Hindawi Publishing Corporation

e Scientific World Journal

Volume 2014, Article ID 393958, 10 pages
http://dx.doi.org/10.1155/2014/393958

Research Article
An Improved Real Time Image Detection System for
Elephant Intrusion along the Forest Border Areas

S. J. Sugumar1 and R. Jayaparvathy2

1
Department of Electrical and Electronics Engineering, Coimbatore Institute of Technology, Avinashi Road,
Civil Aerodrome PO, Coimbatore, Tamil Nadu 641 014, India
2
Department of Electronics and Communication Engineering, SSN College of Engineering, Kalavakkam,
Old Mahabalipuram Road, Chennai, Tamil Nadu 603 110, India

Correspondence should be addressed to R. Jayaparvathy; jayaparvathy14@gmail.com

Received 12 August 2013; Accepted 21 October 2013; Published 20 January 2014

Academic Editors: S. Bourennane and J. Marot

Copyright © 2014 S. J. Sugumar and R. Jayaparvathy. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

Human-elephant conflict is a major problem leading to crop damage, human death and injuries caused by elephants, and elephants
being killed by humans. In this paper, we propose an automated unsupervised elephant image detection system (EIDS) as a solution
to human-elephant conflict in the context of elephant conservation. The elephant’s image is captured in the forest border areas and
is sent to a base station via an RF network. The received image is decomposed using Haar wavelet to obtain multilevel wavelet
coefficients, with which we perform image feature extraction and similarity match between the elephant query image and the
database image using image vision algorithms. A GSM message is sent to the forest officials indicating that an elephant has been
detected in the forest border and is approaching human habitat. We propose an optimized distance metric to improve the image
retrieval time from the database. We compare the optimized distance metric with the popular Euclidean and Manhattan distance
methods. The proposed optimized distance metric retrieves more images with lesser retrieval time than the other distance metrics
which makes the optimized distance method more efficient and reliable.

1. Introduction their size and nature of movement. The time to recover from
the danger is negligible; hence, the loss due to destruction in
The Asian elephant (Elephas maximus) is highly threatened the farms is more. The elephants are also subject to attack by
by habitat fragmentation, habitat loss, and human-elephant humans resulting in danger to the life of elephants. According
conflict. India hosts 60% of Asian elephant population, to the authors in [3], poaching for ivory had indeed become
nearly two-thirds of the elephant population lives either close a threat with 100–150 tuskers being lost annually to illegal
to or within human-dominated landscapes. Southern India killings.
harbors half of India’s elephant population containing about Human-elephant conflict (HEC) is a key example of
6300 elephants [1]. The increase in human population in India the growing competition between people and wildlife for
propelled by agricultural and industrial growth has led to space and resources throughout Africa and Asia. This study
the conversion of the forest lands into human settlements. explores the correlation of reported HEC incidents within
Due to this, the wild elephant and other animal populations 58 villages between 80 km from the boundary of Kallar to
face acute shortage of resources such as water and food, Walayar, Coimbatore, Tamil Nadu, India. Habitat loss and
making them move often into the human habitat. Hence, fragmentation is the biggest threat to the continuing survival
there has been severe man-elephant conflict. The conflict of Asian elephants in this region. In addition to food crops,
has been on the rise in the forest border areas with herds forests are being logged for their timber or cleared to make
of wild pachyderms straying into human habitation [2]. The space for cash crop plantations such as rubber, tea, and palm.
surveillance and tracking of these herds are difficult due to As the human population has inexorably risen, the forest
2 The Scientific World Journal

wild lands in which elephants live have been disappearing. [5], Fernando et al. discussed solar fencing to avoid elephant-
Human-elephant conflict is on the rise and it is a battle that human conflict. In [6], King et al. presented the concept of
the elephant is losing. As elephant habitat diminishes, the using beehives to mitigate elephant crop depredation. In [7],
elephants are pushed into increasingly smaller areas. This Loarie et al. discussed about the role of the artificial water
increases the population density to beyond sustainable levels sources which allow elephants to reside in forests during
and food availability grows short. The shortage of fodder has dry seasons. In [8], the authors discussed the potential use
a negative impact on rates of reproduction; hence, normal of satellite technology for conflict mitigation. The elephants
birth rates begin to decrease. The serious consequence of the tagged with radio collars react violently and damage it and
shortage of wild food leads to a corresponding increase of even the elephants die. In [9], Venter and Hanekom proposed
crop raiding and incidents of human-elephant conflict [3]. the possibility of using the elephant-elephant communication
Human-elephant conflict is a rapidly expanding area of (elephant rumbles) to detect the presence of a herd of
research, with conservationists working hard to understand elephants in close proximity, In this work, the authors have
the circumstances under which tensions are the highest recorded the low frequency infrasound pattern, but they
between humans and elephants. A number of factors con- do not compare with that of other animals to confirm an
tribute to such conflicts, including population density of elephant occurrence. In [10], Vermeulen et al. proposed
humans, elephant habitat structure, weather, time of year, unmanned aircraft system to survey elephants, in which
and animal life [4]. A study made in the region of interest the elephant images are acquired at a height of 100 m but
shows that elephants move into human habitation due to the small flight time and being expensive do not make it
many reasons. viable. In [11], Dabarera and Rodrigo proposed appearance
based recognition algorithms for identification of elephants.
(i) Fences and trenches compromised by people who Given the frontal face image of an elephant, the system
need access to forests. searches the individual elephant using vision algorithms and
(ii) Farm lands may funnel them to unprotected adjacent gives the result as, already identified elephant, or as a new
villages. identification. In [12], Ardovini et al. present an elephant
photo identification system based on the shape comparison of
(iii) Badly planned barriers that do not take elephant the nicks characterizing the elephant’s ears. In [13], Goswami
behavior into consideration. et al. addressed identifying elephants from photographs,
(iv) Denying elephant access to a critical water source or and comparing resultant capture recapture-based population
foraging area. parameter estimates using supervised visual identification of
individual variations in tusk, ear fold and lobe shape. The
(v) Human activities create abundant secondary vegeta-
authors show that this is a reliable technique for individ-
tion that brings elephants closer to human settle-
ual identification and subsequent estimation of population
ments.
parameters. But in real time, the capture of elephant’s front
(vi) Artificially maintained water sources attract ele- image is not possible.
phants during drought. It is easier to chase elephants before they enter fields and
(vii) Traditional migration routes severed by human inter- therefore most damage can be averted [3]. Guarding from
vention (e.g., canals, power installations, and cattle watch towers, patrolling, and trip wire alarms provide farm-
fences). ers with advance warning of approaching elephants. Once
the animals are detected, active crop guarding devices using
The obvious conclusion to be drawn is that there is no light and noise are deployed to chase them away. An early
single cause or explanation to account for human-elephant warning system to minimize the human-elephant conflict in
conflict; situations are circumstantial and complex. Rather, the forest border areas is proposed in this paper. The system
elephants and agriculture mix in numerous ways with varying helps mitigate such conflicts in two ways:
consequences. Human population growth and land occu-
pation for settlement may heighten conflict with elephants. (i) providing warning to people about the anticipated
However, it is generally the borders of forests that are the focal entry of elephants into human habitation;
points of conflicts. Minimizing human-elephant conflict to
reduce the risk of life of both human beings and elephants is
(ii) providing advance information to the authorities to
of utmost importance. Elephant conservation issues can be
take action to chase the pachyderms back to the forest.
divided into two distinct categories:

(1) activities that affect elephants directly such as hunt- An early warning system to minimize the human-ele-
ing/poaching and capture; phant conflict in the forest border areas using image pro-
(2) developmental activities and human activities leading cessing is proposed in this paper. The system helps to detect
either to the loss of elephant habitat or its qualitative the elephants even in the presence of other wild animals like
degradation. Bison, Tiger, and Deer, and so forth. The system also identifies
the elephants coming in groups. The reliability of elephant
Many methods are followed to avoid HEC. Construction detection is tested and the time to detect the elephant images
of elephant proof trenches is being done all over the world. In is optimized with the proposed optimized distance metric.
The Scientific World Journal 3

Figure 1: Map showing study area of the Coimbatore district in Figure 2: Corridors used by Elephants in the forest border area of
Tamil Nadu. Coimbatore.

2. Study Area
to crops and human lives. Each year, nearly 400 people and
Studies were made in the Coimbatore Forest Division, Tamil 100 elephants are killed in conflict related instances in India,
Nadu, India, as shown in Figure 1. The data were collected and nearly 500,000 families are affected by crop damage.
from the Coimbatore Forest Department website and inter- Several reasons including habitat fragmentation, degradation
views with village people affected by elephants and relevant of habitat quality, loss of forest cover, laxity in management
literature. Coimbatore district is richly endowed with hills, of physical barriers, and other causes have been cited for the
forests, rivers, and wildlife. Geographical area of the district human-elephant conflict (HEC) in the country.
is 74,433.72 sq. km with a forest area of 693.48 sq. km (9.33%). The human habitats bordering the forests around Coim-
The forest of Coimbatore district is divided into two divisions. batore city in south western India are seeing severe human-
South of Palghat Gap lies in the Anamalai Wildlife elephant conflict as the expanding human population pro-
Sanctuary, which has been designated as a Tiger Reserve in pelled by industrial and agricultural growth is increasingly
2008. North of Palghat lies in the Coimbatore forest division. fragmenting elephant habitat in this landscape. The number
This division is bounded in the north and northwest by of incidents of elephants straying into farm lands was 680 in
Sathyamangalam, Erode, Nilgiris North, and Nilgiris South 2011, 844 in 2010, and 560 in 2009. The number of people
forest divisions and in the west and southwest by Palghat killed in elephant attacks in Coimbatore was 13 in 2012, 8 in
forest division of Kerala. Coimbatore Forest Division is 2011, 15 in 2010, and 11 in 2009. The number of elephants killed
spread over 693.48 sq. km in six ranges, of which 400 sq. km by such conflicts was 4 in 2012, 1 in 2011, 1 in 2010 and 2 in
is conflict prone. The division has 58 villages and 315 km of 2009 as shown in Figure 3 (Courtesy: the Hindu, Coimbatore,
forest boundary. February 5, 2013).
Six elephant corridors within the Coimbatore forest divi- In this region, train hit accidents occur frequently when
sion are shown in Figure 2. Namely, Jaccanari-Vedar Colony the elephants try to pass the rail track Walayar, Coimbatore,
(Corridor 1) in which the length of the corridor is about railway section which is on the forest border area. Thus it has
12 km and width ranges from 0.2 to 1.5 km. In Kallar-Jaccanari resulted in the death of 20 elephants in the last five years.
(Corridor 2), the length of the corridor is 7 km and width However, it is generally the edges of forest that are to be
ranges from 0.2 to 1.5 km. In Kallar-Nellithurai (Corridor 3), protected as those are the focal points of conflict.
the length of the corridor is 10.8 km and width ranges from Humans go into the forest to graze cattle in day time
0.25 to 3 km. In Anaikatti-Veerapandi (Corridor 4), the length and guard crops at night and therefore run a higher risk of
of the corridor is 21 km and effective width ranges from 0.1 to being killed by elephants. They also try to ride elephants by
1.5 km. In Maruthamalai-Thanikandy (Corridor 5), the length throwing stones and sticks for which the elephants react hard
of the corridor is 13 km and effective width ranges from 0.4 to and even kill the humans. Elephants that wait near villages for
1.5 km. And in Kalkothi-Walayar (Corridor 6), the length of nightfall to eat crops have also been known to kill people [3].
the corridor is 21 kms the effective width ranges from 0.2 to During the period 1999–2011, there had been 1,822 incidents
0.9 km. of crop damage and 53 cases of property damage caused by
In total, there are 85 kms of elephant corridor in the elephants and the compensation disbursed was Rs. 2.19 crore.
district needs to be protected from human-elephant conflict.
The region is also a part of the crucial elephant corridor in 3. Material and Methods
this region [4] amounting to a total of 691–914 elephants
found in this region. The elephant is one of the most conflict- In our previous work [2], seismic geophones used as sensors
prone wildlife species in India, causing large-scale damage are buried under the ground to detect the movements of
4 The Scientific World Journal

16 different postures. Images of elephants in the Sadivayal ele-

phant camp in Coimbatore, South India, have been taken and
14
used in this work. The elephant database images are feature-
12
extracted using Haar wavelet technique and clustered into
groups by using 𝐾-means clustering. A similarity comparison
Number of incidents

10 is made by determining the number of significant coefficients

in common between the query signature and the signatures
8 of the database using 𝐹 Norm theory. The searched elephant
images are then arranged according to the similarity value
6 obtained in a decreasing order. If the matched images are
4
more than 5, an elephant detected message is sent through
the GSM to the mobile phone numbers stored in the system.
2 Once the procedure is complete, the system captures the
next image and performs the same steps to detect elephants.
0 The Haar wavelet decomposition of elephant image in RGB
2009 2010 2011 2012 color space is represented at multiple scales [14]. The Haar
Year wavelet decomposition is computed by iterating difference
Elephant death 𝑑𝑖 and average 𝑎𝑖 between odd and even samples 𝑠𝑖 of the
Human death elephant image. Averaging and differencing the elephant
image elements are done as follows:
Figure 3: Human and elephant loss in Coimbatore district.
𝑠𝑖 + 𝑠𝑖+1 𝑠𝑖 − 𝑠𝑖+1
𝑎𝑖 = , 𝑑𝑖 = . (1)
2 2
elephants in forest border areas. Elephants walking in the
sensing range of the geophones produce vibration which is If an elephant image data set 𝑆1 , 𝑆2 , . . . , 𝑆𝑁−1 contains 𝑁 ele-
converted to an electrical signal, processed in an embed- ments, there will be 𝑁/2 averages and 𝑁/2 wavelet coefficient
ded controller and an SMS is sent to the forest officials values [15]. The averages are stored in the upper half of the 𝑁
for necessary action. In such intrusion detection systems, element array and the difference coefficients are stored in the
there is a possibility of insufficient vibrations sensed by the lower half of the array. The averages become the input for the
geophones as a consequence of weather conditions like rain next step in the wavelet computation, for iteration 𝑖 + 1, 𝑁𝑖 =
and soil moisture. Due to this, it may miss the event, that 𝑁𝑖/2 . The recursive iterations continue until a single aver-
is, elephant movement. An image processing based approach age and a single difference are calculated [16]. The scaling and
is developed as a solution to the above said problem to wavelet values are represented by ℎ𝑖 and 𝑔𝑖 , respectively, and
identify an intruding elephant in human living areas. The are given in (2) and (3). The values of scaling coefficients are
elephant comes out of the forest through certain pockets to given as
enter into human living areas for getting food and water. ℎ0 = 0.5, ℎ1 = 0.5, (2)
The cameras mounted on towers or trees capture the image
of the intruding elephant which is sent to the base station and the values of wavelet coefficients are given as
using RF network. The received image is processed in a PC
at the base station and is compared with the stored database 𝑔0 = 0.5, 𝑔1 = −0.5. (3)
image of elephants. The snapshot from the video is taken
every 5 seconds and compared with the database image. The Haar transform is shown in matrix form as follows:
This image is also updated in database and added. On an ℎ0 ℎ1 0 0 ⋅⋅⋅
image match, an SMS is sent to the forest officials through 𝑔0 𝑔1 0 0 ⋅⋅⋅
the GSM transceiver connected with the PC. The hardware (4)
0 0 ℎ0 ℎ1 ⋅⋅⋅
setup consists of wireless camera, PIR sensor with signal 0 0 𝑔0 𝑔1 ⋅⋅⋅
control module, Atmega microcontroller, GSM module, and
the power supply. PIR motion sensor detects the movement The Haar transform for an eight element signal is shown in
and it switches the camera to the capture mode; the camera the following equation. Here, the signal is multiplied by the
captures the image over a 20-meter distance. The whole forward transform matrix 𝐴:
hardware setup is shielded with metal cast to protect from the
𝑎0 𝑠0
rain. The elephant pockets in the corridors are identified and
[ 𝑎1 ] [𝑠1 ]
these setups are installed to monitor the movement of these [ ] [ ]
[ 𝑎2 ] [𝑠2 ]
herds. [ ] [ ]
[ 𝑎3 ] [𝑠 ]
[𝑑 ] = 𝐴 ⋅ [𝑠3 ] , (5)
[ 0] [ 4]
[𝑑 ] [𝑠 ]
3.1. Elephant Image Detection System. Elephant Image Detec- [ 1] [ 5]
tion System (EIDS) algorithm is developed in this work. A [𝑑 ] [𝑠 ]
2 6
database of 114 images is created by capturing 2 elephants in [𝑑3 ] [𝑠7 ]
The Scientific World Journal 5

where 3.3. Proposed Optimized Distance Metric. In this paper, a

0.5 0.5 0 0 0 0 0 0 novel distance metric called optimized distance measure
[0.5 −0.5 0 0 0 0 0 0 ] integrated with 𝐾-means clustering algorithm to improve
[ ] retrieval time is proposed. We have used the distance metrics
[0 0 0.5 0.5 0 0 0 0 ]
[ ] in the work for (i) finding similarity between two images and
[0 0 0.5 −0.5 0 0 0 0 ]
𝐴=[ .
0 ]
(6) (ii) ordering a set of images based on their distances from
[0 0 0 0 0.5 0.5 0 ]
[0 0 0 0 0.5 −0.5 0 0 ]] a given image. In many image retrieval systems, Euclidean
[
[0 0 0 0 0 0 0.5 0.5 ] [18] and Manhattan [20] are the popular distance measure
[0 0 0 0 0 0 0 −0.5] algorithms used. We carried out a study on the above two
similarity measures and proposed a new distance method
Since the columns of the 𝐴 𝑖 ’s are orthogonal to each other, called optimized distance measure. The proposed method
each of these matrices is invertible with respect to 𝐴 𝑖 . The retrieved more images with faster retrieval rate than the
elephant database images are decomposed into multilevel other two methods. The Euclidean distance measures [18]
coefficients from −1 to −𝐽 levels. After decomposition, feature are suitable for the correlation between quantitative and
vectors for all the elephant images in the database are continuous variables and are not suitable for ordinal data and
obtained using 𝐹-norm theory [17] as given in (7) and (9). it is given as
Every image is considered as a square matrix. 𝐴 is a square
matrix and 𝐴 𝑖 is its 𝑖th-order submatrix. The 𝐹-norm of 𝐴 𝑖 is 2 2 2
given as 𝐷 = √ (𝑅𝑐 − 𝑅𝑔 ) + (𝐺𝑐 − 𝐺𝑔 ) + (𝐵𝑐 − 𝐵𝑔 ) , (12)
𝑖 𝑖 1/2
󵄩󵄩 󵄩󵄩 󵄨 󵄨2 where (𝑅𝑐 , 𝐺𝑐 , 𝐵𝑐 ) are centroids and (𝑅𝑔 , 𝐺𝑔 , 𝐵𝑔 ) are the
󵄩󵄩𝐴 𝑖 󵄩󵄩𝐹 = [ ∑ ∑ 󵄨󵄨󵄨𝑎𝐾𝐼 󵄨󵄨󵄨 ] . (7)
𝐾=1 𝐼=1 pixel points or data points. Most of the time is spent to
Let calculate the square root, so it is basically time consuming.
󵄩 󵄩 󵄩 󵄩 󵄩󵄩 󵄩󵄩 The Manhattan distance is the absolute sum of the horizontal
Δ𝐴 𝑖 = 󵄩󵄩󵄩𝐴 𝑖 󵄩󵄩󵄩𝐹 − 󵄩󵄩󵄩𝐴 𝑖−1 󵄩󵄩󵄩𝐹 , 󵄩󵄩𝐴 0 󵄩󵄩𝐹 = 0. (8) and vertical components of the image data matrix. This
The feature vector of 𝐴 is defined as is essentially a consequence of being forced to adhere to
single-axis movement; one cannot move diagonally in more
𝑉𝐴𝐹 = {Δ𝐴 1 , Δ𝐴 2 , . . . , Δ𝐴 𝑛 } . (9) than one axis simultaneously and, is given in the following
Vector elements in the feature vector are represented by equation:
Δ𝐴 𝑖 and Δ𝐵𝑖 . The similarity between the two images is given 󵄩 󵄩
by the following similarity criteria [16]. Let 𝛼𝑖 be the similarity Distance 𝑑 = ∑ 󵄩󵄩󵄩𝑠𝑖 − 𝜇𝑘 󵄩󵄩󵄩 . (13)
𝑠𝑖 ∈𝑐𝑘
of Δ𝐴 𝑖 and Δ𝐵𝑖 as follows:

{ min (Δ𝐴 𝑖 , Δ𝐵𝑖 ) Whenever each pair is in nonempty intersection, there exists
{ , Δ𝐴 𝑖 ≠ 0 or Δ𝐵𝑖 ≠ 0,
𝛼𝑖 = { max (Δ𝐴 𝑖 , Δ𝐵𝑖 ) (10) an intersection point for the whole collection; therefore,
{ the Manhattan distance forms an injective metric space.
{ 1, Δ𝐴 𝑖 = Δ𝐵 𝑖 = 0.
In the proposed optimized distance method for the given
Similarity between the two images lies in between 0 ≤ 𝛼 ≤ query image only the distance related cluster is searched. The
1. The images in the database are arranged according to the optimized distance metric is given as
similarity match with the query image.
󵄩󵄩 𝑛 󵄩󵄩
󵄩󵄩 3 3 3 󵄩 󵄩
3.2. 𝐾-Means Clustering Algorithm. Clustering is a process 𝐷 = 󵄩󵄩󵄩∑ [(𝑅𝑐 − 𝑅𝑔 ) + (𝐺𝑐 − 𝐺𝑔 ) + (𝐵𝑐 − 𝐵𝑔 ) ]󵄩󵄩󵄩 . (14)
󵄩󵄩𝑖=0 󵄩󵄩
of grouping the similar objects from a given data set. The 󵄩 󵄩
most popular and reliable clustering algorithm is the 𝐾 means The optimized method computes the cube power of the
clustering algorithm that classifies the input data points into distance between the centroid and color pixel points of
multiple classes based on their inherent distance from each the three colors and determines the summation of all the
other. Let 𝑆 = {𝑆𝑖 , 𝑖 = 1, 2, . . . , 𝑁} be the 𝑛-dimensional added values. The modulus of the whole summed values is
data points to be clustered into a set of 𝐾-clusters, 𝐶 = calculated to get the distance value.
{𝐶1 , 𝐶2 , 𝐶3 , . . . , 𝐶𝐾 } [18] from the given elephant data set
𝑋 = {𝑥1 , . . . , 𝑥𝑁}, 𝑥𝑛 ∈ 𝐸𝑑 . The 𝑀-clustering problem aims
at partitioning the elephant data set into 𝑀 disjoint subsets
4. Results and Discussion
(clusters) 𝐶1 , . . . , 𝐶𝑀. The most widely used clustering crite- The field observations are carried out in the forest border
rion is the Euclidean distance [19]. Based on this criterion, areas in Sadivayal elephant camp. The hardware setup was
the clustering of elephant images is grouped depending on arranged to capture the image of elephants. The wireless
the cluster centers 𝑚1 , . . . , 𝑚𝑀 as given below camera was mounted on a wood stick and the camera was
𝑀 𝑀 battery powered with 12 V. Using RF receiver, the video
󵄩 󵄩2
𝐸 (𝑚1 , . . . , 𝑚𝑀) = ∑ ∑ 𝐼 (𝑥𝑖 ∈ 𝐶𝑘 ) 󵄩󵄩󵄩𝑥𝑖 − 𝑚𝑘 󵄩󵄩󵄩 , (11) received is converted to image frames using camcorder
𝑖=1 𝑘=1 software in PC. The elephant image frames are stored in the
where 𝐼(𝑋) = 1 if 𝑋 is true and 0 otherwise. PC memory and updated every 5 seconds. Figure 4 shows
6 The Scientific World Journal

Figure 4: Retrieved online elephant image. Figure 5: 3 Level decomposed query elephant image.

the online EIDS window in which 12 elephant images are

retrieved in 6.33 seconds. The elephant images are arranged in
the order of similarity value obtained. As the retrieved images
are more than 5, a GSM message “Elephant Detected” is sent
using the AT command.
In this work, we perform Haar wavelet decomposition on
the raw elephant image, by determining the scaling coefficient
and largest wavelet coefficients. The scaling coefficient is
stored along with the difference and location (𝑖, 𝑗) of each
wavelet coefficient for every image [15]. The 3-level decom-
posed query elephant image is shown in Figure 5. Using the
3-level wavelet decomposition, the highest and informative
elephant image features are extracted from the coefficients.
These features are used during the process of the query
and database image comparison of the elephants. The Haar
wavelet transform of the elephant is calculated by passing it
through a series of filters (high and low pass filters) and then
Figure 6: Downsampled decomposed elephant image.
downsampled, as we can see from Figure 6.
At each level, the elephant image is decomposed into
low and high frequencies, and this decomposition halves the elephant images from the database image for the given query
resolution since only half the number of samples is retained image.
to characterize the entire image. The Haar wavelet leads to a The EIDS system is tested offline with elephant and nonel-
decomposition of coefficients at level 𝑗 in four components, ephant images. All the images used in this work are in the
and at level 𝑗 + 1. Due to successive downsampling by 2, the dimensions of 3648 × 2763. We also tested the system for
image length must be a power of 2, or a multiple of a power group of elephant images and it is shown in Figure 8.
of 2, and the length of the image determines the maximum The elephant normally moves in herds in the forest
levels into which the elephant can be decomposed. borders during the period of migration. For the given query
The Haar wavelet coefficients of different species are image, the system retrieved 13 images in 6.462 seconds. We
plotted and shown in Figure 7. The Haar wavelet coefficient tested the system with group of elephant images of different
of each species varies with the elephant image and can be sizes and postures. Nonelephant images like bison, bear, deer,
distinguished from other animals. monkey, and human which are the most commonly seen
The value of wavelet coefficients of elephant and bison is species around forest border areas are given as query, which
closer because the major color is black for both species and produced zero image search result. We tested the case with a
the value lies in between 90 and 95. The coefficients for tiger bison image as it is of similar color texture as that of elephants
and deer possess higher band in between 120 and 130. The for which the system retrieved zero search images. On zero
obtained elephant coefficient is averaged to get the threshold search result, no alert is made; hence, GSM message is not
value. We fix 0.6 as threshold value to obtain the similar sent.
The Scientific World Journal 7

140

130
Wavelet coefficients

120

110

100

90

80

70
1 11 21 31 41
Number of values
Elephant Tiger
Bison Monkey
Cow Deer Figure 9: Retrieved images for optimized distance metric.

Figure 7: Wavelet Coefficients for Different Species.

1
Image retrieval time in seconds
0.95 Euclidean distance = 21.544 s
Manhattan distance = 15.067 s
Optimized distance metric = 15.028 s
0.9

0.85
Image similarity

0.8

0.75

0.7

0.65

0.6

0.55
1 3 5 7 9 11 13 15 17 19 21 23 25 27
Number of elephants
Figure 8: Retrieved elephant group image.
Euclidean
Manhattan
We compared the three distance measures with the num- Optimized distance metric
ber of images retrieved, retrieval time, and the retrieval rate Figure 10: Image similarity metric comparison.
per image. The Euclidean scheme produced 20 retrieved
images with retrieval time of 21.544 seconds and the retrieval
rate per image is 1.077 seconds. The Manhattan scheme the three distance measures and the results are shown in
produced 22 retrieved images with retrieval time of 15.067 Figure 11. The retrieval time per image is less for the proposed
seconds and the retrieval rate per image is 0.684 seconds. For distance metric compared to other methods as the metric
the proposed optimized distance metric, the images retrieved retrieves more images which are closer to the cluster center.
were 27 as shown in Figure 9 with retrieval time of 15.028 The images are also retrieved in lesser time interval. Due to
seconds and the retrieval rate per image is 0.556 seconds. reduction of computational time and higher image retrieval
A comparison between all the distance methods is carried rate, the time to react for the elephant intrusion is improved
out in this work and the results are shown in Figure 10. It is using the proposed method.
observed that the proposed method retrieves more images The recall rate [21] is defined as the ratio of the number
with lesser time compared to the other two methods. of relevant (same shape and position) retrieved images to the
The optimized retrieval rate improvement over another total number of images in the database
distance metric is 18%. To assess the retrieval effectiveness,
Number of relevant images retrieved
10 query elephant images are selected and tested. The average Recall rate = . (15)
retrieval time per query elephant image is calculated for all Total number of images in database
8 The Scientific World Journal

1.2 Table 1: Database clustering for different 𝐾 values.

1.15
Number of
1.1 𝐾 Retrieval time Retrieval rate
retrieved
1.05 for images per image
images
1 2 14 10.884 0.777
Retrieval time/image (s)

0.95 3 12 7.604 0.6336

0.9 4 12 7.242 0.6032
0.85 5 9 4.949 0.5498
0.8
0.75 Table 2: Comparison between the online and offline observations.
0.7
Image distance Retrieved images
0.65 Image threshold
Offline Online Offline Online
0.6
0.4 1.0–0.49 0.57–0.36 22 12
0.55
0.5 1.0–0.49 0.44–0.34 22 6
0.5
0.6 1.0–0.59 0.47–0.38 17 5
Query 1

Query 2

Query 3

Query 4

Query 5

Query 6

Query 7

Query 8

Query 9

Query 10
0.7 1.0–0.74 0 2 0

Query
Euclidean We varied the cluster 𝐾 value from 2 to 5 and recorded the
Manhattan corresponding number of retrieved images and the retrieval
Distance optimized rate per image from the database for the query image given.
Figure 11: Image retrieval rate comparison. It is seen that for small number of clusters more images were
retrieved and for higher number of clusters the retrieved
images were less in number as shown in Figure 13. So for our
0.25 analysis, we fix 2 clusters in order to retrieve more images
from the database.
A comparison was made between of offline and online
0.2 elephant image detection system. In the online system for the
threshold values 0.4 to 0.6, the number of images retrieved
varied from 5 to 12 and for offline system, it produced 17 to 22
images. This variation is due to the camera posture and image
Image recall rate

0.15 capture in frame. The required number of 5 images is achieved

in online system with the proposed optimized distance
method that makes the system efficient and reliable. The
0.1 result comparison between the online and offline detection
system is recorded in the Table 2.

0.05 5. Conclusion
In conclusion, the findings of our work contribute to elephant
conservation issues. The work provides solutions to human-
0 elephant conflict. The study provides insights to protect
Euclidean Manhattan Optimzed elephants from human activities and reduces the work effort
distance metric of forest officials. The real time elephant identification system
Figure 12: Elephant image recall rate comparison. provides solutions to the problem of human elephant conflict
and provides solution for unsupervised process of individual
species identification specifically for elephants. The system is
We compare image recall rate for the three methods as shown completely automated; the strength of this approach stems
in Figure 12. It is observed that the optimized distance metric from the ability to narrow down the collection of potential
is 16% better than Manhattan and 18.5% better than Euclidean matches in the database with the query image. Optimal
methods. We also tested the algorithm by varying the cluster results for automated identification of individual elephants
formation. Table 1 shows the variation of number of images are obtained with the algorithm developed and is used to rank
retrieved, retrieval time, and retrieval rate per image. the most likely matches, followed by final supervised visual
When we fix 2 clusters, 14 images were retrieved and the identifications and also with an early warning sent to the
retrieval time for the 14 images was found to be 10.884 and forest officials about the arrival of elephants from the forest
the retrieval rate per image is 0.777. borders into the human habitat. The real time automated
The Scientific World Journal 9

0.75 References
0.7 [1] H. N. Kumara, S. Rathnakumar, M. Ananda Kumar, and M.
Singh, “Estimating Asian elephant, elephasmaximus, density
0.65 through distance sampling in the tropical forests of BiligiriRan-
gaswamy Temple Tiger Reserve, India,” Journal Tropical Conser-
Retrieval rate

0.6 vation Science, vol. 5, no. 2, pp. 163–172, 2012.

[2] S. J. Sugumar and R. Jayaparvathy, “An early warning system for
0.55
elephant intrusion along the forest border areas,” Current Sci-
0.5 ence, vol. 104, pp. 1515–1526, 2013.
[3] J. Lenin and R. Sukumar, “Action plan for the Mitigation of
0.45 Elepant-Human conflict in India,” Final Report to the U.S. Fish
and Wildlife Service, Asian Nature Conservation Foundation,
0.4 Bangalore, India, 2011.
K=2 K=3 K=4 K=5
[4] C. Arivazhagan and B. Ramakrishnan, “Conservation perspec-
Image clusters
tive of Asian Elephants (Elephasmaximus) in Tamil Nadu,
Retrieval rate Southern India,” International Journal of Biotechnology, vol. 1,
pp. 15–22, 2010.
Figure 13: Image retrieval rate per cluster. [5] P. Fernando, M. A. Kumar, A. C. Williams, E. Wikramanayake,
T. Aziz, and S. M. Singh, “Review of human-elephant conflict
mitigation measures practiced in South Asia,” AREAS Technical
approach minimizes the manual work which is not possible Support Document Submitted to World Bank, WWF-World
all the time because it is difficult to monitor the presence Wide Fund for Nature, Vaud, Switzerland, 2008.
of elephants manually when the herds march towards the [6] L. E. King, A. Lawrence, I. Douglas-Hamilton, and F. Vollrath,
forest borders. More importantly, our results demonstrate the “Beehive fence deters crop-raiding elephants,” African Journal
importance of certainty in identifying approaching elephants of Ecology, vol. 47, no. 2, pp. 131–137, 2009.
in to human living areas and provide early warning about the [7] S. R. Loarie, R. J. V. Aarde, and S. L. Pimm, “Fences and artificial
elephant entry into the human habitat. We therefore recom- water affect African savannah elephant movement patterns,”
Biological Conservation, vol. 142, no. 12, pp. 3086–3098, 2009.
mend the use of the real time image processing technique to
[8] A. B. Venkataraman, R. Saandeep, N. Baskaran, M. Roy, A. Mad-
identify an approaching individual elephant as well as a group
hivanan, and R. Sukumar, “Using satellite telemetry to mitigate
of elephants. The system can be implemented in forest border elephant-human conflict: an experiment in northern West
areas. We also solve the traits for distinguishing species where Bengal, India,” Current Science, vol. 88, no. 11, pp. 1827–1831,
animals are differentiated on the basis of Haar wavelet color 2005.
attributes. In the context of elephant conservation, the real [9] P. J. Venter and J. J. Hanekom, “Automatic detection of African
time automated image processing system can be used for elephant (Loxodonta africana) infrasonic vocalisations from
the elephant ecology learning, population monitoring, ele- recordings,” Biosystems Engineering, vol. 106, no. 3, pp. 286–294,
phant habitat usage, commercial ivory poaching and human- 2010.
elephant conflict. The system can also be deployed along [10] C. Vermeulen, P. Lejeune, J. Lisein, P. Sawadogo, and P. Bouche,
forest border migration routes or at water holes and food “Unmanned aerial survey of elephants,” PLoS ONE, vol. 8, no. 2,
plantation areas for elephant tracking and monitoring. pp. 1–7, 2013.
[11] R. Dabarera and R. Rodrigo, “Vision based elephant recognition
In this paper, we proposed and analyzed an elephant
for management and conservation,” in Proceedings of the 5th
image detection system via wavelet decomposition of images, International Conference on Information and Automation for
followed by feature extraction and similarity match under Sustainability (ICIAfs ’10), pp. 163–166, December 2010.
𝐹-norm theory. We compared the retrieval performance of [12] A. Ardovini, L. Cinque, and E. Sangineto, “Identifying elephant
optimized distance metric based 𝐾-means clustering with the photos by multi-curve matching,” Pattern Recognition, vol. 41,
existing techniques like Euclidian distance and Manhattan no. 6, pp. 1867–1877, 2008.
distance. It turns out that optimized distance metric calcula- [13] V. R. Goswami, M. V. Lauretta, M. D. Madhusudan, and K.
tion has 18.5% improvement in recall rates. Field observations U. Karanth, “Optimizing individual identification and survey
show that the proposed method can be used as an effective effort for photographic capture-recapture sampling of species
scheme to detect elephants in the forest border areas even with temporally variable morphological traits,” Animal Conser-
in the presence of different species. This system has been vation, vol. 15, no. 2, pp. 174–183, 2012.
rigorously tested through the various phases of the project [14] P. Porwik and A. Lisowska, “The haar-wavelet transform in
and found to be efficient compared to the existing systems. digital image processing: its status and achievements,” Machine
Graphics and Vision, vol. 13, no. 1-2, pp. 79–98, 2005.
[15] S. Arivazhagan and L. Ganesan, “Texture classification using
Conflict of Interests wavelet transform,” Pattern Recognition Letters, vol. 24, no. 9-
10, pp. 1513–1521, 2003.
The authors declare that there is no conflict of interests [16] M. Y. Latha, B. C. Jinaga, and V. S. K. Reddy, “Content based
regarding the publication of this paper. color image retrieval via wavelet transforms,” International
10 The Scientific World Journal

Journal of Computer Science and Network Security, vol. 7, no. 12,

p. 38, 2007.
[17] P. Manimegalai and K. Thanushkodi, “Content based color
image retrieval using adaptive lifting,” International Journal of
Advanced Networking and Applications, vol. 1, no. 6, pp. 359–
366, 2010.
[18] D. Doreswamy and K. S. Hemanth, “A novel design specification
distance (DSD) based K-mean clustering performance evalua-
tion on engineering materials’ database,” International Journal
of Computer Applications, vol. 55, no. 15, pp. 26–33, 2012.
[19] G. F. Tzortzis and A. C. Likas, “The global kernel 𝜅-means algo-
rithm for clustering in feature space,” IEEE Transactions on
Neural Networks, vol. 20, no. 7, pp. 1181–1194, 2009.
[20] A. Vadivel, A. K. Majumdar, and S. Sural, “Performance com-
parison of distance metrics in content-based image retrieval
applications,” in Proceedings of the International Conference
on Information Technologies, pp. 159–164, Bhubaneswar, India,
2003.
[21] T. Deselaers, D. Keysers, and H. Ney, “Features for image
retrieval: an experimental comparison,” Information Retrieval,
vol. 11, no. 2, pp. 77–107, 2008.
 International Journal of

Rotating
Machinery

International Journal of
The Scientific
Engineering Distributed
Journal of
Journal of

Hindawi Publishing Corporation

World Journal
Hindawi Publishing Corporation Hindawi Publishing Corporation
Sensors
Hindawi Publishing Corporation
Sensor Networks
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of

Control Science
and Engineering

Advances in
Civil Engineering
Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Submit your manuscripts at

http://www.hindawi.com

Journal of
Journal of Electrical and Computer
Robotics
Hindawi Publishing Corporation
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

VLSI Design
Advances in
OptoElectronics
International Journal of

International Journal of
Modelling &
Simulation
Aerospace
Hindawi Publishing Corporation Volume 2014
Navigation and
Observation
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
in Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com
http://www.hindawi.com Volume 2014

International Journal of
International Journal of Antennas and Active and Passive Advances in
Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration
Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014