Professional Documents
Culture Documents
Abstract—Precise estrus detection is a factor for specifically, the growth rate is 8.6 %. Meanwhile, Laos, and
the reproductive performance of cows. The primary Vietnam both have a 7.5 % growth rate. This is followed by
sign of estrus is the standing heat wherein a cow Indonesia with a growth rate of 3.5% and then, Myanmar
stands still for a few seconds while mating with other at 0.4%. The Philippines, however has a negative growth
cows. Visual monitoring is the most common method rate. The slaughter rate of the cattle was 4.88 % which is
used for detection of estrus that requires farmers’ relatively higher than the previous years [1]. In 2010, the
time and attention for high yield. In this study, an government imported cattle for breeding, which caused
estrus detection using image recognition is used to an increase in the growth of the population of the cattle
detect the standing heat. The system is comprised of from 2010 to 2015 [2]. The efficiency and determination
detection, identification, and notification system. Scale the optimum time for insemination of cattle is beneficial
Invariant Feature Transform (SIFT) is responsible to increase the pregnancy rate and consequently lead to
for the detection and identification of in-heat cows. economic growth resulting to less importation of cattle.
Using SIFT, the images of cows were registered in the The failure to detect estrus (heat) is considered as one of
database, these images were used for detection and the major factors for low fertility of cattle [3]. Hence,
identification of cows and an algorithm for feature for the dairy farmers, it is important to detect estrus in
overlapping were created to detect the standing heat. cattle.
When a standing heat is detected, it is recorded into Due to the insufficient detection of estrus, the
the computer and simultaneously the Global System reproduction growth in cattle decreases which subsequently
for Mobile Communications (GSM) module will send leads to the increase cost due to the artificial insemination.
an alert message. A cow in heat usually stands still to be mounted 20 to
55 times and each mount lasts 3 to 7 seconds. In using
Keywords: estrus, in-heat cows, standing heat, image
artificial insemination, the optimal time is 8 to 12 hours
recognition, artificial insemination
after the first standing heat. Therefore, an effective estrus
detection is very important to predict their heat condition
[3]. The average number of standing events ranges between
I. Introduction 6 to 9 per estrus to be consider in heat [4]. There are new
T he Philippine cattle industry is one of the least technologies developed to detect estrus better than visual
developed commodities in the past years [1]. The observation. There are past studies that have shown a
high slaughter rate causes the declining population of the concrete result of the efficiency and effectiveness of
cattle. The average growth rate of cattle population in the estrus detection. Using only visual monitoring is time-
Asian region is at 2.7%. In Indo-China and Cambodia, consuming and requires repetitive actions that need to be
done frequently [5].
Using an estrus detection for monitoring is not new for
Electronics Engineering Department, Technological the dairy farmers, but the dairy producers is interested to
University of the Philippines, Manila, Philippines (email: a modern, low cost, user-friendly and advance monitoring
nilo_arago@tup.edu.ph.com, august_thio-ac@tup.edu.ph, systems, hence this study. The main objective of this study
miguel.casis.apostol@gmail.com, irwinjamesdg@gmail. is to develop an estrus detection system for Philippine
com, albertelireyes@gmail.com, kathyrdrgz02@gmail.com, cows.
roialdrintoring@gmail.com)
II. Technologies for Estrus Detection define a feature. These features are invariant in rotation,
scaling, and in changes of illumination.
A. Related Works Detection of space-scale extrema, localization of key
Chen and Lin used a Wireless Sensor Network (WSN) points, orientation assignment for each key points, and
to develop a system that recognizes standing-heat signal [6]. creating descriptors are the major steps in implementing
The system used a 3-axial accelerometer that recognizes SIFT. Fig. 1.shows an overview of the SIFT algorithm.
the movements of the cows. The accelerometer is placed
[6]. The on the systemcow’s used forefeet. a 3-axial
The authors accelerometer
utilized ZigBee that and
recognizes
[6]. [6].
The The received
system the
system movements
signal
used used strength
a 3-axial of the
a 3-axial cows.
indicator The
(RSSI)
accelerometer
accelerometer accelerometer
in achieving
that thatestrus
is placed
recognizes
recognizes on
detection.
the the cow’s
the movements
movements forefeet.
of theofcows. The
the cows. authors utilized
The accelerometer
The accelerometer ZigBee
and
is placed received
is placed
on theon signal
the cow’s
Floresca
cow’s strength
et al.
forefeet. forefeet.
[7]The indicator
The authors
presented
authors (RSSI)
autilized
monitoring in
utilized achieving
ZigBee ZigBeecalled
system
estrus
and received detection.
and received signalsignal
DISCOW to detect
strength strength and indicator
indicator observe (RSSI) (RSSI)
mounting in
in achieving achieving
activities of cows.
estrusestrus detection.
detection.The system uses QR codes
Floresca et al. [7] presented a monitoring system called patches, high definition cameras,
DISCOW and to
Floresca a [6].
personal
detect
et
The computer.
[7]and
al.presented
system used
presentedobserve aWhen
a 3-axial
mounting
monitoring the QR accelerometer
code called
activities
system is blocked
of
that
Floresca et al.recognizes
[7] a monitoring
the movements of thesystem cows. called
The accelerometer
cows.
DISCOW by the
The toiscow
system detect from
uses and mounting,
QR codes
observe the computer
patches,
mounting high will analyze
definition
activities the
DISCOW to detect placed
and on the
observe cow’s forefeet.
mounting The authors
activities of of
utilized ZigBee
cameras,
cows.cows. and
detection
The system
The system a
and
usespersonal
and uses
received signals
QR codes QR computer.
the
signalcodes camera
strength
patches, When
patches, to
high high
indicator the
capture QR
(RSSI)
definition code
image,
definition is
in achievinggiving
blocked
cameras,
cameras, and allby the
aand the
estrus cow
necessary
a personal
personal from
detection. information
computer. mounting,
computer. When about
Whenthethein-heat
QR computer
the QR
code cows. is will
code is
analyze
blocked
blocked by the the
byBycowdetection
utilizing
the frominfrared
cow
Floresca
and
from signals
[7]cameras,
et al. mounting,
mounting, the computer
the
presented
camera
Yuang,
the computer
a monitoring
towill
Lin, capture
and
system Peng
will [8]
called
image,
analyze
analyze giving
the developed
detection all athe
the detection
DISCOW system
and necessary
toand that
detect
signals signals
aims
and
the information
therecognize
to
observe
camera camera toabout
mounting to
estrus
capture in-heat
capturedetection
activities of
cows.
image,image,
giving atgivingall cows.
night theall
usingthe
The a
necessary necessary
system
non-contactuses information
information QR codes
video about about
patches,
monitoring
in-heat in-heat
high definition
system.
cameras, and a personal computer. When the QR code is
cows.cows. ByImage utilizing processing
blocked infrared will
by thecameras,
be usedmounting,
cow fromYuang,
to identify Lin,theand estrus-specific
Peng will
computer
[8] By behaviors
developed
utilizing a
analyze of dairy
system
the detection
infrared cows.
that
cameras, The
aims authors
and Yuang, to
signals recognize
the developed
Lin, camera
and to acapture
estrus
Peng video
By utilizing infrared cameras, Yuang, Lin, and Peng
detection imaging
[8] developed at image,
night technology
giving
using
a system all
a that
the uses
necessary
non-contact the length
information
video of the
about
monitoring cowsin-heat to
[8] developed a cows.
system that that aims aims to recognize
to recognize estrusestrus Fig. 1. Overview of SIFT Algorithm
system.
detection
detection Image
detect
at night at night processing
estrus.
usingusing A length will of be
a non-contact
a non-contact 2 used
cows to
shows
videovideo identify
specific
monitoring estrus-
monitoring mounting
specific behaviors Fig. 1. OverviewFig. 1.Fig. of SIFT Algorithm
1. Overview ofAlgorithm
SIFT Algorithm
system.
system. Image behavior
Image Byofutilizing
and
processing
processing dairy
then cows.
willa length
will
beinfraredbeof
used The 1.5authors
used
cameras,
to cows
to identify
identify Yuang,developed
will be
estrus- theand
estrus-
Lin, a Peng
basis of Overview
Scale-space extrema detection identifies key points that
of SIFT
video
specific
specific imaging
the
behaviors [8]technology
of developed
detected
behaviors dairy estrus.
of dairy
cows. that
acows.
The uses
system The the
authors length
thatauthors
developed ofrecognize
aims todeveloped the acowsaestrus
are invariant Scale-space
Scale-space to extrema
orientation
extrema detection
detection and scale identifies
identifies through keythe
key pointsusethatof
to
videovideodetect
imaging imaging detection
estrus.
In 2014,
technology A
technologyTsai at
that
night
length
and that
uses
using
of
Huanguses
the 2 a non-contact
cows
the introduced
[9]
length length
of shows
the
video
ofcowsthe monitoring
aspecific
cows
motion and Scale-space
difference-of-Gaussian
are
extrema detection
Fig. 1. Overview function
identifies
of SIFT (1). Algorithm
key points that
mounting system.and
behavior Image then processing will be cowsused to identify theestrus- are invariant areinvariantinvariantto orientationtotoorientation
orientation and scale and
and scale through
scale through
through the use the
the of use
use of of
to detect
to detect image
estrus. estrus.
analysis
A length
specific Abehaviors
length
methodof a2of length
of
for
cows
dairy
of
2automatic
cows 1.5
shows
cows. Theshows will
detection
specific
authors
be of
specific
developed estrus difference-of-Gaussian
adifference-of-Gaussian function (1).
basis
mountingof the detected
behavior andestrus. then a length of 1.5 cows will be the difference-of-Gaussian
Scale-space extrema detection function function (1). (1).
identifies key points that
mounting behavior and forvideo and then a length
monitoring
imaging mating
technology of 1.5 thatcows
behavior uses will of be
the alength theof A
cattle. theregion
cows 𝐷𝐷𝐷𝐷(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦, 𝜎𝜎𝜎𝜎) = to
are invariant �𝐺𝐺𝐺𝐺(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦, 𝑘𝑘𝑘𝑘𝜎𝜎𝜎𝜎) and
orientation − 𝐺𝐺𝐺𝐺(𝑥𝑥𝑥𝑥,
scale𝑦𝑦𝑦𝑦,through 𝜎𝜎𝜎𝜎)� ∗ 𝐼𝐼𝐼𝐼(𝑥𝑥𝑥𝑥, the 𝑦𝑦𝑦𝑦)
use of(1)
basisbasis
of the ofdetected
of the detected
to estrus.
interest detect
for estrus.
estrus.video
each
In 2014, Tsai and Huang [9] introduced a motion and A length frame ofis2analyzed cows shows where specific
high
mounting behavior and then a length of 1.5 cows will be the 𝐷𝐷𝐷𝐷(𝑥𝑥𝑥𝑥, 𝐷𝐷𝐷𝐷(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦,difference-of-Gaussian
𝜎𝜎𝜎𝜎) 𝑦𝑦𝑦𝑦,
=𝜎𝜎𝜎𝜎) = �𝐺𝐺𝐺𝐺(𝑥𝑥𝑥𝑥,
�𝐺𝐺𝐺𝐺(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦,function
𝑦𝑦𝑦𝑦, 𝑘𝑘𝑘𝑘𝜎𝜎𝜎𝜎) − 𝑘𝑘𝑘𝑘𝜎𝜎𝜎𝜎)
𝐺𝐺𝐺𝐺(𝑥𝑥𝑥𝑥,− (1).
𝐺𝐺𝐺𝐺(𝑥𝑥𝑥𝑥,
𝑦𝑦𝑦𝑦, 𝜎𝜎𝜎𝜎)�𝑦𝑦𝑦𝑦,∗𝜎𝜎𝜎𝜎)� 𝐼𝐼𝐼𝐼(𝑥𝑥𝑥𝑥,∗𝑦𝑦𝑦𝑦)
𝐼𝐼𝐼𝐼(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦)
(1) (1)
image In levels
analysis
2014, of
Tsaimotion
method and forthat
Huang hasestrus.
automatic[9] occurred.detection
introduced After aof analyzing
estrus
motion and
and video Key points are highly selected based on their stability
In 2014, Tsai and Huang [9] introduced a motion and
basis of the detected
for
image monitoring
frames,
analysis mating
foreground
method behavior
for segmentation
automatic of detection
a cattle.
then A
follows. region
of estrus of
It separates
and measures.
Key
𝐷𝐷𝐷𝐷(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦,To
Key
points pointsareget are
highlyaccurate
𝜎𝜎𝜎𝜎) = �𝐺𝐺𝐺𝐺(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦, 𝑘𝑘𝑘𝑘𝜎𝜎𝜎𝜎)key
highly selected selected point
− 𝐺𝐺𝐺𝐺(𝑥𝑥𝑥𝑥,
based
𝑦𝑦𝑦𝑦, 𝜎𝜎𝜎𝜎)�localization,
based ∗ 𝐼𝐼𝐼𝐼(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦) (1)
on their on their stability astability
second
image analysis method Infor 2014, automatic
Tsai the and detection
Huang [9] ofintroduced
estrus andaeach
motion and Key points areexpansion
highly selected based on (2). their stability
interest
for monitoringfor
the each
moving video
matingcow frame
from
behavior is analyzed
region
of of where
a cattle. interest highin
A region levels offrame. order
measures. Taylor-series To getare accurate is
key point implemented localization, a second
for monitoring mating behavior
image analysis method of afor cattle.
automatic A detection
region ofof estrus andmeasures. measures. To Key get accurate
points
To get accurate key selected
highly pointkey localization,
point basedlocalization, on theira stabilitysecond a second
of motion By that
fordetermining hasvideo occurred.the ischanges After analyzing
of lengths video
between frames,
interest
interest for each each
video frame
for monitoring frame is analyzed
analyzed
mating where
behavior where
ofhigh high
levels
a cattle. Athe
levels moving
region oforderorder Taylor-series Taylor-series
measures. Toexpansion expansion
get accurate iskey is
implemented
point implemented
localization, (2). (2).
a second
foreground
of motion objects, segmentation
that has following
the occurred. then After
of follows.
cows analyzing
and It separates
mounting video frames,
behaviorsthelevels are order Taylor-series expansion𝜕𝜕𝜕𝜕𝐷𝐷𝐷𝐷 is 1𝑇𝑇𝑇𝑇 is implemented 2
𝜕𝜕𝜕𝜕 𝐷𝐷𝐷𝐷 (2). (2).
of motion that has occurred.
interest for each After video analyzing
frame video
is analyzed frames,
where high order Taylor-series 𝐷𝐷𝐷𝐷(𝑥𝑥𝑥𝑥) = 𝐷𝐷𝐷𝐷expansion + implemented
𝑇𝑇𝑇𝑇
moving cow from the region of interest in each frame. By 𝑇𝑇𝑇𝑇 + 2𝑥𝑥𝑥𝑥 𝑥𝑥𝑥𝑥 (2)
foreground
foreground identified. segmentation
of motion that has
segmentation then follows. It separates the then follows.
occurred. After It separates
analyzing video theframes, 𝜕𝜕𝜕𝜕𝐷𝐷𝐷𝐷 𝑇𝑇𝑇𝑇 𝜕𝜕𝜕𝜕𝐷𝐷𝐷𝐷 1 𝜕𝜕𝜕𝜕 2 𝐷𝐷𝐷𝐷
1𝑇𝑇𝑇𝑇 𝑇𝑇𝑇𝑇 𝜕𝜕𝜕𝜕 𝐷𝐷𝐷𝐷𝑇𝑇𝑇𝑇2𝜕𝜕𝜕𝜕𝑥𝑥𝑥𝑥
𝜕𝜕𝜕𝜕𝑥𝑥𝑥𝑥 2 2
determining 𝐷𝐷𝐷𝐷(𝑥𝑥𝑥𝑥) = 𝐷𝐷𝐷𝐷 + + 𝑥𝑥𝑥𝑥 𝑥𝑥𝑥𝑥 (2)
movingmoving
cow cow fromthe thechanges
foreground
from the region
region ofofinterest
segmentation lengths
of interest in between
then
each follows.
in each
frame. theframe.
It moving
separates
By By the 𝐷𝐷𝐷𝐷(𝑥𝑥𝑥𝑥) = 𝐷𝐷𝐷𝐷 +
𝐷𝐷𝐷𝐷(𝑥𝑥𝑥𝑥) = 𝐷𝐷𝐷𝐷
+ 𝑥𝑥𝑥𝑥 1 2 𝑇𝑇𝑇𝑇 𝜕𝜕𝜕𝜕2 𝑥𝑥𝑥𝑥𝜕𝜕𝜕𝜕𝑥𝑥𝑥𝑥
𝜕𝜕𝜕𝜕𝐷𝐷𝐷𝐷
+ 𝜕𝜕𝜕𝜕𝑥𝑥𝑥𝑥
𝜕𝜕𝜕𝜕𝑥𝑥𝑥𝑥 2 + 𝜕𝜕𝜕𝜕𝑥𝑥𝑥𝑥
𝑥𝑥𝑥𝑥
𝐷𝐷𝐷𝐷 2 (2)
𝑥𝑥𝑥𝑥 (2)
objects,
determining
determining the following
moving
the changes cow of lengths
cows
from the
of lengthsandregionmounting
of
between thebehaviors
interest in each
the moving are
frame. By After the selection of 𝜕𝜕𝜕𝜕𝑥𝑥𝑥𝑥key2points, 𝜕𝜕𝜕𝜕𝑥𝑥𝑥𝑥 it will be assigned in
2
mouth,
classify
classify them. chin,
them.
[10]. These and
These
mouth, other
include chin, facial
include and other
distinct structures,
distinct facial
features features which
structures,
such such sets
which
as eyes, as them
eyes,
sets them �𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦+1)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦−1)�
�𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦+1)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦−1)�
2
apart
mouth, from chin, other
apart
and objects.
from
other other Face
facial recognition
objects. Face
structures, algorithms
recognition
which algorithms
sets and
them and
𝜃𝜃𝜃𝜃(𝑥𝑥𝑥𝑥,
𝜃𝜃𝜃𝜃(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦)
𝑦𝑦𝑦𝑦)==tan tan � � 2 2 � 22� (3)
(3)
mouth, chin, and In other
2004,facial Lowestructures, developedwhich Scale-invariant
sets them feature �𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦+1)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦−1)�
�𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥+1,𝑦𝑦𝑦𝑦)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥−1,𝑦𝑦𝑦𝑦)�
�𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥+1,𝑦𝑦𝑦𝑦)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥−1,𝑦𝑦𝑦𝑦)�
�𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦+1)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦−1)�
𝜃𝜃𝜃𝜃(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦) = tan � � (3)
apartbiometrics are oneobjects.
of the applications ofalgorithms
feature recognition 𝜃𝜃𝜃𝜃(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦) = tan � 2 � 2 (3)
apart biometrics are one of the applications of feature recognition
from from other
transform other
objects.
[10]. (SIFT) Face[11], Face anrecognition
recognition algorithm algorithms
that and and
extracts a set of �𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥+1,𝑦𝑦𝑦𝑦)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥−1,𝑦𝑦𝑦𝑦)�
�𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥+1,𝑦𝑦𝑦𝑦)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥−1,𝑦𝑦𝑦𝑦)�
[10].
biometrics
biometrics are oneareof one theofapplications
the applications of
of feature feature
recognitionrecognition
unique points from gray-level images. From a gray-level C. K-NearestNeighbor
C. K-Nearest Neighbor (KNN) (KNN)Machine Machine LearningLearning
[10]. [10]. image, a list
In of 2004, 2D points
Lowe developedwill be extracted.
Scale-invariant Each point
featureis Algorithm
In 2004,transform Lowe (SIFT) developed Scale-invariant feature a set of C. K-Nearest C. K-Nearest Algorithm Neighbor (KNN) Machine Learning
accompanied withan a vector[11], of anlow-level
algorithm descriptors
that extractsand TheNeighbork-nearest neighbors (KNN) Machine
(KNN) isLearning a machine learning
transform
In In 2004,
2004, (SIFT)
Lowe Lowe
unique [11],
points
developed algorithm
developed
from gray-level thatimages.
Scale-invariant
Scale-invariant extracts From
feature a feature
set of thus,
a gray-level Algorithm The
algorithmAlgorithm
k-nearest that is neighbors commonly used (KNN) in, is buta not machine limited learning
to,
unique
transform
transform points
(SIFT) (SIFT) from[11],
image,
[11], a gray-level
an list of 2D
an
algorithm algorithmimages.
points
thatwill that From
be extracts
extracts aa set
extracted. gray-level
Each
aof set of point is algorithmclassification
The that
k-nearest is commonly
problems.neighbors The (KNN) used
KNN in,
algorithm
is but
a not
machine works limitedby to,
learning
image, a list of 2D
accompanied points withwill a bevectorextracted.
of From
low-level Each point
descriptors is and The k-nearest neighbors (KNN) is a machine learning
unique
unique points from points from
gray-level gray-level images. a gray-level classification
algorithmstoring all problems.
available
thatcommonly is commonly cases The
and KNN
classifies
used algorithm
new
in, not available
but limitednot limited worksto, by
cases to,
accompanied thus,
with define
a vector a images.
feature.
of
From features
These
low-level
a gray-level
descriptors are invariantand inalgorithm by that aallmajorityis vote of its used in, but
k classifies
neighbors. Measuring the
image,
image, a list of 2Da list of 2D points
pointsscaling, will be will be
extracted.extracted. Each Each
point is point is storing
classification
classification available
problems. problems. cases
The and
The
KNN KNN algorithm new
algorithm available
works works
by casesby
thus, define rotation, and in changes of illumination. distance by a distance function will assign the case to a
accompanied
accompanied withawith a feature.
vectora vector These
of of features
low-level low-level are
descriptors invariant
descriptors and and in
storing by
storingalla available
class
majority
all
that available
is
votecases
cases
most and
common
of classifies its
and among
kclassifies
neighbors.
its new K newMeasuring
available
nearest available cases
neighbors
the
cases
thus,rotation,
thus,
definedefine scaling,
a feature. and
a feature.
Detection in changes
These These
of features offeatures
space-scale illumination.
are are invariant
invariant
extrema, in inof keyby adistance
localization bymajority a majority
[12]. by vote a distance of itsoffunction
vote kitsneighbors. will assign
k neighbors. Measuring the case
Measuring the tothe a
5.8 meters in width, 3.80 meters in height and 88.16 square
= �(𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥 + 1, 𝑦𝑦𝑦𝑦) − 𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥 − 1, 𝑦𝑦𝑦𝑦))2 + (𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦 + 1) − 𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦 − 1))2 meters in area. Four high definition cameras were placed on
2
top of the barn and are 3.85 meters apart. In Fig. 2, an
�𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦+1)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥,𝑦𝑦𝑦𝑦−1)� equation was derived in determining the camera focal
𝜃𝜃𝜃𝜃(𝑥𝑥𝑥𝑥, 𝑦𝑦𝑦𝑦) = tan � 2� (3) length.
Development�𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥+1,𝑦𝑦𝑦𝑦)−𝐿𝐿𝐿𝐿(𝑥𝑥𝑥𝑥−1,𝑦𝑦𝑦𝑦)�
of an Automated Cows In-Heat Detection and Monitoring System Arago et al. 11
(6) (6) 8.1 operating system with an Intel Core i5 series 3rd
Philippines – Los Baños. generation (2.80Ghz) processor
Unit (CPU) along with a 4 GB RAM.
for the Central Processing
B. Hardware Development The GSM module used is the A6 mini with a working
D. METHODOLOGY frequency between 850-1900 MHz and an Arduino UNO.
III. Methodology
A. Data Gathering The barn has the dimensions Tof
a b l15.2
The Arduino UNO
e meters
1 was t a n d ain
Sconnected rtod length,
it asS the
e n scontroller
o r S iof
z e
the notification system.
A total of 235.8 meters cows inwerewidth,
registered3.80
and meters in height and 88.16 square
A. Data Gathering
Holstein Based on the computation, the focal length to be used
A total of 23 Holstein cows were2017.
registered
The and C. should The
monitored in
system monitored
meters
the system
in thetested
was deployed,
inandstarting
system area.
starting November
Fourin high
November
evaluated 2017.definition
Dairy The cameras
Software
Based
be
should that
Development
on
less
be less
the
than were
computation,
3.14
than 3.14
mm and
mmmm.
placed
the
3.55focal
mm. onsmaller
length
The to be used
and 3.55 mm. The smaller
focal video fe
length is available is 2.8
Trainingsystem
and was top
Research of
deployed, theand
tested
Institute ofbarn and
evaluated of are
in Dairy
University 3.85 meters
Training
the focalThe
length apart.
that
system is
was In isFig.
available
composed IP 2,
2.8ofmm. an laptop, and
cameras,
identify
was first
Philippines
and–Research
Los Baños.
Institute of University of the Philippines – Los
equation was derived in determining
Baños.Development
the GSM module.
The system the
wasThe camera
StarvisofVesper
composed focal
IP camera
IP cameras, laptop,serves
and the user
image a
B. Hardware the GSM module. The Starvis Vesper IP camera serves as
length.
The barn has the dimensions of 15.2 meters in length,
5.8 meters in width, 3.80 meters in height and 88.16 square
meters in area. Four high definition cameras were placed on
top of the barn and are 3.85 meters apart. In Fig. 2, an
equation was derived in determining the camera focal
length.
asthe
theinput
inputdevice
deviceofofthe
thesystem.
system.ItIt detects
detects the
the cows
cows and
and The camera of the system was used to capture live video
standing
standingheat.
heat. feed of cows in the barn. It was used to detect and identify
The laptop used in this study ran on is 64-bit Windows the cows inside the barn. The XAMPP application was first
The laptop used in this study ran on is 64-bit Windows
8.1 operating system with an Intel Core i5 series 3rd opened to give the access to the main GUI where the user
6) 8.1 operating system with an Intel Core i5 series 3rd
generation
generation(2.80Ghz)
(2.80Ghz) processor
processor for
for the
the Central
Central Processing
Processing must log in. Upon logging in, the cows top view image and
Unit
Unit(CPU)
(CPU)along
along with
with aa 44 GB
GB RAM.
RAM. its data information (like; date and time of registry, and
The GSM module used is the A6 mini with a working ID) will be inputted by the user in the Register New Cow
The GSM
frequency module
between used isMHz
850-1900 the A6andmini with a working
an Arduino UNO. Menu. Such image is also known as a Model that will be
frequency between 850-1900 MHz
The Arduino UNO was connected to it as the and an Arduino UNO.
controller of used for identifying, tracking and detecting the cows that
The Arduino UNO was connected to it as the controller of
the notification system. has been registered in the system. Each Cow that has
the notification system.
ered and been registered inside the barn will have its own identity
017. The C.C.Software
Software Development
Development in the system and will be named by the number imprinted
n Dairy on their back.
of the Using SIFT feature extraction, the cows will be
detected by their different assigned pattern on the system.
The number of Models in the registry will determine the
n length, accuracy of the detection. Afterward, the system will
16 square initialize the FLANN matcher that performs a K-Nearest
placed on Neighbor search. Consequently, the system will evaluate
ig. 2, an the features if it is a good match or not. The standing-heat
era focal
will be detected by the overlapping of two cows’ detection.
If the threshold reaches three to seven seconds, the system
will record the threshold and simultaneously alert the
caretaker of the barn using the GSM module that a cow is
set to be inseminated.
The GSM module will send a text message that a certain
cow had a standing heat. The data of the standing-heat will
then be stored in the database. A PDF file of the gathered
data containing information of the said standing-heat that
is stored on the database is ready to be printed out for
the user’s copy. A log file of all the standing heat image
𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 captured is also generated as well as a tabulated data under
(1) the Cows Detected Menu. A Print All option is also provided
to allow the user to print out the said data in PDF file.
𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿