Rancang Bangun Sistem Pendeteksi Gerak B

RANCANG BANGUN ALGORITMA DETEKSI
GERAK ADAPTIF BERBASIS WEB
DENGAN MENGGUNAKAN METODE FRAME
DIFFERENCES DAN DYNAMIC TEMPLATE
MATCHING
Naskah Publikasi
untuk memenuhi sebagian persyaratan
mencapai derajat Sarjana S-2
Program Studi S2 Teknik Elektro
Konsentrasi Sistem Komputer dan Informatika
Jurusan Teknik Elektro dan Teknologi Informasi

diajukan oleh
Muhammad Ihsan Zul
10/305739/PTK/06833
kepada

PROGRAM PASCASARJANA
FAKULTAS TEKNIK
UNIVERSITAS GADJAH MADA
YOGYAKARTA

2012

DESIGN OF WEB BASED ADAPTIVE MOTION
DETECTION ALGORITHM USING
FRAME DIFFERENCES AND DYNAMIC
TEMPLATE MATCHING METHODS
Computer and Informatic Systems
Department of Engineering Science
Faculty of Engineering

Proposed by:
Muhammad Ihsan Zul
10/305739/PTK/06833

To

GRADUATE SCHOOL
FACULTY OF ENGINEERING
GADJAH MADA UNIVERSITY
YOGYAKARTA

2012

Naskah Publikasi
RANCANG BANGUN ALGORITMA DETEKSI GERAK
ADAPTIF BERBASIS WEB DENGAN MENGGUNAKAN
METODE FRAME DIFFERENCES DAN DYNAMIC
TEMPLATE MATCHING
yang dipersiapkan dan disusun oleh
MUHAMMAD IHSAN ZUL
10/305739/PTK/06833

Pembimbing Utama

…………………………………………..
Widyawan, S.T., M.Sc., Ph.D

Pembimbing Pendamping

…………………………………………..
Ir. Lukito Edi Nugroho, M.Sc., Ph.D

Pengelola Program Studi : S2 Teknik Elektro

Ir. P. Insap Santosa, M.Sc., Ph.D.
NIP. 196101081985031002

Mengetahui,
Ketua Jurusan / Wakil Penanggung Jawab Program Studi Teknik Elektro

Ir. Lukito Edi Nugroho, M.Sc., Ph.D
NIP. 196603271991031002

DESIGN OF WEB BASED ADAPTIVE MOTION
DETECTION ALGORITHM USING FRAME
DIFFERENCES AND DYNAMIC TEMPLATE
MATCHING METHODS
Proposed by :
MUHAMMAD IHSAN ZUL
10/305739/PTK/06833

Supervisor

Co-Supervisor

…………………………………………..
Widyawan, S.T., M.Sc., Ph.D

…………………………………………..
Ir. Lukito Edi Nugroho, M.Sc., Ph.D

Programme Director : Magister of Electrical Engineering

Ir. P. Insap Santosa, M.Sc., Ph.D.
NIP. 196101081985031002

Head of Electrical Engineering and Information Technology Department

Ir. Lukito Edi Nugroho, M.Sc., Ph.D
NIP. 196603271991031002

Rancang Bangun Algoritma Deteksi Gerak yang Adaptif Berbasis Web
Menggunakan Metode Frame Differences dan Dynamic Template
Matching
Design of Web Based Adaptive Motion Detection Algorithm using Frame
Differences and Dynamic Template Matching Methods
Muhammad Ihsan Zul1, Widyawan2, Lukito Edi Nugroho3
Program Studi S2 Teknik Elektro
Program Pascasarjana Universitas Gadjah Mada
INTISARI
Ada banyak cara yang dilakukan untuk mendeteksi gerak dalam kajian computer vision.
Metode yang umum digunakan untuk mendeteksi objek yang bergerak dilakukan dengan
membandingkan dua atau lebih citra yang ditangkap berurutan. Pembandingan dengan
melakukan analisis setiap piksel dari dua atau lebih citra dikenal dengan nama Frame
Differences. Selain, template matching juga dikenal dengan sebuah metode penentuan citra

pembanding atau citra referensi. Citra referensi yang ditentukan secara dinamis dikenal
dengan nama dynamic template matching. Penelitian ini mengajukan algoritma penentuan
citra referensi secara adaptif dengan menggunakan metode dynamic template matching.
Algoritma ini menggunakan tiga metode penentuan citra referensi berdasarkan perbuahan
kondisi area tangkapan kamera. Algoritma ini direalisasikan dengan menggunakan bahasa

pemrograman web dan menggunakan IP Camera sebagai alat pendeteksi gerakan. Algoritma
ini menghasilkan akurasi pendeteksian hingga 95,5%.

Kata Kunci - frame differences, dynamic template matching, deteksi gerak berbasis web
1

Fakultas Teknik, Universitas Gadjah Mada, Yogyakarta, Indonesia
Fakultas Teknik, Universitas Gadjah Mada, Yogyakarta, Indonesia
3
Fakultas Teknik, Universitas Gadjah Mada, Yogyakarta, Indonesia
2

Design of Web Based Adaptive Motion Detection Algorithm using Frame
Differences and Dynamic Template Matching Methods
Rancang Bangun Algoritma Deteksi Gerak yang Adaptif Berbasis Web
Menggunakan Metode Frame Differences dan Dynamic Template
Matching
Muhammad Ihsan Zul1, Widyawan2, Lukito Edi Nugroho3
Program Studi S2 Teknik Elektro
Program Pascasarjana Universitas Gadjah Mada

ABSTRACT
There are many ways to detect the moving object in term of image motion detection. A
common method that is used to detect moving object recognize by comparing two or more
sequence images. Comparing image by analysing all of image pixel is known as frame
differences method. Template matching is a method that used to determine the reference

image. Reference image which determined dynamically is known a dynamic template
matching. This research proposes an algorithm to determine the reference image by using
dynamic template matching adaptively. In the system, there are three ways to determine the
reference image base on environment condition. This research realizes an algorithm by using
web based system and using IP Camera as measures device. This algorithm provide detection
accuracy rate 95.5%.

Keywords- frame differences, dynamic template matching, web based motion detection
1

Faculty of Engineering, Gadjah Mada University, Yogyakarta, Indonesia.
Faculty of Engineering, Gadjah Mada University, Yogyakarta, Indonesia.
3
Faculty of Engineering, Gadjah Mada University, Yogyakarta, Indonesia.

2

I. PENDAHULUAN
Kamera pemantau merupakan perangkat yang digunakan
untuk memantau suatu area atau objek. Terdapat berbagai
jenis kamera pemantauan yang digunakan untuk sistem
keamanan. Salah satu
kamera pemantau yang umum
digunakan dalam pemantauan adalah IP Camera . Penerapan
sistem pemantauan memiliki fitur pendeteksian gerak
berdasarkan citra yang terdeteksi. Pendeteksian gerak ini
dilakukan dengan menganalisis citra-citra yang ditangkap.
Mekanisme pendeteksian gerak dimulai dari penentuan
citra referensi dengan citra pembanding. Citra pembanding
dianggap sebagai kondisi normal sebuah ruangan. Citra
tersebut dibandingkan dengan kondisi setelah dilakukan
penangkapan citra. Proses penangkapan citra dilakukan secara
berkala sesuai dengan kebutuhan sistem.
Menurut penelitian yang dilakukan oleh Mishra et al. [1]
ada tiga metode yang umum digunakan untuk mendeteksi

gerak. Metode tersebut adalah background subtraction ,
optical flow dan temporal differences. Background
subtraction dilakukan dengan membandingkan citra tertentu
dengan citra yang dijadikan sebagai referensi. Background
subtraction
melakukan pendeteksian gerak dengan
menggunakan teknik penentuan gambar referensi secara statis
[2, 3, 4].
Penelitian [5, 6, 7, 8] menggunakan optical flow dalam
penelitian tentang deteksi gerak yang dilakukan. Penelitian
tersebut cukup sulit diterapkan untuk real time video
surveillance. Penerapan optical flow membutuhkan perangkat
keras tambahan untuk mendukung kinerja dan performa
sistem pemantauan. Metode temporal differences juga dikenal
dengan nama frame differences. Metode ini dilakukan dengan
membandingkan frame-frame citra yang ditangkap. Penelitian
lain yang dilakukan oleh Kenchannavar et al. [9] menjelaskan
tentang algoritma yang diterapkan dalam metode background
subtraction dan frame differences. Penelitian yang dilakukan
dengan menerapkan konsep SAD. SAD merupakan singkatan

dari Sum of Absolute Difference . SAD inilah yang digunakan
untuk menyatakan ada atau tidaknya pergerakan suatu pasang
citra.
Metode frame differences menggunakan citra referensi
tertentu dalam mendeteksi gerak. Metode yang digunakan
dalam penggunaan citra referensi ini dikenal dengan nama
template matching . Ada dua metode penentuan template yang
digunakan, antara lain : static template matching (background
subtraction ) dan dynamic template matching.
Penelitian ini menggunakan metode dynamic template
matching dalam menentukan citra referensi. Metode dynamic
template matching dikembangkan dan dimodifikasi agar
adaptif terhadap perubahan lingkungan. Metode ini
selanjutnya disebut dengan nama dynamic and adaptive
template matching. Pendeteksian gerak dengan menggunakan
metode ini dikembangkan dan diimplementasikan untuk
aplikasi berbasis web, sehingga bahasa pemrograman yang
digunakan adalah bahasa pemrograman web.

II. PENELITIAN TENTANG DETEKSI GERAK

Penelitian yang dilakukan oleh Yong et al. [10]
menggunakan empat metode pendeteksian gerak. Metodemetode tersebut antara lain metode frame differences,
background subtraction , pixellate filter dan blob counter .
Metode frame differences menggunakan citra ke t-1 sebagai
citra referensi. Penelitian ini menggunakan bahasa
pemrograman C# dalam melakukan pendeteksian. Support
Vector Machine (SVM) juga diterapkan dalam mendeteksi
gerak [4]. Penelitian ini tidak hanya melakukan pendeteksian
gerak, akan tetapi juga melakukan segmentasi terhadap objekobjek tersebut. Pendeteksian ini dirancang dengan
menggunakan bahasa pemrograman C++ dan OpenCV.
Penelitian yang terkait dengan deteksi gerak juga
dilakukan oleh Zheng et al [11]. Penelitian ini menggunakan
metode frame differences yang digabungkan dengan
pengaturan ambang batas yang adaptif (adaptive threshold ).
Pendeteksian gerak juga direalisasikan dengan menggunakan
metode statistical correlation method [12]. Metode ini
digunakan setelah dilakukan proses temporal differences
dalam menganalisis beberapa frame citra. Pendeteksian gerak
dengan mengkombinasikan teknik frame differences dengan
optical flow. Metode ini dilanjutkan dengan teknik
morphological filter [13].
Penelitian yang dilakukan oleh [14] menerapkan konsep
vektor untuk mendeteksi pergerakan. Metode ini dilakukan
dengan membandingkan beberapa frame dan menandai titiktitik perbedaan antar frame. Metode ini juga menghasilkan
informasi tentang arah pergerakan objek. Zheng et al. [11]
menjelaskan bahwa terdapat metode lain yang digunakan
untuk mendeteksi gerak. Metode ini dikenal dengan nama
Statistical Learning Algorithm. Penelitian dengan menerapkan
metode serupa juga dilakukan oleh Murali dan Girisha [12].
Sama seperti optical flow, metode ini membutuhkan waktu
komputasi yang besar. Hal ini terjadi karena algoritma ini
membutuhkan langkah-langkah yang kompleks.
Terkait dengan teknik penentuan citra referensi, salah satu
metode pendeteksian gerak dilakukan dengan menggunakan
teknik double differences. Teknik komparasi citra ini
dikembangkan oleh Kameda dan Minoh [15]. Double
differences dilakukan dengan membandingkan citra dengan
waktu t dengan citra t-1, selanjutnya dilakukan pembandingan
kedua antara citra t-1 dengan citra t-2. Berbeda dengan
metode yang dikembangkan oleh Collins et al. [16],
pembandingan dilakukan antara citra t dengan citra t-1, dan
antara citra t dengan citra t-2.
Berdasarkan tinjauan tersebut, disimpulkan bahwa terdapat
banyak penelitian yang dilakukan dalam pendeteksian gerak
dengan menggunakan citra. Penelitian yang diajukan di dalam
paper ini memiliki perbedaan dalam hal metode, teknik
penentuan citra referensi dan bahasa pemrograman yang
digunakan dalam pendeteksian gerak.
III.

PENELITIAN DETEKSI GERAK

A. Frame Differences

Penelitian ini menggunakan metode frame differences
yang dilakukan dengan membandingkan piksel rata-rata
komponen RGB. Pengujian dengan langkah ini dilakukan
dengan menggunakan persamaan 1 dan persamaan 2.
g ( x, y)  g G ( x, y)  g B ( x, y)
g o ( x, y)  R
3
(1)
f R ( x, y)  f G ( x, y)  f B ( x, y)
f o ( x, y) 
3
( g o ( x, y)  T )  fo ( x, y)  ( g o ( x, y)  T )
(2)
g R , gG , g B
Dimana:
adalah komponen RGB citra yang

START

Capture Image

No
Set image as
reference

Yes

Motion Detection

f R , fG , f B

ditangkap pada waktu t, sedangkan
adalah
go
fo
komponen RGB citra referensi. dan adalah nilai rata-rata
dari penjumlahan nilai komponen warna RGB. T merupakan
threshold atau ambang batas nilai RGB.
Selanjutnya dilakukan penghitungan persentase piksel
objek yang terdeteksi. Pendeksian ini dilakukan dengan
menggunakan persamaan 3.

f
D

R
i

( x, y)   f i ( x, y)   f i ( x, y)

 f  f  f
G

R

G

B

100%

(3)

B

,  f i ( x, y) ,  i
adalah
terdeteksi, sedangkan  i
jumlah piksel yang terdeteksi berbeda berdasarkan komponen
Dimana D adalah persentase total perbedaan yang
R

G

f ( x, y)

B

f ( x, y)

 R  G  B adalah jumlah piksel
warna RGB.
keseluruhan citra yang diambil dari tiga komponen warna.
f ,

f ,

f

Already have
reference image?

Save as Reference Image
And Save the Detection
Coordinates (XY)

No
0,1% < D < 40%

Yes

Latest (XY) != Existing
(XY)

No

Delete All
the
References
Image

Yes
Mark the object and then
Save as detected image

END

Gbr 1. Diagram Alir Metode Dynamic and Adaptive Template Matching
(DATM)

IV. ANALISIS DETEKSI GERAK
A. Analisis Metode DTM t-1, Static Template Matching dan
DATM


Dynamic Template Matching t-1
Gambar 2 dan gambar 3 merupakan hasil tangkapan objek
yang teridentifikasi bergerak dengan menggunakan metode
DTM t-1. Kedua gambar ditangkap secara berurutan dengan
menggunakan IP Camera.
Pada gambar 2, gambar 2 (b) dan 2 (c) mendeteksi
pergerakan yang telah terdeteksi pada citra 2 (a) dan 2 (b).
Keadaan ini merupakan kelemahan dari metode DTM t-1. Jika
diteliti lebih lanjut, kelemahan semakin terlihat pada gambar 3.

B. Dynamic and Adaptive Template Matching

Terkait dengan penenetuan citra referensi, penelitian ini
mengajukan sebuah metode penentuan citra referensi dengan
teknik yang berbeda. Metode referensi yang digunakan adalah
kombinasi antara referensi pada saat t-1, t-n dan penetapan
citra referensi baru jika lingkungan (area tangkapan kamera)
mengalami perubahan yang signifikan.
Perubahan-perubahan signifikan yang dimaksudkan dapat
berupa: (1) perubahan kecerahan objek, dalam hal ini bisa
terjadi jika lampu ruangan dimatikan atau ruangan diterangi
cahaya matahari, (2) perubahan area tangkapan kamera,
kondisi ini terjadi jika posisi kamera diubah, (3) perubahan
kondisi lingkungan jika terdapat objek yang datang dan secara
statis berada di posisi tertentu secara terus-menerus.
Algortima penentuan citra referensi ini selanjutnya disebut
dengan nama teknik dynamic and adaptive template matching .
Algortima dynamic and adaptive template matching bekerja
berdasarkan diagram pada gambar 1.

Gbr 2. Motion Detection t-1 Reference Image

Gbr 3. Deteksi Gerak dengan Referensi t-1 Ditandai dengan Kotak Merah

Gambar 3 merupakan pendeteksian gerak dengan metode
DTM t-1 dengan dilengkapi penanda (marker) kotak merah.
Gambar 3 (c) menandai objek kosong yang sebelumnya
merupakan objek yang terdapat di gambar 3 (b). Pengujian ini
semakin menegaskan kelemahan metode DTM t-1 dalam
mendeteksi pergerakan.


Static Template Matching
Pengujian selanjutnya dilakukan dengan menggunakan
citra referensi yang telah ditetapkan oleh sistem. Metode ini

bisa disebut dengan nama static template matching. Pengujian
dengan menggunakan metode ini dapat dilihat pada gambar 4.

Gbr 4. Metode Static Template Matching

Pada gambar 4, sistem berhasil menanggulangi kelemahan
yang terjadi pada saat menerapkan metode DTM t-1. Setiap
citra yang terdapat pada gambar 4 berhasil mendeteksi objek
yang bergerak dan menandai persis pada area objek tersebut.
Namun, metode ini masih meninggalkan permasalahan
ketika suatu objek datang dan secara terus-menerus berada di
area tangkapan kamera secara statis. Hal ini terlihat pada
gambar 5. Terdapat objek (mouse) yang masuk pada gambar 5
(b). Objek tersebut dideteksi bergerak oleh hasil tangkapan
pada citra selanjutnya (gambar 5 (c), (d), (e)). Terlihat bahwa
sistem tidak adaptif terhadap perubahan area tangkapan. Hal
ini tentu akan menjadi masalah jika terjadi perubahan
kecerahan karena faktor cahaya.

adalah referensi saat t-1. Metode ini digunakan untuk kondisi
ketika sistem pertama kali menangkap pergerakan, maka
referensi t-1 akan bereperan sebagai pembanding untuk citra
yang terdeteksi tersebut. Jika pada frame selanjutnya sistem
masih mendeteksi pergerakan, maka referensi yang digunakan
adalah citra pada saat t-2. Jika pada frame selanjutnya masih
berlanjut dan mendeteksi pergerakan, maka citra referensi
yang digunakan adalah pada saat t-3, begitu seterusnya.
Metode ini disebut dengan teknik penentuan citra referensi
saat t-n. Metode t-n merupakan metode kedua yang digunakan
sistem. Dimana nilai n disesuaikan dengan urutan frame hasil
pendeteksian secara berurutan
Metode ketiga yang digunakan adalah penentuan ulang
citra referensi. Metode ini bekerja dengan cara menangkap
ulang citra referensi yang akan digunakan dan menghapus
semua citra refernsi yang telah ada. Citra referensi akan
berubah berdasarkan hasil pengujian koordinat objek yang
terdeteksi. Sistem akan menentukan ulang citra referensi jika
mendeteksi pergerakan di koordinat yang sama. Kesamaan
titik koordinat diuji terhadap dua frame citra yang dinyatakan
mendeteksi pergerakan secara berurutan.
Metode ketiga digunakan untuk menanggulangi
permasalahan yang muncul ketika menggunakan metode static
template matching . Hasil pengujian dengan menggunakan
metode
DATM
dapat
dilihat
pada
gambar
6.

Gbr 6. Hasil Pengujian Dynamic and Adaptive Template Matching

Gbr 5. Permasalahan Deteksi Gerak Pada Citra Referensi Statis

Berdasarkan pengujian yang dilakukan terhadap dua
metode penentuan citra referensi sebelumnya, terdapat
kelemahan yang terkait dengan metode penentuan citra
referensi. Rancangan dan diagram alir algoritma metode
penentuan citra referensi yang teradapat pada gambar 1,
merupakan algoritma yang modifikasi dari metode dynamic
template matching . Metode ini diperkenalkan dengan nama
Dynamic and Adaptive Template Matching (DATM). Metode
ini bekerja dengan cara melakukan penyesuaian citra referensi
berdasarkan kondisi area yang ditangkap.


Dynamic and Adaptive Template Matching
Secara umum, sistem ini bekerja dengan menggunakan
tiga metode penentuan citra referensi. Metode tersebut
berubah-ubah secara dinamis dan adaptif. Metode pertama

Gambar 6 merupakan urutan frame citra yang ditangkap
pada saat pengujian Dynamic And Adaptive Template
Matching (DATM). Urutan citra yang ditangkap diurutkan
berdasarkan angka yang terdapat di pojok kanan atas masingmasing citra. Pada gambar 6, citra 1, 4, 7, 10, dan 13
merupakan citra referensi yang secara adaptif berubah sesuai
dengan kondisi area yang ditangkap. Pada gambar terlihat
jelas adanya objek statis yang dalam waktu tertentu berada di
area tangkapan kamera tanpa melakukan pergerakan (diam).
Sehingga citra dengan objek tersebut dianggap sebagai
referensi oleh sistem. Penerapan metode DTM t-1 terlihat
ketika terdeteksi gerak pada citra 2, 5, 8, dan 1, dimana citra
referensi (1, 4, 7, 10 dan 13) yang digunakan merupakan citra
yang ditangkap sebelum citra tersebut ditangkap. Sedangkan
penerapan citra referensi t-n terlihat pada citra 3, 6, 9 dan 12.
Citra tersebut menggunakan citra referensi pada saat t-2.
Berdasarkan pengujian tersebut, metode DATM mampu
menyelesaikan maslaah yang dihadapi ketika menggunakan
metode DTM t-1 dan Static Template Matching.

B. Perbandingan Metode DTM t-1 dengan Metode DATM

Pengujian pada bagian ini dilakukan untuk menguji tingkat
akurasi dari metode pendeteksian yang dijelaskan pada bagian
sebelumnya. Metode yang akan dibandingkan adalah metode
DTM t-1 dan DATM. Metode static template matching tidak
dibandingkan karena metode ini tidak adaptif dengan
perubahan lingkungan. Oleh karena itu, pengujian ini hanya
dilakukan untuk metode dynamic template matching .
Pengujian dan pengumpulan data dilakukan selama tiga
hari. Setiap hari dilakukan pendeteksian selama 2 jam pada
saat jam kantor. Pengujian dilakukan dengan menggunakan
sebuah IP Camera berjenis PTZ (Pan Tilt Zoom) dengan merk
Vivotek. Kamera ditempatkan di lobi gedung jurusan Teknik
Elektro dan Teknologi Informasi FT UGM. Pengumpulan data
dilakukan dengan menjalankan algoritma deteksi gerak DTM
t-1 dan DATM secara bersamaan. Pengaturan fungsi frame
differences sistem untuk kedua algoritma disamakan. Dimana,
nilai threshold yang digunakan adalah 45, rentang nilai
persentase terdeteksi (D) adalah dari 0,5% - 40%, resolusi
citra yang digunakan adalah 256 × 192 piksel. Terkait dengan
jenis citra, citra yang digunakan adalah citra RGB, dan
metode frame differences yang digunakan adalah
perbandingan rata-rata piksel komponen warna RGB.
Hasil pendeteksian sistem diklasifikasikan menjadi dua
kondisi, yaitu: True Positive dan False Positive. True Positive
berarti sistem menangkap citra dan menandai objek yang
teridentifikasi bergerak. Sedangan False Positive merupakan
kondisi dimana sistem menangkap citra tetapi menandai objek
kosong, atau tidak ada objek yang teridentifikasi bergerak.
Hasil pengujian ini menghasilkan 1643 citra yang
mendeteksi pergerakan untuk metode DTM t-1. Sedangkan
metode DATM menghasilkan 1278 citra yang mendeteksi
pergerakan. Hasil analisis dan perbandingan dari kedua
metode ini dapat dilihat pada tabel 1. Perbandingan pada tabel
1 menjadi tabel acuan untuk tingkat akurasi pendeteksian yang
dilakukan oleh kedua metode pendeteksian gerak ini.
Table 1. Perbandingan Akurasi DTM -1 dan DATM
Metode
Perbandingan
DTM t-1
DATM
True Positive
1474
89,8%
1221
95,5%
False Positive
169
10,2 %
57
4,5%

Berdasarkan tabel 1 dapat disimpulkan bahwa metode
pendeteksian gerak dengan menggunakan teknik DATM
memiliki tingkat akurasi pendeteksian yang lebih tinggi
(95,5 %) jika dibandingan dengan metode DTM t-1 (89,8 %).
Metode DATM juga behasil menekan jumlah citra hasil
pendeteksian yang berpengaruh terhadap penggunaan
harddisk untuk pendeteksian.
Table 2 Perbandingan Hasil Pendeteksian Sistem
Metode
Hasil Deteksi
DTM t-1
DATM
Terdeteksi Sempurna
817
55,4%
713 58,3%
Terdeteksi Kurang
657
44,6%
508 41,7%
Sempurna

Pada tabel 2 hasil pendeteksian dikelompokkan menjadi 2
bagian yang dibedakan berdasarkan hasil penandaan. Kedua
bagian tersebut antara lain: Terdeteksi Sempurna dan
Terdeteksi Kurang Sempurna. Terdeteksi sempurna berarti
sistem mampu menandai objek dengan sempurna berdasarkan
pada koordinat objek tersebut. Sedangkan terdeteksi kurang
sempurna terjadi ketika sistem menandai objek tetapi
penandaan mengalami pelebaran. Sehingga tidak fokus
menandai objek yang diidentifikasi bergerak. Dari tabel 2
terlihat bahwa metode DTM t-1 menghasilkan 55,4% untuk
keadaan Terdeteksi Sempurna dan 58,3 % untuk metode
DTAM.
V. KESIMPULAN DAN SARAN
Metode pendeteksian gerak Dynamic and Adaptive
Template Matching (DATM) secara dinamis dan adapatif
menjadi solusi terhadap kelemahan yang terjadi ketika
menggunakan metode penentuan citra referensi pada saat t-1
(DTM t-1) dan static template matching. Pengujian
pendeteksian gerak yang dilakukan menghasilkan suatu
metode pendeteksian gerak dengan menggunakan teknik
frame differences. Dimana metode penentuan citra referensi
yang digunakan dikenal dengan Dynamic and Adaptive
Template Matching (DATM). Berdasarkan perbandingan
metode DTM t-1 dengan DATM, metode DTAM memiliki
tingkat akurasi pendeteksian 95,5 %, sedangan metode DTM
t-1 hanya memiliki akurasi 89,8 %.
Penelitian lanjutan yang akan dilakukan adalah segmentasi
objek. Sehingga setiap objek yang terdeteksi bisa
dikelompokkan berdasarkan kedekatan titik piksel yang
terdeteksi. Dengan demikian sistem dapat mengenali jumlah
objek yang terdeteksi.
DAFTAR PUSTAKA
[1] Sumita Mishra, Prabhat Mishra, Naresh K Chaudhary, and
Pallavi Asthana, "A Novel Comprehensive Method for Real
Time Video Motion Detection Surveillance," International
Journal of Scientific & Engineering Research Volume 2, Issue
4, 2011.
[2] P. Spagnolo, T. D'Orazio, M.Leo, and A.Distante, "Moving
Object Segmentation by Background Subtraction and Temporal
Analysis," Image and Vision Computing , vol. 24, pp. 411-423,
May 2006.
[3] Zhen Tang and Zhenjiang Miao, "Fast Background Subtraction
Using Improved GMM and Graph Cut," in Congress on Image
and Signal Processing, 2008. CISP '08. , 2008, pp. 181 - 185.
[4] Hongyan Li and Hongyan Cao, "Detection and Segmentation of
Moving Objects Based on Support Vector Machine," in 2010
Third International Symposium on Information Processing ,
Shandong China, 2010, pp. 193-197.
[5] M. Allili, M.-F. Auclair-Fortier, P. Poulin, and D. Ziou, "A
Computational Algebraic Topology Approach for Optical
Flow," in ICPR '02 Proceedings of the 16 th International
Conference on Pattern Recognition (ICPR'02) Volume 1 Volume 1 , Washington DC, USA, 2002.
[6] J Gallego, M. Pardas, and J.-L. Landabaso, "Segmentation and

[7]

[8]

[9]

[10]

[11]

Tracking of Static and Moving Objects in Video Surveillance
Scenarios," in ICIP 2008. 15th IEEE International Conference
on Image Processing , 2008, pp. 2716 - 2719.
Ho Gi Jung, Jae Kyu Suhr, Kwanghyuk Bae, and Jaihie Kim,
"Free Parking Space Detection Using Optical Flow-based
Euclidean 3D Reconstruction," in Proceedings of the IAPR
Conference on Machine Vision Applications (IAPR MVA 2007) ,
Tokyo, Japan, 2007, pp. 16-18.
Jens Klappstein, Tobi Vaudrey, Clemens Rabe, Andreas Wedel,
and Reinhard Klette, "Moving Object Segmentation using
Optical Flow and Depth Information," Advances in Image and
Video Technology, 2009.
H H Kenchannavar, Gaurang S Patkar, U P Kulkarni, and M M
Math, "Simulink Model for Frame Difference and Background
Subtraction comparision in Visual Sensor Network," in 2010
The 3rd International Conference on Machine Vision (ICMV
2010), Hongkong China, 2010.
Yee Ching Yong, Rubita Sudirman, and Kim Mey Chew,
"Motion Detection and Analysis with Four Different Detectors,"
in 2011 Third International Conference on Computational
Intelligence, Modelling and Simulation , Langkawi, 2011, pp.
46-50.
Xiaoshi Zheng, Yanling Zhao, Na Li, and Huimin Wu, "An
Automatic Moving Object Detection Algorithm for Video
Surveillance Applications," in 2009 International Conference
on Embedded Software and Systems , Hangzhou China, 2009,
pp. 541-543.

[12] S Murali and R Girisha, "Segmentation of Motion Objects from
Surveillance Video Sequences using Temporal Differencing
Combined with Multiple Correlation," in 2009 Sixth IEEE
International Conference on Advanced Video and Signal Based
Surveillance, Genova, Italy , 2009, pp. 472-477.
[13] li Fang, Zhang Meng, Claire Chen, and Qian Hui, "Smart
Motion Detection Surveillance System," in 2009 International
Conference on Education Technology and Computer ,
Singapore, 2009, pp. 171-175.
[14] Takanori Yokoyama, Toshiki Iwasaki, and Toshinori Watanabe,
"Motion Vector Based Moving Object Detection and Tracking
in the MPEG Compressed Domain," in 2009 Seventh
International Workshop on Content-Based Multimedia Indexing ,
Chania, Crete, 2009, pp. 201-206.
[15] Y Kameda and M Minoh, "A Human Motion Estimation
Method Using 3-Successive Viedo Frames," in International
Conference on Virtual Systems and Multimedia , 1996, pp. 135–
140.
[16] Robert T. Collins et al., "A System for Video Surveillance and
Monitoring," 2000.

I. INTRODUCTION
The camera is device is used to monitor an area or object.
There are various types of cameras used for monitoring that
aim for security system. One of the camera monitors are
commonly used in monitoring is the IP Camera. Application
of a monitoring system has features motion detection based on
the detected image. Motion detection is done by analyzing the
images captured.
Motion detection mechanism starts from the determination
of the reference image by image comparison. The image
contrast is considered as the normal condition of a room. The
image is compared with the imagery condition after the arrest.
Image capture process carried out at regular intervals in
accordance with the requirements of the system.
According to a study conducted by Mishra et al. [1], there
are three methods that commonly is used to detect a motion.
They are background subtraction, optical flow and temporal
differences. Background subtraction is done by comparing a
specific image with the image used as a reference.
Background subtraction makes motion detection by using the
technique of determining the reference image statically [2, 3,
4].
Study [5, 6, 7, 8] use optical flow in research on motion
detection performed. The study is quite difficult to implement
for real time video surveillance. The application of optical
flow requires additional hardware to support the performance
and monitoring systems. Method of temporal differences is
also known by the name of the frame differences. This method
is performed by comparing image frames are captured.
Another study conducted by Kenchannavar et al. [9] describes
the algorithm implemented in the method of background
subtraction and frame differences. Research carried out by
applying the concept of SAD. SAD stands for Sum of
Absolute Difference. SAD is used to declare whether or not
the movement of an image pair.
Frame differences using the method specified in the
reference image to detect motion. The method applied in using
of a reference image is known as template matching. There
are two methods of determination of the template uses; static
template matching (background subtraction) and dynamic
template matching
This study uses a dynamic template matching method in
determining the reference image. Dynamic template matching
method is developed and modified to be adaptive to
environmental changes. This method is here in after referred
to as the dynamic and adaptive template matching. Motion
detection using this method was developed and implemented
for web-based applications, so the programming language
used is a web programming language.
II. RELATED WORKS
Research conducted by Yong et al. [10] using four
methods of motion detection. These methods include methods
of frame differences, background subtraction, pixellate filter
and blobcounter. Frame differences method uses an image to
t-1 as the reference image. This study uses the C# to do the

detection. Support Vector Machine (SVM) is also applied in
the motion detection [4]. This study is not only performing
motion detection, but also the segmentation of the objects.
This detection is designed by using the C++ and OpenCV.
Research related to motion detection is also performed by
Zheng et al. [11]. This study uses frame differences are
coupled with an adaptive threshold setting (adaptive
threshold). Motion detection is also realized by using the
method of statistical correlation method [12]. This method is
used after the process of analysing the temporal differences in
some of the image frame. The detection of motion by
combining the frame differences technique with optical flow.
This method is followed by a morphological filter technique
[13].
Research conducted by Yokoyama et al. [14] also applying
the concept of vectors to movement detection. This method is
performed by comparing multiple frames and marks the points
of difference between the frames. This method also yields
information about the direction of movement of the object.
Zheng et al. [11] explains that there are other methods used to
detect motion. This method is known as the Statistical
Learning Algorithm. Research by applying a similar method is
also performed by Murali and Girisha [12]. Just like optical
flow, this method requires large computational time. It occurs
because these algorithms require complex steps
Regarding with the technique of determining the reference
image, one of the methods of motion detection was done by
using double differences technique. Image comparison
technique was developed by Kameda and Minoh [15]. Double
differences done by comparing the image with time t to image
t-1, then performed a second comparison between the image t1 with the image of t-2. In contrast to the method developed
by Collins et al. [16], the comparisons were made between the
image t with the image of t-1, and the image t with the image
of t-2.
Based on such review, it was concluded that there are a lot
of research done in the motion detection by using the image.
The research that proposed in this paper have differences in
the methods, determination techniques of the reference image
and programming language used in the motion detection.
III. PROPOSED MOTION DETECTION ALGORITHM
A. Frame Differences

This study uses frame differences method by comparing
the average RGB components of each pixel. Equation 1 and
equation 2 are used for the calibration step.
g ( x, y)  g G ( x, y)  g B ( x, y)
g o ( x, y)  R
3
(1)
f R ( x, y)  f G ( x, y)  f B ( x, y)
f o ( x, y) 
3
( g o ( x, y)  T )  fo ( x, y)  ( g o ( x, y)  T )

Where:

g R , gG , g B
f R , fG , f B

(2)

are component of RGB images captured

the RGB components of the reference
at time t,
go
fo
image.
and
the average value of the sum of the RGB

colour components. T is the threshold or thresholds RGB
value and then performed counting the percentage of pixels
detected object. The detection is done using equation 3.

weakness of the method of DTM t-1. When examined further,
the weakness can be seen in figure 3.

B. Dynamic and Adaptive Template Matching

Regarding with the determination reference image, this
study proposed a method of determining the reference image
with different techniques. Reference method that used is a
combination of reference at the time t-1, t-n and the
establishment of a new reference image if the environment
(captured area of the camera) undergoing significant change.
Significant changes are intended to be: (1) changes in the
brightness of the object; in this case it can occur if the room
lights turned off, or the sun-lit room; (2) changes in captured
area of the camera, this condition occurs when the camera
position is changed; (3) changes in environmental conditions
if there are objects come and statically bases in a particular
position on an ongoing. The algorithm to determining the
reference image is referred to as the Dynamic and Adaptive
Template Matching. Dynamic algorithms and adaptive
template matching works based on the diagram in figure 1 .

Figure 9. Motion detection with reference t-1 is characterized by the Red
Box

Figure 3 is motion detection with DTM t-1 method is
equipped with a red box marker. Figure 3 (c) marks the empty
object that an object previously contained in figure 3(b). This
test further confirms the weakness of DTM t-1 methods in
detecting movement.


Static Template Matching
Testing is then performed using a reference image that
has been set by the system. This method can be called by the
name of the static template matching. Testing that using this
method can be seen in Figure 4.

START

Capture Image

No
Set image as
reference

Already have
reference image?

Yes

Motion Detection

Save as Reference Image
And Save the Detection
Coordinates (XY)

No
0,1% < D < 40%

Yes

Latest (XY) != Existing
(XY)

No

Delete All
the
References
Image

Yes

Figure 10. Static Template Matching Method
Mark the object and then
Save as detected image

END

Figure 7. Dynamic and Adaptive Template Matching (DATM) Method
Flow Chart

IV. MOTION DETECTION ANALYSIS

In figure 4, the system managed to overcome weaknesses
that occur when applying the DTM t-1 method. Each image
contained in Figure 4 successfully detects moving objects and
mark exactly on the object area.

A. DTM t-1, Static Template Matching and DATM Methods


Dynamic Template Matching t-1
Figure 2 and figure 3 is the result of the moving objects
captured that identified by using the DTM method t-1. Both
images are captured sequentially by using the IP Camera.

Figure 8. Motion Detection t-1 Reference Image

Figure 11. Error on motion detection with Static Template Matching Method

In Figure 2, figure 2 (b) and 2 (c) motion detection has
been detected in the image 2 (a) and 2 (b). This situation is a

However, this method still leaves the problem when an
object comes and constantly stated a static in camera captured

area. This can be shown in figure 5. There are objects (mouse)
are entered in figure 5 (b). The object is detected as moving
by the result of the captured on the next image (figure 5 (c),
(d), (e)). It can be seen that the system is not adaptive to
changes in catchment area. This will certainly be a problem if
the brightness changes because of the light.
Based on tests performed on two methods of determining
the reference image before, there are drawbacks regarding
with the method of determining the reference image. The
design of new algorithm in determining the reference image is
shown on Figure 1. This algorithm is a modification of the
method of dynamic template matching. This method was
introduced by the name of Dynamic and Adaptive Template
Matching (DATM). This method works by adjusting the
reference image based on the condition of the captured area.


Dynamic and Adaptive Template Matching
In general, this system works by using three methods of
determining the reference image. The method is dynamically
changing and adaptive. The first method is the reference at the
time t-1. This method is used for conditions when the system
first captures the movement, then the reference t-1 will have a
role as a comparison to the detected image. If the next frame
the system still detects movement, the references used are the
images at t-2. If the next frame and detect the movement
continues, then the reference image is used as t-3, and so on.
This method is called with the technique of determining the
reference image when t-n. T-n method is the second method
that used by the system. The value of n is adapted to the result
of the detection the sequence of frames.
The third method is used by determining the new of the
reference image. This method works by capturing a reference
image to be re-used and remove all existing reference image.
Reference image is changed based on results of testing the
coordinates of detected objects. The system will redefine the
reference image if it detects movement in the same coordinate.
The similarities of the point coordinates are tested on two
image frames are declared movement detection in a sequence.
The third method used to overcome the problems that arise
when using the static template matching method. The test
results using the DATM method can be seen in figure 6.
Figure 6 is a sequence of image frames are captured at the
time of testing Dynamic and Adaptive Template Matching
(DATM) method. Sequence of images captured is sorted by
the numbers in the upper right corner of each image. In the
figure 6, the image of 1, 4, 7, 10, and 13 are reference image
is adaptively changed according to the condition of the
captured area. It clearly visible of a static object in a certain
time in the catchment area without moving the camera. Thus
the image of the object is considered as a reference by the
system. Applying the method of DTM t-1 can be seen when
motion is detected in the image of 2, 5, 8, and 1, where the
reference image (1, 4, 7, 10 and 13) that used a captured
image before the image is captured. While the applying of the
reference image t-n, it can be seen in 3, 6, 9 and 12 images.
Based on these tests, a DATM method capable of resolving a
problem encountered when using the DTM t-1 and Static
Template Matching.

Figure 12. Dynamic and Adaptive Template Matching Testing Results

B. The Comparison of DTM t-1 Method with DATM Method

Testing in this section was conducted to examine the
accuracy of the detection methods described in the previous
section. Methods to be compared dare the method of t-1 and
DTM. Static template matching methods are not compared
because the method is not adaptive to the environmental
changing. Therefore, this test is only performed for dynamic
template matching method.
Testing and data collection conducted over three days.
Every day it takes 2 hours to do detection during office hours.
Experiments were done by using a type IP Camera PTZ (Pan
Tilt Zoom) with Vivotek brands. The camera is placed in the
lobby of the building department of Electrical Engineering
and Information Technology UGM. The data was collected by
running the motion detection algorithm DTM t-1 and DATM
simultaneously. The Setting function of frame differences
system for both algorithms comparable. That is, the threshold
value used is 45, the range of percentage values was detected
(D) is from 0.5% - 40%, the image resolution used is 256
×192 pixels. Regarding with this type of image, the image
used is RGB image, and the methods used frame differences is
the comparison of the average pixel RGB colour components.
The results of the detection system are classified into two
conditions, namely: True Positive and False Positive. True
Positive means the system captures the image and mark the
identified moving objects. Whereas, False Positive is a
condition in which the system captures the image, but mark
the empty object or no identifiable objects moving.
There are 1643 images that detects the movement when
used t-1 DTM method. While this method produces 1278
images movement for DATM method. The analysis and
comparison of these two methods can be seen in table 1.
Table 3. Comparison of DTM -1 And DATM Accuracy
Method
Comparison
DTM t-1
DATM
True Positive
1474
89,8%
1221
95,5%
False Positive
169
10,2 %
57
4,5%

According to the table 1 it can be concluded that the
motion detection method using DATM technique has
detection accuracy rate is higher (95.5%) when compared with
the DTM method t-1 (89.8%). DATM method can also
suppress the image of the detection result of the influence on

storage usage. The storage usage case will be concern to the
motion detection system for future research.
Table 4. Comparison of detection system result

Detection
Perfect Detection
Less Perfect
Detection

Method
DTM t-1
DATM
817
55,4% 713 58,3%

[6]

657

[7]

44,6%

508

41,7%

In table 2 the results of detection are grouped into two
parts which are distinguished based on labelling. Both parts
include: Detected Less Perfect and Perfect. Perfect detected
means the system is able to perfectly detected objects based
on the coordinates of the object. While detected Less Perfect
occur when the system mark object but the tagging system is
experiencing widening. So it is not the focus of identified
moving object mark. Based on table 2 show that the DTM t-1
method produced 55.4% for the state and 58.3% detected
Perfect for DTAM method.
V. CONCLUSION AND FUTURE WORKS
Method of motion detection Dynamic and Adaptive
Template Matching (DATM) dynamically and adaptive be a
solution to the weakness that occurs when using the method of
determining the reference image at time t-1 (DTM t-1) and the
static template matching. Motion detection test is performed
to produce a method of motion detection by using frame
differences. Where the reference image determination method
used is known as Dynamic and Adaptive Template Matching
(DATM). Based on the comparison of both methods, DATM
method has detection accuracy rate 95.5%, Whereas, DTM t-1
method only has an accuracy of 89.8%.
However, further research will be done is the object
segmentation. So that each detected object can be grouped
by the proximity of the detected pixel point. Thus the system
can identify the number of detected objects.

[8]

[9]

[10]

[11]

[12]

[13]

REFERENCES
[14]
[1] Sumita Mishra, Prabhat Mishra, Naresh K Chaudhary, and
Pallavi Asthana, "A Novel Comprehensive Method for Real
Time Video Motion Detection Surveillance," International
Journal of Scientific & Engineering Research Volume 2, Issue
4, 2011.
[2] P. Spagnolo, T. D'Orazio, M.Leo, and A.Distante, "Moving
Object Segmentation by Background Subtraction and Temporal
Analysis," Image and Vision Computing , vol. 24, pp. 411-423,
May 2006.
[3] Zhen Tang and Zhenjiang Miao, "Fast Background Subtraction
Using Improved GMM and Graph Cut," in Congress on Image
and Signal Processing, 2008. CISP '08. , 2008, pp. 181 - 185.
[4] Hongyan Li and Hongyan Cao, "Detection and Segmentation of
Moving Objects Based on Support Vector Machine," in 2010
Third International Symposium on Information Processing ,
Shandong China, 2010, pp. 193-197.
[5] M. Allili, M.-F. Auclair-Fortier, P. Poulin, and D. Ziou, "A

[15]

[16]

Computational Algebraic Topology Approach for Optical
Flow," in ICPR '02 Proceedings of the 16 th International
Conference on Pattern Recognition (ICPR'02) Volume 1 Volume 1 , Washington DC, USA, 2002.
J Gallego, M. Pardas, and J.-L. Landabaso, "Segmentation and
Tracking of Static and Moving Objects in Video Surveillance
Scenarios," in ICIP 2008. 15th IEEE International Conference
on Image Processing , 2008, pp. 2716 - 2719.
Ho Gi Jung, Jae Kyu Suhr, Kwanghyuk Bae, and Jaihie Kim,
"Free Parking Space Detection Using Optical Flow-based
Euclidean 3D Reconstruction," in Proceedings of the IAPR
Conference on Machine Vision Applications (IAPR MVA 2007) ,
Tokyo, Japan, 2007, pp. 16-18.
Jens Klappstein, Tobi Vaudrey, Clemens Rabe, Andreas Wedel,
and Reinhard Klette, "Moving Object Segmentation using
Optical Flow and Depth Information," Advances in Image and
Video Technology, 2009.
H H Kenchannavar, Gaurang S Patkar, U P Kulkarni, and M M
Math, "Simulink Model for Frame Difference and Background
Subtraction comparision in Visual Sensor Network," in 2010
The 3rd International Conference on Machine Vision (ICMV
2010), Hongkong China, 2010.
Yee Ching Yong, Rubita Sudirman, and Kim Mey Chew,
"Motion Detection and Analysis with Four Different Detectors,"
in 2011 Third International Conference on Computational
Intelligence, Modelling and Simulation , Langkawi, 2011, pp.
46-50.
Xiaoshi Zheng, Yanling Zhao, Na Li, and Huimin Wu, "An
Automatic Moving Object Detection Algorithm for Video
Surveillance Applications," in 2009 International Conference
on Embedded Software and Systems , Hangzhou China, 2009,
pp. 541-543.
S Murali and R Girisha, "Segmentation of Motion Objects from
Surveillance Video Sequences using Temporal Differencing
Combined with Multiple Correlation," in 2009 Sixth IEEE
International Conference on Advanced Video and Signal Based
Surveillance, Genova, Italy , 2009, pp. 472-477.
Li Fang, Zhang Meng, Claire Chen, and Qian Hui, "Smart
Motion Detection Surveillance System," in 2009 International
Conference on Education Technology and Computer ,
Singapore, 2009, pp. 171-175.
Takanori Yokoyama, Toshiki Iwasaki, and Toshinori Watanabe,
"Motion Vector Based Moving Object Detection and Tracking
in the MPEG Compressed Domain," in 2009 Seventh
International Workshop on Content-Based Multimedia Indexing ,
Chania, Crete, 2009, pp. 201-206.
Y Kameda and M Minoh, "A Human Motion Estimation
Method Using 3-Successive Viedo Frames," in International
Conference on Virtual Systems and Multimedia , 1996, pp. 135–
140.
Robert T. Collins et al., "A System for Video Surveillance and
Monitoring," 2000.