. In the proposed multi-resolution approach

based on the above method, the author used blocks of

varying sizes to capture features in different spatial

frequencies. An important consequence of the feature

hierarchy is that, both more-global (low resolution) and

more-local (high resolution) features are encoded.

IV. EXPERIMENTAL RESULTS

The proposed technique for object detection is

experimented on PASCAL Visual Object Classes

challenge 2006 database (VOC 2006). There are 10

groups of objects are presented in the database such as

bicycles, buses, cats, cars, cows, dogs, horses,

motorbikes, people and sheep. The multi-resolution

approach is applied for the sheep and car categories. The

proposed technique obtained the best results among all

participants on these two categories.

Figure 3. Comparison between Proposed and Existingmethods using Detection Error Trade-off curves on sheep

category in VOC2006 database

Figure 4. Comparison between Proposed andExisting methods using Detection Error Trade-off

curves on car category in VOC2006 database.

Figure 3 shows the comparison of proposed method with

existing object detection method using DET curves in

case of sheep category. From the graph it can be observed

that the resulted miss rate for the proposed system with

Hybrid ELM is lesser when compared to the existing

system with SVM and ELM. This shows that the Hybrid

ELM gives better classification result for the proposed

technique. Figure 4 shows the comparison of proposed

method with existing object detection method using DET

curves in case of car category. From the figure, it can be

clearly seen that the miss rate is lesser for the proposed

technique for all levels of false positive per window.

V. CONCLUSION

Object detection from the image is an important tech-

nique applied in many fields such as image browsing,

robotics, etc. The existing techniques lack accuracy and

take more time detection. To overcome these drawbacks,

the multi-resolution based detection system is proposed

in this paper. The first step is to discard the negative

windows using multi-resolution framework, the next step

is learning and detection in which the training samples



223

are supplied for training the system and the detection is

performed accordingly. To improve the accuracy of clas-

sification, Hybrid ELM is applied. The input weights

and hidden biases are created with the help of integrated

Analytic Hierarchy Process (AHP) and Bayesian Net-

work (BN) model. Finally, with the help of Histogram

of Oriented Gradients (HOG) features, the object is de-

tected from the image. The VOC 2006 database is used

for the purpose of experimentation and result shows that

this technique yields better accuracy for detection and

also miss rate for the proposed technique is lesser when

compared to the conventional techniques.

REFERENCES

[1] Jiqiang Song, Min Cai and M. S Lyu, “Edge color

distribution transform: an efficient tool for object

detection in images”, Conference on Pattern

Recognition, Proceedings. 16th International

conference, Pp: 608 – 61, vol.1, 2002.

[2] Saad Ali and Mubarak Shah, “A supervised learning

framework for generic object detection in images”,

Tenth IEEE International Conference on Computer

Vision, ICCV 2005, Pp: 1347 – 1354, Vol. 2, 2005.

[3] S. Sheraizin and S. Itzikowitz , “Unmanned object

detection for image surveillance systems”, IEEE

conference on EUROCON’09, Pp: 302 – 307, 2009.

[4] A. Steel and D. Brunner, “Detection and

Characterization of Urban Objects from VHR

Optical Image Data”, IEEE International conference

on Geoscience and Remote Sensing Symposium,

IGARSS 2008. Pp: III - 1256 - III – 1259, Vol.3,

2008.

[5] R.P. Broussard, R.P, S.K. Rogers, M.E. Oxley and

G.L. Tarr, “Physiologically motivated image fusion

for object detection using a pulse coupled neural

network”, IEEE Transactions on Neural Networks,

Pp: 554 – 563, Vol.10, 1999.

[6] J.P Kerekes and J.E Baum, “Spectral imaging system

analytical model for subpixel object detection”,

IEEE Transactions on Geoscience and Remote

Sensing, Page(s): 1088 – 1101, vol.40, 2002.

[7] T. Kubo, M. Obuchi, G. Ohashi and Y. Shimodaira,

“Image processing system for direction detection of

an object using neural network”, The 1998 IEEE

Asia-Pacific Conference on Circuits and Systems,

IEEE APCCAS 1998. Pp: 571 - 574, 1998.

[8] Wen-jie Wang, Zhong-ming Zhao and Hai-qing Zhu,

“ Object-oriented multi-feature fusion change

detection method for high resolution remote sensing

image”, 17th International Conference on

Geoinformatics, Page(s): 1 – 6, 2009.

[9] S. Belongie, J. Malik, and J. Puzicha. Shape

matching and object recognition using shape

contexts. PAMI, 24(4):509-522, 2002.

[10]S. C. Brubaker, M. D. Mullin, and J. M. Rehg.

Towards optimal training of cascaded detectors. In

ECCV06, pages I: 325-337.

[11] J. Fasola and M. Veloso. Real-time object detection

using segmented and grayscale images. In ICRA06,

pages 4088-4093.

[12]B. Leibe, E. Seemann, and B. Schiele. Pedestrian

detection in crowded scenes. In CVPR05, pages I:

878-885.

[13]S. Mitri, A. Nuchter, K. Pervolz, and H. Surmann.

Robust object detection in regions of interest with

an application in ball recognition. In ICRA05, pages

125-130.

[14]P. Viola, M. J. Jones, and D. Snow. Detecting

pedestrians using patterns of motion and appearance.

IJCV, 63(2):153-161, 2005.

[15]V. Ferrari, T. Tuytelaars, and L. J. V. Gool. Object

detection by contour segment networks. In ECCV06,

pages: 14-28. http://pascallin.ecs.soton.ac.uk/

challenges/VOC/voc2006/index.html

224



N.V. Balaji has obtained his

Bachelor of Science in Computer

Science from Sri Ramasamy Naidu

Memorial College, Sattur in 1997 and

Master of Science in Computer

Science from Dr. GRD College of

Science in 1997. Now he is doing

Ph.D., in Bharathiar University. He commences more

than nine years of experience in teaching field moreover

industrial experience in Cicada Solutions, Bangalore. At

present he is working as Asst. Professor & Training

Officer at Karpagam University. His research interests

are in the area of Image Processing and Networks. He

presented number of papers in reputed national and

international journals and conferences.

Dr. M. Punithavalli received the Ph.D

degree in Computer Science from

Alagappa University, Karaikudi in May

2007. She is currently serving as the

Adjunct Professor in Computer

Application Department, Sri Ramakrishna Engineering

College, Coimbatore. Her research interest lies in the

area of Data mining, Genetic Algorithms and Image

Processing. She has published more than 10 Technical

papers in International, National Journals and

conferences. She is Board of studies member various

universities and colleges. She is also reviewer in

International Journals. She has given many guest

lecturers and acted as chairperson in conference.

Currently 10 students are doing Ph.D., under her

supervision.

Empirical Analysis of Denoising Techniques in Video Processing

225


R.Revathi1, M. Hemalatha2

ABSTRACT

Video broadcast plays a very vital role in traffic applica-

tions. Existence of Noise in the video frames conceals

essential details. It negotiates with intensity of quality

of images in the video frame. So, it is crucial to elimi-

nate the noise from video frames. Removal of noise is

one of the pre-processing tasks in several video process-

ing techniques. Numerous researchers work on differ-

ent types of filters used to remove different types of

noises from video frames. There are some traditional fil-

ters, some filters derived from traditional filters and some

filters are new innovations. In this proposed method we

adapted the spatial video denoising methods, where noisy

image are applied to each frame individually. Since there

is a immense deal of eliminate noise from video con-

tent, this paper has been devoted to noise detection and

filtering technology with the aspire of eradicate unwanted

noise without affecting it negatively the clarity of scenes

that contain necessary detail and rapid motion. In this

proposed work we made a survey on different denoising

filters and concluded which works better among all..

Keywords:

Image Processing, Video Processing, Denoising and

Filters

I.INTRODUCTION

Image processing is a process to convert an image into

digital form and perform some operations on it, in order

to get an enhanced image or to extract some useful

information from it. Some type of signal allowance in

which input is image, like video frame or photograph

and output may be image or characteristics associated

with that image. Generally Image Processing system

includes treating images as two dimensional signals while

applying already set signal processing methods to them

[1].

Image Processing forms research area in engineering and

computer science disciplines too.

Image processing includes the following three steps.

Importing the image with the help of optical scanner or

by digital photography[1].

Data compression and image enhancement and spotting

patterns are analyzed and manipulating the image that

are not to human eyes like satellite photographs [1].

Output is the last stage in which result can be altered

image or report that is based on image analysis[1].

II. Purpose of Image Processing:

The purpose of image processing is divided into 5 groups.

They are:

Visualization - The objects are observed which are not

visible [1].

Image sharpening and restoration - To generate a

better image [1].1 Research Scholar, Karpagam University, Coimbatore,TamilNadu, India [email protected]. Head Dept of Software Systems, KarpagamUniversity, Coimbatore, TamilNadu, [email protected]

226


Image retrieval - Seek for the image of interest [1].

Measurement of pattern – Measures various objects

in an image [1].

Image Recognition – Distinguish the objects in an

image.

Digital Processing techniques help in manipulation of

the digital images by using computers. As raw data from

imaging sensors from satellite platform contains

deficiencies. To get over such flaws and to get originality

of information, it has to undergo various phases of

processing. The three general phases that all types of

data have to undergo while using digital technique are

Pre- processing, enhancement and display, informationextraction [1].

III. APPLICATIONS

Intelligent Transportation Systems – This technique

can be used in Automatic number plate recognition and

Traffic sign recognition [1].

Remote Sensing – For this application, sensors capture

the pictures of the earth’s surface in remote sensing

satellites or multi – spectral scanner which is mounted

on an aircraft. These pictures are processed by

transmitting it to the Earth station. Techniques used to

interpret the objects and regions are used in flood control,

city planning, resource mobilization, agricultural

production monitoring, etc [1].

Moving object tracking – This application enables to

measure motion parameters and acquire visual record of

the moving object. The different types of approach to

track an object are:

Motion based tracking

Recognition based tracking

Defense surveillance – Aerial surveillance methods are

used to continuously keep an eye on the land and

oceans. This application is also used to locate the types

and formation of naval vessels of the ocean surface. The

important duty is to divide the various objects present in

the water body part of the image. The different parameters

such as length, breadth, area, perimeter, compactness are

set up to classify each of divided objects. It is important

to recognize the distribution of these objects in different

directions that are east, west, north, south, northeast,

northwest, southeast and south west to explain all

possible formations of the vessels. We can interpret the

entire oceanic scenario from the spatial distribution of

these objects [1].

Biomedical Imaging techniques – For medical

diagnosis, different types of imaging tools such as X-

ray, Ultrasound, computer aided tomography (CT) etc

are used. The diagrams of X- ray, MRI, and computer

aided tomography (CT) are given below.

Some of the applications of Biomedical imaging

applications are as follows:

Heart disease identification– The important diagnostic

features such as size of the heart and its shape are

required to know in order to classify the heart diseases. To

improve the diagnosis of heart diseases, image analysis

techniques are employed to radiographic images [1].

Lung disease identification – In X- rays, the regions

that appear dark contain air while region that appears

lighter are solid tissues. Bones are more radio opaque

than tissues. The ribs, the heart, thoracic spine, and the


227

diaphragm that separates the chest cavity from the

abdominal cavity are clearly seen on the X-ray film [1].

Digital mammograms – This is used to detect the breast

tumors. Mammograms can be analyzed using Image

processing techniques such as segmentation, shape

analysis, contrast enhancement, feature extraction, etc

[1].

Automatic Visual Inspection System – This application

improves the quality and productivity of the product in

the industries.

Automatic inspection of incandescent lamp filaments –

This involves examination of the bulb manufacturing

process. Due to no uniformity in the pitch of the wiring

in the lamp, the filament of the bulb gets fused within a

short duration. In this application, a binary image slice

of the filament is created from which the silhouette of

the filament is fabricated. Silhouettes are analyzed to

recognize the non uniformity in the pitch of the wiring

in the lamp. This system is being used by the General

Electric Corporation[1].

Automatic surface inspection systems – In metal

industries it is essential to detect the flaws on the surfaces.

For instance, it is essential to detect any kind of aberration

on the rolled metal surface in the hot or cold rolling mills

in a steel plant. Image processing techniques such as

texture identification, edge detection, fractal analysis etc

are used for the detection [1].

Faulty component identification – This application

identifies the faulty components in electronic or

electromechanical systems. Higher amount of thermal

energy is generated by these faulty components. The

Infra-red images are produced from the distribution of

thermal energies in the assembly. The faulty components

can be identified by analyzing the Infra-red images [1].

IV. VIDEO PROCESSING

Video processing uses hardware, software, and

combinations of the two for suppression the images and

sound recorded in video files. General algorithms in the

processing software and the peripheral equipment allow

the user to perform editing functions using various filters.

The most wanted things can be shaped by suppressed

frame by frame or in larger batches.

Video files are obtained from the recording device using

a universal standard bus (USB) cable or fire wire

attachment. The records are then full into a computer

software program or peripheral device. Most computers

used personal come with software that allows the user to

accumulate images and videos, change images, and create

videos on a incomplete level. Storyboards allows the

adding together of audio files and the alteration of chart

images, transitions, and audio files, which, together,

determine the overall length of the video. Videographers,

electrical engineers, and computer science professionals

use programs that are capable of a wider range of

functions. Signal processing frequently involves

applying a combination of pre filters, intra filters, and

post filters [2].

Video Processing used in Research Areas:

Now-a-days many researches focuses on applying image/

video processing techniques, like recognition of pattern

and various signal processing techniques, to work out

definite problems in real-world applications.

In an application is the computerized age-assessment of

skeletal babies development, children and young adults,

using X-rays of the human hand. A mismatch of the

skeletal age with the biological age indicates growth

abnormalities, and thus serves as good diagnostic tool

for a number of diseases. New methods for matching

228


the patterns were developed to create algorithms for the

evaluation of the skeletal age of the patient.

Though the image processing tribulations carry over to

video processing, but exaggerated by the larger data sizes

and the new dimension of time and motion. We have

developed video segmentation techniques will open the

door to some type of divide-&-conquer advance to many

of the video processing problems.

V. VIDEO DENOISING:

Video denoising is the process of extracting the noise

from a video signal. Video denoising methods can be

divided into spatial, temporal and spatio-temporal. Spatial

video denoising method is adopted in this paper. In this

method image noise reduction is applied to each frame

of a video individually. Images are frequently corrupted

by noises. Noises occur during image capture,

transmission, etc. [3] The most important process in video

processing is Noise removal. In this the results of the

noise removal have a strong control on the quality of the

video processing technique. Various techniques for noise

removal are well expanded in color image processing.

The nature of the noise removal problem depends on the

kind of the noise corrupting the image. In the field of

image noise reduction several linear and non linear

filtering methods have been projected. Linear filters are

not able to efficiently eradicate impulse noise as they

have a propensity to blur the edges of an image. On the

other hand non linear filters are suitable for dealing with

impulse noise. Several non linear filters based on

Classical and fuzzy techniques have emerged in the past

few years. For example most classical filters that

confiscate concurrently blur the edges, while fuzzy filters

have the ability to merge edge preservation and

smoothing. Evaluated to other non linear techniques,

fuzzy filters are able to symbolize acquaintance in an

understandable way. In this paper we present results for

different filtering techniques and we evaluated the results

for these technique

Image sequence restoration has been progressively

achieving importance with the increasing prevalence of

visual digital media. The demand for content increases

the pressure on archives to automate their restoration

activities for preservation of the cultural heritage that

they hold. The goal of this paper is to find a pre-

processing method based on noise removal.

Type of Denoising:

• Spatial video denoising methods, in this noise in

image is reducing by applying to each frame

individually.

• Temporal in this video denoising method the noise

between frames is reduced. [4]Motion compensation

may be used to avoid ghosting artifacts when

blending together pixels from several frames.

• Spatial-Temporal is another video denoising methods

which is a combination of spatial and temporal

denoising.

TYPES OF NOISE:

• Amplifier noise (Gaussian noise)

• Salt-and-pepper Noise

• Periodic Noise

Amplifier noise (Gaussian noise):The Amplifier noise

is also called as additive or Gaussian or independent at


229

each pixel and independent of the signal intensity, caused

primarily by Johnson–Nyquist noise (thermal noise),

including that which comes from the reset noise of

capacitors [3]. In color cameras more amplification is

used in the blue color channel other than in the green or

red channel, there can be more noise in the blue channel

[5].

Salt-and-pepper Noise: Salt-and-pepper noise or spike

noise is also called as Fat-tail distributed or “impulsive”

noise. The images of salt-and-pepper noise will have dark

pixels in bright regions and bright pixels in dark regions.

[4]This type of noise can be caused by the errors of analog-

to-digital converter, bit errors in transmission, etc. This

can be eliminated in large part by using dark frame

subtraction and by interpolating around dark/bright

pixels[6].

Periodic Noise: The signal-to-periodic noise ratio is the

ratio in decibels, of the nominal amplitude of the

luminance signal (100 IRE units) to the peak-to-peak

amplitude of the noise. Different performance objectives

are sometimes specified for periodic noise (single

frequency) between 1 kHz and the upper limit of the

video frequency band and the power supply hum,

including low-order harmonics.

Types of Filters

1. Average Filter

2. Median Filter

3. Wiener Filter

4. Rank Order Filter

5. Gaussian Filter

6. Non-Linear Filter

7. Outlier Filter

Average filter: Mean filter or average filter is a

windowed filter of linear class, which smoothens the

signal (image). The filter works as low-pass one. The

basic idea behind filter is for any element of the signal

(image) take an average across its neighbourhood.

Median filter: Median filtering is a nonlinear process is

used to reduce impulsive or salt-and-pepper noise. It is

also useful in preserving edges in an image while

reducing random noise. Impulsive or salt-and pepper

noise can occur due to a random bit error in a

communication channel. In a median filter, a window

slides along the image, and the median intensity value of

the pixels within the window becomes the output

intensity of the pixel being processed. For example,

suppose the pixel values within a window are 5, 6, 55,

10 and 15, and the pixel being processed has a value of

55. The output of the median filter and the current pixel

location is 10, which is the median of the five values [7].

Wiener Filter: A restoration technique for deconvolution

is inverse filtering, i.e., when the image is blurred by a

known low pass filter, it is possible to recover the image

by inverse filtering or generalized inverse filtering.

Additive noises are sensitive to inverse filtering. The

approaches for reducing degradation for restoring the

algorithms are developed for each type of degradation

and simply combine them. The Wiener filtering is

executed by an optimal tradeoff between inverse filtering

and noise smoothing. The mean square error is optimized

in wiener filtering. It minimizes the overall mean square

error in the process of inverse filtering and noise

smoothing [8]. The original images are linear estimated

to wiener filtering. [9]The orthogonally theory of implies

that the Wiener filter in Fourier domain can be expressed

as follows:

230


�

),(),(),(

),(),(),(

2121

2

21

2121*

21ffSffSffH

ffSffHffW

nnxx

xx

+=

Where ),(),,( 2121 ffSffS nnxx power spectra of the

original image and the additive are respectively noise,

and ),( 21 ffH is the blurring filter. The Wiener filter

is easy to separate two parts, an inverse filtering part

and a noise smoothing part. It not only performs the

disconsolation by inverse filtering (high pass filtering)

but also removes the noise with a compression operation

(low pass filtering).

Rank Order Filter: Rank order filters have a certain size,

but do not have any matrix values or a gain factor. A

rank order filter of size 3x3 for example, examines 9

pixel values of the input map at a time, sorts the values

from small to large, and selects for the output value that

value which is encountered at a certain rank order

number. So one value of the pixel values examined

becomes the output value, without any calculation

performed on the values itself. When a threshold is set,

the value of the centre pixel will only be replaced with

the new value if the difference between the original and

new value is smaller than or equal to the threshold.

Gaussian Filter: Gaussian filter is a filter which has the

impulse response is the Gaussian function. When

minimizing the rise and fall time these filters are

developed to overshoot. The minimum possible group

delay is connected closely with the Gaussian Filter. A

Gaussian filter is modified by the input signal of the

convolution with a Gaussian function mathematically;

this transformation is also known as the Weierstrass

transform.

The one-dimensional Gaussian filter has an impulse

response given by

�

2..)( xaea

xg −

Π=

or with the standard deviation as parameter

2

2

2...2

1)( σ

σ

x

exg−

Π=

In two dimensions, it is the product of two such

Gaussians, one per direction:

2

22

222

1),( σ

σ

yx

eyxg+−

Π=

Where x is the distance from the origin in the horizontal

axis, y is the distance from the origin in the vertical axis,

and ó is the standard deviation of the Gaussian

distribution[10].

Non-Linear Filters: A nonlinear filter is a signal-

processing device whose output is not a linear function

of its input. Terminology concerning the filtering problem

may refer to the time domain (state space) showing of

the signal or to the frequency domain representation of

the signal. When referring to filters with adjectives such

as “band pass, high pass, and low pass” one has in mind

the frequency domain. When resorting to terms like

“additive noise”, one has in mind the time domain, since

the noise that is to be added to the signal is added in the

state space representation of the signal. The state space

representation is more general and is used for the

advanced formulation of the filtering problem as a

mathematical problem in probability and statistics of

stochastic processes [11].

Outlier Filter: An adaptive filter is a filter that self-

adjusts its transfer function according to an optimization


231

algorithm driven by an error signal. Most adaptive filters

are digital filters has more complexity of algorithms in

optimizations. A non-adaptive filter has a static transfer

function. An error signal that has to be refined to its

transfer function to match the changing parameters is

used in adaptive filters. [12]The adaptive process is

involved the use of a cost function, which is also used as

a criterion for optimum performance of the filter, which

determines how to modify the filter algorithm to transfer

its function to minimize the cost on the next iteration.

Adaptive filters have become much more common and

are now often used in devices such as mobile phones

and other communication devices, camcorders and digital

cameras, and medical monitoring equipment.

VI. EVALUATION MEASURES FOR FILTERS

Mean Square Error : In the image coding and computer

vision literature, the most frequently used measures are

deviations between the original and coded images of

which the mean square error (MSE) or signal to noise

ratio (SNR) being the most common measures. The

reasons for these metrics widespread popularity are their

mathematical tractability and the fact that it is often

straightforward to design systems that minimize the MSE

but cannot capture the artifacts like blur or blocking

artifacts [13]. The effectiveness of the coder is optimized

by having the minimum MSE at a particular compression

[David Bethel (1997)] and MSE is computed using

Eq.(2),

MSE=2

1 1

' )),(),((1

jifjifMN

M

i

N

i∑∑

= =− (1)

Peak Signal – To – Noise-Ratio: Larger SNR and PSNR

indicate a smaller difference between the original

(without noise) and reconstructed image. This is the most

widely used objective image quality/ distortion measure.

The main advantage of this measure is ease of

computation but it does not reflect perceptual quality.

An important property of PSNR is that a slight spatial

shift of an image can cause a large numerical distortion

but no visual distortion and conversely a small average

distortion can result in a damaging visual artifact, if all

the error is concentrated in a small important region. This

metric neglects global and composite errors PSNR is

calculated using Eq. (2),

PSNR= dBRMSE

N

10log20 (2)

Average Difference: A lower value of Average Difference

(AD) gives a “cleaner” image as more noise is reduced

and it is computed using Eq. (3).

Average Difference (AD)

= [ ]∑∑= =

−M

i

N

j

jifjifMN 1 1

' ),().(1

(3)

Maximum Difference: Maximum difference (MD) is

calculated by using Eq. (4) and it has a good correlation

with MOS for all tested compression techniques so it is

preferred as a very simple measure by referring the

measure of the compressed picture quality in different

compression systems (Marta Mrak et al 2003).

Maximum Difference (MD) =

)),(),(( ' jifjifMax − (4)

Normalized Correlation: The proximity between two

digital images can also be quantified in terms of

correlation function. These measures measure the

similarity between two images, hence in this sense they

232


are complementary to the difference based measures. All

the correlation based measures tend to 1, as the difference

between two images tend to zero. As difference measure

and correlation measures complement each other,

minimizing Distance measures are maximizing

correlation measure [Thomas Kratochevil et al (2005)]

and Normalised Correlation is given by Eq. (5)

Normalised Correlation (NK)

=

[ ]

∑∑

∑∑

= =

= =M

i

N

j

M

i

N

j

jif

jifjif

1 1

2

1 1

'

),(

),()..(

(5)

F. Mean Absolute Error:

Mean Absolute Error (MAE)

= )),(),((1

1 1

'∑∑

= =

−M

i

N

j

jifjifMN (6)

MAE is calculated used Eq. (6) and large value of MAE

means that the image is of poor quality [Ratchakit

Sakuldee et al (2007)].

Normalized Absolute Error

Normalised absolute error computed by Eq. (7) is a

measure of how far is the decompressed image from the

original image with the value of zero being the perfect

fit [Harker (1987)]. Large value of NAE indicates poor

quality of the image [Ratchakit Sakuldee et al (2007)].

Normalized Absolute Error (NAE)

=

[ ]

∑∑

∑∑

= =

= =M

i

N

M

i

N

j

jif

jifjif

1 1

1 1

'

),(

),().,(

(7)

Structural Correlation/Content:

A recognizable concept in image processing, that

estimates the likeness of the structure of two signals is

Correlated,. The measure are compared with the total

weight of an original signal to that of a coded or given.

This measure is also called as structural content. The

Structural content is given by Eq. (8) and if it is spread

at 1, then the decompressed image is of better quality

and large value of SC means that the image is of poor

quality [Ratchakit Sakuldee et al (2007)].

Structural Correlation/Content (SC)

=

[ ]

∑∑

∑∑

= =

= =M

i

N

j

M

i

N

j

jif

jif

1 1

2'

1 1

2

),(

),(

(8)


233

Table: 1 Evaluation Measures for Gaussian Noise

From the above Table: 1 for Gaussian Noise, the video for Wiener Filter out performs. The remaining

filters in the video are denoising

Original Data Noise Data Median Filter Winer Filter

Rank Order Average Data Gaussian Filter NL Filter

Outliner Filter

Fig: 1 Screen Shot for Gaussian Noise

Gaussian Noise

MSE PSNR MNCC AD SC MD NAE

Average Data 435.7165 21.9992 0.9616 1.8718 1.0534 161.5750 0.0962

Median Filter 256.0075 24.2621 0.9798 0.8614 1.0256 159.7333 0.0828

Wiener Filter 175.3028 25.9706 0.9798 0.8615 1.0308 85.9333 0.0678

Rank Order 202.2178 25.2960 0.9886 0.3427 1.0107 116.2333 0.0831

Gaussian Filter 226.5426 24.8181 0.9789 0.8713 1.0294 116.4417 0.0796

Non-Linear Filter 256.0075 24.2855 0.9798 0.8614 1.0256 159.7333 0.0828

Outlier Filter 345.1225 22.9543 0.9964 0.1580 0.9866 150.0750 0.1178

VII. RESULTS AND DISCUSSIONS :

234


Table: 2 Evaluation Measures for Salt and Pepper Noise

Salt & Pepper Noise


Average Data 512.6008 21.2961 0.9582 104849 1.0558 162.2667 0.1181

Median Filter 255.9669 24.0606 0.9958 0.4679 0.9930 189.8667 0.0272

Wiener Filter 421.1141 22.1408 0.9673 1.0192 1.0417 208.1167 0.1010

Rank Order 187.3607 25.6188 0.9892 0.3037 1.0104 215.5667 0.0378

Gaussian Filter 435.6573 21.9709 0.9746 0.4647 1.0253 141.7417 0.1169

Non-Linear Filter 230.6177 24.7355 0.9802 0.8514 1.0265 198.0917 0.0501

Outlier Filter 257.5257 24.2373 0.9956 0.4818 0.9933 189.3417 0.0273

From the above Table: 2 for Salt and Pepper Noise, the video for Median Filter and Rank Order Filter performs

well. The remaining filters in the video are denoising.

Original Image Noise Data Median Filter Winer Rank Order

Average Data Gaussian NLFilter Outlier Filter

Fig: 2 Screen Shot for Salt & Pepper Noise


235

Periodic Noise


Average Data 612.4158 16.7406 0.9381 0.6814 1.0452 171.7083 0.2343

2D Filter 150.0259 20.2936 0.9654 0.5665 1.0240 125.9833 0.1002

Wiener Filter 512.355 16.9289 0.9314 0.0260 1.0636 176.4833 0.2264

Rank Order 235.313 9.5164 0.9632 0.4357 0.7345 252.5833 0.5757

Gaussian Filter 356.218 17.1290 0.9486 1.6681 1.0302 172.0333 0.2260

Non-Linear Filter

235.51 14.4934 0.9622 2.5876 0.9885 249.2000 0.3010

Outlier Filter 178.86 15.6354 0.9526 0.7225 0.9917 225 0.2560

From the above Table: 3 for Periodic Noise, the video for 2D Filters performs well. The remaining filters in thevideo are denoising.

Original Data Noise Data Wiener Data Average Data Gaussian Data

NLFilter Median Filter 2D Filter Outlier Filter

Fig: 3 Screen Shot for Periodic Noise

Table: 3 Evaluation Measures for Periodic Noise

VIII. CONCLUSION

In this article, we conversed about different filtering

techniques for removing noises in video. Furthermore,

we presented and compared results for these filtering

techniques. From the results obtained we conclude that

with three different noises salt and pepper noise,

Gaussian noise and periodic noise applied for denoising

of the spatial video produces variant results over different

filtered techniques. From the results obtained using

various filtering techniques it is observed that for salt

and pepper noise median and rank order filter works

better than other techniques. In case of Gaussian noise

Weiner and rank order filter works fine. For Periodic

noise 2D filter works better than other filters.

236


REFERENCES

[1] http://www.engineersgarage.com/articles/image-

processing-tutorial-applications.

[2] http:/ /www.wisegeek.com/what-is-video-

processing.htm.

[3] C. Mythili,Dr. V. Kavitha, .”Efficient Technique

for Color Image Noise Reduction”, T h e R e s e a

r c h B u l l e t i n o f J o r d a n A C M, V o l. I I (I

I I ) P a g e | 4 1-44.

[4] http://en.wikipedia.org/wiki/Video_denoising

[5] http://en.wikipedia.org/wiki/Image_noise.

[6] Amit Yerpude, Dr. Sipi Dubey, Amit Yerpude, Dr.

Sipi Dubey, “Robust Method for Noisy Image

Segmentation”, IJCSET |February 2012| Vol 2,

Issue 2,891-895.

[7] http://nptel.iitm.ac.in/courses/Webcourse-

contents/IIT-KANPUR/Digi_Img_Pro/chapter_8/

8_16.html

[8] http://www.owlnet.rice.edu/~elec539/Projects 99/

BACH/proj2/wiener.html.

[9] Disha Sharma and Gagandeep Jindal, “Computer

Aided Diagnosis System for Detection of Lung

Cancer in CT Scan Images”, International Journal

of Computer and Electrical Engineering, Vol. 3,

No. 5, October 2011.

[10] http://en.wikipedia.org/wiki/Gaussian_filter.

[11] http://en.wikipedia.org/wiki/Nonlinear_filter

[12] http://en.wikipedia.org/wiki/Adaptive_filter.

[13] Sumathi Poobal, G.Ravindran,” The Performance

of Fractal Image Compression On Different

Imaging Modalities Using Objective Quality

Measures”, International Journal of Engineering

Science and Technology (IJEST), and Vol. 3 No.

1 Jan 2011, PP: 525-530.


M. Hemalatha completed MCA

MPhil., PhD in Computer

Science and currently working

as an Assistant Professor and

Head, Department of software

systems in Karpagam

University. More than ten years

of Experience in teaching and published more than fifty

papers in International Journals and also presented

seventy papers in various National conferences and one

International Conferences Area of research is Data

mining, Software Engineering, bioinformatics, Neural

Network. She is also reviewer in several National and

International journals.

R.Revathi was born and brought

up Trichy She received her

Bachelor of Computer Science

from Bharathidasan University

and Master degree in Information

Technology from Bharathidhasan

University. She completed her

M.Phil from Bharathidasan University. Currently

pursuing Ph.D., in computer science at Karpagam

University under the guidance of Dr.M.Hemalatha Head,

Dept of Software System, Karpagam University, and

Coimbatore. Area of Interest is Image Processing and

Networks. Area of Research is Image Processing.

Karpagam Journal of Computer Sciencewww.karpagameducation.com

Volume : 06 Issue : 04 May - June 2012

CONTENTS PAGE NO.

1. Acoustic Signature Recognition of Moving Vehicles Using Elman 183Neural NetworkPaulraj M P, Abdul Hamid Adom, Hema C R, Sathishkumar Sundararaj

2. Fuzzy MPPT Based Voltage Regulation on Photovoltaic Power Supply 190System For Continuously Varying Illumination ConditionA. Durgadevi, S. Arulselvi

3. Design and Development of a Gripper System for an Indoor 196Service RobotHema C. R., Vivian Tang Sui Lot, Paulraj M.P.

4. Power Adaption Routing Protocol For Realtime Applications 204in Wireless Sensor NetworksR.Prema, R.Rangarajan

5. Path Planning in AI (Artificial Intelligence) 210T.C. Manjunath, G.V. Jayaramaiah

6. Efficient Object Detection and Classification using Hybrid ELM 217


N.V. Balaji, M. Punithavalli

7. Empirical Analysis of Denoising Techniques in Video Processing 225

R.Revathi, M.Hemalatha

acoustic signature recognition of moving vehicles … · 183 acoustic signature recognition of...

Documents