Hand Motion Gesture for Human-Computer Interaction Using Support Vector Machine and Hidden Markov Model - Binus e-Thesis

  ISSN 1828-6003 Vol. 11 N. 5 May 2016

  International Review on Computers and Software (IRE COS) Contents: Hand Motion Gesture for Human-Computer Interaction

  374 Using Support Vector Machine and Hidden Markov Model by Suharjito, F itra B. A dinugraha F uzzy E xpert System for Classifying Pests and Diseases of Paddy 381 Using Bee Colony Algorithms by Yovita Tunardi, Suharjito A F ramework for Social Media and Text-Based Content Analysis 388 for E vent Management Purposes by Qusai A buein, Mohammed Q. Shatnawi, Muneer Bani Yassein, Radwan Batiha Optimized Implementation of H.264/ AVC Motion E stimation on a Mixed Architecture

  395 Using SynDE x-Mix by Oussama F ek i, Thierry Grandpierre, N ouri Masmoudi, Mohamed A k il Integration E lectronic Patients’ Records with Open Life Sciences Datasets 403 Using Semantic Web Tools by Bassam N ajeeb, Bassel A l Khatib E nhancing Relay Selection Scheme for Connecting VANE Ts to Internet Over IE E E 802.11p 410 in Congested and F ading E nvironment Scenarios by D. A bada, A . Massaq, A . Boulouz Key F rames Based Video Authentication Using F ragile Watermarking 420 and Singular Value Decomposition by A ssma A zeroual, Karim A fdel Localization of Mobile Stations from ONE Base Station in GSM Systems 427 by Khalid G. Samarah Word E xtraction from Arabic Handwritten Documents Based on Statistical Measures 436 by A yman A l-Dmour, Raed A bu Zitar A New Compression Scheme Based on Adaptive Vector Quantization

  445 REPRINT and Singular Value Decomposition by Imene Soussi, Mohamed Ouslim Real-Time E lectroencephalography-Based E motion Recognition System 456 by Riyanarto Sarno, Muhammad N adzeri Munawar, Brilian T. N ugraha

  

International Review on Computers and Software

(IRECOS)

Editor-in-Chief: Prof. Marios Angelides

  Brunel University London Department of Electronic and Computer Engineering Uxbridge, Middlesex - UB8 3PH, U.K.

  Dr. Harry Agius Co-Editor:

  Brunel University London Department of Electronic and Computer Engineering Uxbridge, Middlesex - UB8 3PH, U.K.

  Editorial Board: Mikio Aoyama (Japan) Pascal Lorenz (France) Francoise Balmas Marlin H. Mickle

  (France) (U.S.A.) Vijay Bhatkar (India) Ali Movaghar (Iran) Arndt Bode Dimitris Nikolos

  (Germany) (Greece) Rajkumar Buyya (Australia) Mohamed Ould-Khaoua (U.K.) Wojciech Cellary Witold Pedrycz

  (Poland) (Canada) Bernard Courtois (France) Dana Petcu (Romania) Andre Ponce de Carvalho Erich Schikuta

  (Brazil) (Austria) David Dagan Feng (Australia) Arun K. Somani (U.S.A.) Peng Gong Miroslav Švéda

  (U.S.A.) (Czech) Defa Hu (China) Daniel Thalmann (Switzerland) Michael N. Huhns Luis Javier García Villalba

  (U.S.A.) (Spain) Ismail Khalil (Austria) Brijesh Verma (Australia) Catalina M. Lladó Lipo Wang

  (Spain) (Singapore) The International Review on Computers and Software (IRECOS) is a publication of the Praise Worthy Prize S.r.l.. The Review is published monthly, appearing on the last day of every month. Published and Printed in Italy by Praise Worthy Prize S.r.l., Naples, May 31, 2016.

  Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved.

  

This journal and the individual contributions contained in it are protected under copyright by Praise Worthy Prize S.r.l.

and the following terms and conditions apply to their use: Single photocopies of single articles may be made for personal use as allowed by national copyright laws.

  

Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic

copying, copying for advertising or promotional purposes, resale and all forms of document delivery. Permission may be

sought directly from Praise Worthy Prize S.r.l. at the e-mail address: administration@praiseworthyprize.com

  REPRINT

Permission of the Publisher is required to store or use electronically any material contained in this journal, including any

article or part of an article. Except as outlined above, no part of this publication may be reproduced, stored in a retrieval

system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without

prior written permission of the Publisher. E-mail address permission request: administration@praiseworthyprize.com Responsibility for the contents rests upon the authors and not upon the Praise Worthy Prize S.r.l..

  

Statement and opinions expressed in the articles and communications are those of the individual contributors and not the

statements and opinions of Praise Worthy Prize S.r.l.. Praise Worthy Prize S.r.l. assumes no responsibility or liability

for any damage or injury to persons or property arising out of the use of any materials, instructions, methods or ideas

contained herein.

  

Praise Worthy Prize S.r.l. expressly disclaims any implied warranties of merchantability or fitness for a particular

purpose. If expert assistance is required, the service of a competent professional person should be sought.

ISSN 1828-6003

  May 2016

Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved DOI: 10.15866/irecos.v11i5.8641

  

Hand Motion Gesture for Human-Computer Interaction

Using Support Vector Machine and Hidden Markov Model

Suharjito, Fitra B. Adinugraha

  Abstract

• – Hand gesture recognition for human-computer interaction has become very popular

  in recent times. The main problem of this technology is how the system can recognize the presence

  of a gesture in a streaming video. In this paper, we propose a model that can recognize hand

  motion gesture in avideo stream using Support Vector Machine and Hidden-Markov model.

  Support Vector Machine has the advantage of generalizing classification. On the other hand,

  Hidden Markov Model is a statistical model that is capable of modeling Spatial-temporal time

  series. This system is divided into two main processes. First, this system recognizes hand posture

  using SMV (static gesture recognition) and generates sequence observation which is used for the

  second process later. The second process is recognizing dynamic hand gesture with the sequence

  observation from the static gesture. The implementation shows that static hand gesture recognition

  achieves average accuracy at 91% using testing dataset. Meanwhile,dynamically isolated hand

  gesture recognition gets average accuracy at 89% using testing dataset. We also have tested the

  system with continuous dynamic gesture using video stream. The system can recognize the gesture

  very well with accuracy of 83%. This achievement shows that the model can be used in human

  computer-interaction with specific supports such as Vector Machine and Hidden Markov Model

  parameter. Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved.

  Keywords: Gesture Recognition, Hand Gesture Recognition, Human Computer Interaction I.

  Introduction

  In daily life, gesture becomes one of main elements of non-verbal communication. We use gesture to convey some meaningful information. Gesture recognition is the process in which gestures made by a user is made known to the system. On the other hand, human-computer interaction using motion gesture provides something more natural, innovative and user-friendly [1]. Gesture recognition has a wide range of applications, such as virtual reality, augmented reality, sign language recognition, video games, and medical application [2].

  Recognition using the hand can be done using the glove-based technique, device accelerator-based [3] and visual-based technique [2]. While glove-based and accelerator device sensing are very effective tools for capturing hand motion, they are very expensive, unnatural, and difficult to calibrate and setup the procedure. On the other hand, the vision-based technique provides natural gesture. It is inexpensive but still has many challenges to improve the accuracy and speed [2].

  One of the problems to achieve accuracy is how we can extract the right feature that represents the hand shape. Hand gesture recognition can be classified into two types, static and dynamic. The static gesture is a configuration or pose of the hand represented by a single image while the dynamic gesture is represented by more than one image [2]. In this paper, we focus on problems regarding both of them.

  After we have acquired static hand gesture, we use the result to create a feature for the dynamic hand gesture. In the static hand gesture, we use Support Vector Machine (SVM) to classify single hand pose image with another pose. Then, Hidden Markov Model (HMM) is used to model dynamic hand gesture from the sequence of pose hand image.

  II. Related Work

  In the last decades, there has been some research that performs hand gesture recognition. Some of them employ device based recognition like glove-based gesture recognition [4], This glove-based gesture recognition is very efficient in recognizing gesture because the glove contains embedded sensors that can detect motion very well. However, this method is uncomfortable for the hand because this glove must be attached to the wire. The other method using device- based is MEMS accelerator recognition [3]. In this method, the author tries to recognize motion gesture through a device that can detect acceleration of hand motion. Similar tothe glove-based gesture recognition, this method is also efficient. However, we cannot recognize the shape of the hand.

  SVM is apopular model used for classification and regression. In gesture recognition, SVM is commonly paired with another feature extraction or model like HU

  REPRINT moments feature extraction. It has been used to classify We segment image color using YCrCb color space hand shape for static hand gesture [5].This pairing information first and select the appropriate threshold method has an advantage of extracting moment of the algorithm with additional threshold configuration to be hand efficiently, such as hand orientation, angle, center calibrated in the current environment. The default moment. The other method that has been paired with threshold in the Eq. (1) [11]. Later, we will remove the SVM is Fourier descriptor used by Gamal et al. [2]. noise using dilate and mean filtering to smooth the

  This method has some advantages, including HU images. This process left the image of binary color: moments with additional fast processing because Fourier descriptor only uses a limited number of feature for each 77 ≤ Cb ≤ 127 and 133 ≤ Cr ≤ 173 image. SVM is also often used to be combined with

  (1) another classification model as proposed by Demidova. where Y Cb Cr,, [0,255] He combines SVM and fuzzy clustering algorithms [6].

  Then, AbAzziz, who combines Fast Artificial Immune and SVM for maximum load margin improvement [7]. The other method like HMM has been used to develop Arabic continuous hand gesture from color image sequence by applying trajectory of a single hand [8]. Malgireddy et al. [9] develop dynamic hand gesture. They suggest a model explicitly (gesture grammar) and use it to learn a model for each gesture. This method

  (a) (b)

  divides gesture to sub-gestures and then uses the sub- gestures to recognize the start point and end point of gesture. With this method, Malgireddy uses HMM that has been modified to use multiclass model and compare HCRF. Hsiang-Yueh [10] and Hui-Shyong Yeo [11] suggestthe use of YCrCb color space and convexity defect character point to extract fingertip position and perform robust dynamic hand gesture recognition with

  (c) (c)

  simple algorithms. SVM HMM is also commonly used to

  Figs. 1. Hand Segmentation,(a) original image, (b) filter color image

  be combined with other classification models, like

  (c) smooth image proposed by Regan and Srivatsa.

  They present a hybrid Canny Edge Detection and The next step iscalculating contour of the hand

  Weibull Probability Density Function based on forfurther use in feature extraction. Hierarchical Dirichlet Process HMM [12]. In this paper, we focus on combining the base on these effective _{Image Frame Filterd Image} methods. First, we suggest combining these advantages _{Capture Image Filtering Feature Extraction (YCrCb, MEDIAN, (CONVEX HULL, CONVEXITY} from the previous research, such as using sub-gesture _{Video COUNTOUR) DEFECTS, FINGERTIP)} from Malgireddy [9] with little modification.

  We only use a single image to detect static gesture and ^{Feature Vector} _{1:52 2:31 3:42 ... 1:50 2:32 3:40 ...} then perform the static gesture usingSVM. Second, the used method is performing extract feature using _{Class Label (1-5-5-2-1-1-2)} convexity defect method from Hsiang-Yueh to get robust Motion Classification Hand Classification _{Gesture Prediction (HMM) (SVM)} finger detection. Then, we also use HMM to recognize _{(Left Click)} continuous dynamics hand motion gesture. This process _{Models Models SVM} will be explained in the methodology. _{Module Training Module Training Topology HMM (A, B, ) (alpha, bias, SV, Kernel)}

III. Methodology

  _{Motion Classification Hand Classification (HMM) (SVM)}

  In this section, we will present the proposed model of human-computer interaction using hand motion gesture

  Fig. 2. Proposed model of hand motion gesture recognition recognition. The system is divided into four processes.

  REPRINT

  There are preprocessing, feature extraction, hand static III.2. gesture and hand dynamic gesture recognition. Fig. 2 Feature Extraction Using Convexity Defect shows the proposed model.

  Our proposed feature extraction method evolved from previous research regarding hand tracking and gesture recognition system [10], [11]. The result of preprocessing III.1.

   Preprocessing

  is the contour of the hand. The contour is a list of the The purpose of this process is to prepare images for point that represents a curve in an image. the next phase.

  

Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved International Review on Computers and Software, Vol. 11, N. 5

  Now, we have a set of feature hand shape images. Then, the next step is to run the SVM classifier model and get the best parameter to classify the images. SVM is a supervised machine learning method used for classification and regression.

  We calculate the maximum inscribed circle using Voronoi diagram [13]. The maximum point of inscribed circle lies on the point generated from the Voronoi diagram. Next, the convexity defect will be illustrated in Fig. 3. In this paper, we only use several convexity defect points that match the following criteria:

  This method has andisadvantage. When the class that will be classified multiplies, the time to classify will increase, that can affect the performance of the application. Moreover, the strategy of one-vs-all SVM does createa model in which the model will detrain with all class of existing dataset so that the decision will depend on the model [14].The other ability of the SVM is providing a transformation dataset from dimensional data for other higher dimensional data defined by kernel function [6].

  First, we calculate polygon represented the approximate to build contour [11]. Second, it is to extract interior of the hand contour to gain more information about hand shape image. This process can be done by calculating palm center, and extracting convex hull and convexity defect. Palm Center is an area determined as the maximum inscribed circle inside the contour.

• Depth point (Pd)and start point (Ps) must be above the center of maximum inscribed circle (Ca)
• Each start point (Pa) and end point (Pe) must be above Depth Point (Pd)
• Depth of each defect (ld) must be longer than palm center radius (ra)

  Fig. 3. Convex hull and convexity defect extracted from [11] hand tracking and recognition system for human-computer interaction using low-cost hardware

  Each is trained to distinguish one class from the others. The decision is taken with a winner-takes-all approach. However, there is no clear indication that this approach results in an excellent decision. In the 1-vs-1 strategy, the problem is divided into c (c – 1) / 2 sub- problems considering there are only two classes at a time.

  REPRINT

  This step requires the recognition ofdynamic hand gesture since the process considered is the dynamic process and needs to be handled with the learning model that can be accommodated with temporal time.

   Dynamic Gesture Recognition using HMM

  III.4.

  In this paper, we train and test four kernel types: RBF, Linear, Polynomial, and sigmoid. Then, we propose ten poses of static hand gesture as illustrated in Fig. 4. We use multiclass classifier of SVM to classify ten classes of static hand gesture. Fig. 4 shows static hand gesture representing sequence observation label. This static gesture is not used for human-computer interaction.

  This leaves the problem of evaluating an increased number of machines every time a new instance is classified

  At the decision phase, each machine casts a vote for one of the classes and the label with the highest number of votes wins.

  There is an immediate problem arising from the SVM’s original hyperplane formulation. It is not very obvious how to make the model applicable to more than two classes. Several approaches have been proposed to overcome this limitation. Two of them are known as the 1-vs-1 and 1-vs-all strategies for multiple class classification. For the decision problem over classes, 1- vs-all requires the creation of classifiers.

  The next step is calculating or extracting the feature from convexity defect that matches the criteria. In this paper, we use two feature vectors.

  The SVM algorithm separates the training data in feature space by a hyperplane, which maximizes the margin between two data. Hence, it is also known as maximum margin classifiers [2]. SVM is based on the principle of structural risk minimization (SRM). The SRM induction principle has two main objectives. First, it is to control the empirical risk on the training data, and the second is to control the capacity of the decision functions used to obtain this risk value [14].

• – which can become troublesome or prohibitive in time sensitive applications easily [15].

  

Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved International Review on Computers and Software, Vol. 11, N. 5

  The function of the hyperplane is to maximize the distance between the two pieces of data [2]. Basically, SVM is used to classify the two pieces of data or classes. However, the application of SVM applied is to classify more than two classes using method 1-vs-1 and 1-vs-all strategy 1-vs-1 by comparing one class with another class so the class that has the highest value will become the solution.

  Support vector machine is a supervised classification model that is often used to classify and regress. The basic algorithm SVM is to separate the feature of training data with a line or field separator, which is called hyperplane.

   Static Gesture Recognition Using SVM

  The first feature is taken from the angle between depth point (Pd) and end point (Pe) of each convexity defects (θa), whilethe second one is taken from the depth distance (distance between depth point and end point).

III.3.

  (a) (b) Fig. 4. 10 static hand gesture pose (later this pose used for sequence observation label for dynamic hand gesture)

  We propose HMM model [16] to accommodate this dynamic motion gesture model. Hidden Markov Model is one of the models based on the statistic. HMM is often used in applications that intersect with time. HMM possesses three main parameters,

  λ = (Π, A, B) [17], (c)

  where Π represents the vector beginning, A is the matrix of transition while B is the matrix emission. Furthermore,

  Figs. 6. (a) ergodic, (b) left-right, (c) parallel-left-right [18] extracted

  HMM has three major problems, which are Evaluation,

  froma tutorial on hidden markov model

  Decoding, and Training. However, these three problems can be solved using the method of Forward-Backward In this process, we train HMM model with HMM type Ergodic and left-right then for each type of HMM.We algorithm, Viterbi, and Baum-Welch [17]. In the application, HMM is often implemented using train it usingthe total hidden state from 2 until 9. three topologies. There are main Fully Connected (Ergodic model) where all the states are connected as a

  IV. Result and Discussion

  whole between the states; (Left-Right model) where each state can only be returned to the state itself and in the Our proposed system is implemented in .net using next state; (Parallel-Left-Right model) where each state

  EmguCV wrapper library, libsvm, and accord hmm. can only be repopulated themselves or towards the next Fig. 7 shows the system implementation screenshot. state. Shown in Figs. 6, the input of the model is the

  We use the standard webcam with 240 × 320 pixels result of sequence classification static hand gesture from resolution. Moreover, we use conditional indoor lighting

  SVM. Figs. 5 show dynamic hand gestures in this environment. In this process, we use three scenarios to system. test the system.

  The first scenario that we find is the matching of SVM parameter _{Left Click (a)} C and γ with the value from -3 to 3 for C parameter and - 15 to 3 for γ parameter. _{Right Click}

  We use four kernels (RBF, Linear, Polynomial, and _{Swipe Right (b) (c)} Sigmoid) in here.

  (e) gesture for “Grap”

Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved International Review on Computers and Software, Vol. 11, N. 5

  Swipe _{Left (d) Grap} REPRINT _(e) Figs. 5. 5 Dynamic hand motion gesture that replacing mouse functionality (random sample images from frame 1 until the end of videos), (a) gesture for “Left Click”,(b) gesture for “Right Click”,(c) gesture for “Swipe Left”,(d) gesture for “Swipe Right” Fig. 7. Hand motion gesture recognition system

  TABLE

  III

  The second scenario we see is the matching of HMM

  R ESULT T EST S CENARIO

  1 U SING L

  INEAR K ERNEL SVM

  parameter topology (Ergodic, left-right, and parallel-left-

  HMM Number of Training Testing

  right) and we test the number of the hidden state with the

  Topology Hidden State Accuracy (%) Accuracy (%)

  value from 2-9. In this scenario, we will test it using

  Ergodic

  2

  51.34

  38.67 Left-Right

  2

  94.67

  83.34 isolated hand motion gesture.

  Parallel-Left-Right

  2

  98

  82 The third scenario is that we will test the best

  parameter of SVM and HMM with continuous hand

  TABLE

  IV

  motion gesture. Then, we calculate the accuracy with the

  R ESULT T EST S CENARIO

  1 U SING P OLYNOMIAL K ERNEL SVM

  following equation:

  Number of Training Testing HMM Topology Hidden State Accuracy (%) Accuracy (%)

  Ergodic

  2

  45.34

  41.34

  (2) Left-Right

  2

  98

  94 parallel-Left-Right

  4

  98

  94.67 We use two datasets that were created by our camera. TABLE

  V R ESULT T EST S CENARIO

  1 U SING S

  IGMOID K ERNEL SVM

  The first dataset contains 50 images used to train and 50

  Number of Training Testing Accuracy images to test each static gesture. HMM Topology Hidden State Accuracy (%) (%)

  In this dataset, every image has various orientation

  Ergodic

  2 94 77.33333

  and scale. The second dataset contains 30 videos used to

  Left-Right 2 96.66667 82.66667 Parallel-Left-Right 4 95.33333

  82

  train and 30 videos to test each isolated dynamic gesture recognition.

  Fig. 8 below shows the result of graph in finding the On the other hand, we have two videos to test the best topology and number of hidden state in HMM. continuous dynamic hand gesture. One video has five

  The last scenario is to test the best model of SVM and gestures that are performed sequentially, and another HMM for continuous dynamic hand gesture recognition. video contains five gestures done randomly. In other

  The system shows a good result, that continuous hand words, our dataset contains 1000 images and 302 videos. gesture with convergent dataset testing (videos that

  Table I shows that kernel RBF with parameter C = 3 and contain only one gesture variance) can classify the

  γ = -1 achieves maximum accuracy training at 100% and

  gesture with achievement up to 83%. Otherwise, testing 93.4% for the testing dataset. with videos that have random gesture achieves 63%. This

  TABLE

  I

  happens because convergent gesture videos have the

  R ESULT T EST S CENARIO

1 F OR SVM P ARAMETER

  same sequence of image observation over all. Therefore,

  C Value Training Testing _x

Kernel Type ) HMM gets high probability to choose the same sequence

_{x γ Value (2} (2 ) Accuracy (%) Accuracy (%)

  of image observation. If we test it with videos that have

  RBF 3 -1 100

  93.4

  the random sequence of gesture, the system gets false

  Linear 3 -15

  99.4

89.4 Polynomial

  0.5

  1

  76.6

  79.4 recognition.

  Sigmoid 3 -3

  97

  83.8 In Table II topology parallel-left-right gets the

  maximum accuracy from training HMM with the number of hidden state = 4 and kernel SVM Polynomial.

  On the other hand, Figs. 8, 9, 10, 11 show that the best accuracy for hand motion gesture is HMM with topology Parallel-Left-Right and with 2-4 number of hidden state.

  The highest accuracy is achieved by using SVM kernel polynomial with parameter C = 0.5 and γ = -1. Then, Ergodic topology has the worst accuracy in all

  Fig. 8. HMM topology and number of hidden state testing result

  SVM kernel.This is because the sequence observation

  for kernel RBF SVM

  only needs various three sequences for each gesture, whileErgodic topology, which has more transitions than lef-right topology,was used to solve more complex problem [17].

  REPRINT TABLE

  II R ESULT T EST S CENARIO

  1U SING RBF K ERNEL SVM Testing HMM Number of Training Accuracy (%)

  Topology Hidden State Accuracy (%) Ergodic

  2

  94

80 Left-Right

  2

  97

  89.34 Parallel-Left-

  4

  99

  89.34 Fig. 9. HMM topology and number of hidden state testing result Right for kernel LINEAR SVM

  

Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved International Review on Computers and Software, Vol. 11, N. 5

  [6] Demidova, L., Sokolova, Y., Nikulchev, E., Use of Fuzzy Clustering Algorithms Ensemble for SVM Classifier Development, (2015) International Review on Modelling and Simulations (IREMOS) , 8 (4), pp. 446-457. [7] Ab Aziz, N., Abdul Rahman, T., Zakaria, Z., Reactive Power Planning for Maximum Load Margin Improvement Using Fast

  Artificial Immune Support Vector Machine (FAISVM), (2014) International Review of Automatic Control (IREACO) , 7 (5), pp. 436-447. [8] M. Elmezain, A. Al- Hamadi, J. Appenrodt, and B. Michaelis, “A

  Hidden Markov Model-based continuous gesture recognition system for hand motion trajectory,” 2008 19th Int. Conf. Pattern , pp. 3 Recognit.

• –6, 2008. [9] M. R. Malgireddy, J. J. Corso, S. Setlur, V. Govindaraju, and D.

  Mandalapu, “A framework for hand gesture recognition and Fig. 10. HMM topology and number of hidden state testing result spotting using sub- gesture modeling,” Proc. - Int. Conf. Pattern for kernel POLYNOMIAL SVM

  Recognit. , pp. 3780 –3783, 2010.

  [10]

  H. Y. Lai and H. J. Lai, “Real-Time Dynamic Hand Gesture Recognition,” 2014 Int. Symp. Comput. Consum. Control, no. 1, pp. 658 –661, 2014. [11]

  H. Yeo, B. Lee, and H. Lim, “Hand tracking and gesture recognition system for human-computer interaction using low- cost hardware,” Multimed Tools Applications 74(1) pp. 2687– 2715, 2015.

  [12] Regan, D., Srivatsa, S., Mixed Pixel Wise Characterization Based on HMM and Hyper spectral Image Gradient Enhancement for Classification Using SVM-FSK, (2014) International Review on Computers and Software (IRECOS) , 9 (6), pp. 1017-1026. [13] M. Schuster, “Largest Empty Circle Problem,” https://www.cs.swarthmore.edu/~adanner/cs97/s08/papers/schuste r.pdf pp. 28

• –37, 2008. [14] L. Gericke, M. Wenzel, R. Gumienny, C. Willems, and C. Meinel, Fig. 11. HMM topology and number of hidden state testing result “Handwriting recognition for a digital whiteboard collaboration for kernel SIGMOID SVM platform,” Proc. 2012 Int. Conf. Collab. Technol. Syst. CTS 2012, pp. 226 –233, 2012.

  [15] C. R. Souza and E. B. Pizzolato , “Sign Language Recognition with Support Vector Machines and Hidden Conditional Random

V. Conclusion and Future Works

  Fields: Going from Fingerspelling to Natural Articulated Words" Machine Learning and Data Mining in Pattern Recognit ion,” vol.

  This paper suggests a system for human-computer

  7988, of the series Lecture Notes in Computer Science pp 84-98,

  interaction using hand motion gesture. This system 2013. achieves an average accuracy of about 97%. The highest

  [16] A. Ramamoorthy, N. Vaswani, S. Chaudhury, and S. Banerjee,

  accuracy of static hand gesture is obtained from SVM

  “Recognition of dynamic hand gestures,” Pattern Recognition, vol. 36, no. 9, pp. 2069

• –2081, 2003.

  wit h kernel RBF, C = 3 and γ = -1.

  [17] M. Elmezain and A. Al- hamadi, “A Hidden Markov Model-Based

  On the other hand, the highest accuracy for dynamic

  Isolated and Meaningful Hand Gesture Recognition,”Proceedings

  hand gesture is from HMM with topology parallel-left-

  Of World Academy Of Science, Engineering And Technology (WASET) Vol. 31 July 2008 pp. 393

  right using kernel polynomial with parameter C = 0.5 and –400, 2008.

  γ = -1. It reaches 98% of accuracy.

  [18] L. R. Rabiner, “A Tutorial on Hidden Markov Models and

  Selected Application in Speech Recognition.” Proceeding of the

  By using HMM topology parallel-left-right with four IEEE Vol. 77 No. 2 February 1989 pp. 257 - 286. 1989. hidden states, the system presents 97% of accuracy.

  In the future, our system will focus on recognition gesture spotting using a depth camera to create effective

  Authors’ information skin color segmentation.

  Magister in Information Technology, Binus Graduate Program, Bina Nusantara University, Jakarta, Indonesia. E-mails

  References

  [1] M. Panwar and P. Singh Mehra, “Hand gesture recognition for Fitra B. Adinugraha received his Bachelor human computer interaction,” 2011 Int. Conf. Image Inf. Process., degree in Informatics majoring in Computer no. Iciip, pp. 1 –7, 2011.

  Science from Institute Technology Telkom [2] H. M. Gamal, H. M. Abdul- Kader, and E. a. Sallam, “Hand

   (ITT), Indonesia in 2011. Presently, he is gesture recognition using fourier descriptors,” 2013 8th Int. Conf. pursuing the Graduate Program of Information

  Comput. Eng. Syst. , pp. 274 –279, 2013.

  Technology at Bina Nusantara University, [3] R. Xu, S. Zhou, and W. Li, “MEMS accelerometer based Indonesia. His research interest includes nonspecific-user hand gesture recogn ition,” Sensors Journal,

  Human Computer Interaction (HCI), Natural

  IEEE , vol. 12, no. 5, pp. 1166 –1173, 2012.

  User Interface (NUI), Mobile Applications, [4] P. Kumar, J. Verma, and S. Prasad, “Hand Data Glove: A Cloud Storage, and Enterprise Architecture Design.

  Wearable Real-Time Device for Human- Computer Interaction,” Int. J. Adv. Sci. Technol. , vol. 43, pp. 15 –26, 2012.

  [5] M. K. Bhuyan, D. Ajay Kumar, K. F. MacDorman, and Y.

  Iwahori, “A novel set of features for continuous hand gesture recognition,” J. Multimodal User Interfaces, 2014.

  

Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved International Review on Computers and Software, Vol. 11, N. 5

  Suharjito is the Head of Information Technology Department in Binus Graduate Program of Binus University. He received under graduated degree in mathematics from The Faculty of Mathematics and Natural Science in GadjahMada University, Yogyakarta, Indonesia in 1994. He received master degree in information technology engineering from Sepuluh November Institute of Technology, Surabaya, Indonesia in

  2000. He received the PhD degree in system engineering from the Bogor Agricultural University (IPB), Bogor, Indonesia in 2011.His research interests are intelligent system, Fuzzy system, image processing and software engineering.

  REPRINT

Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved International Review on Computers and Software, Vol. 11, N. 5

  

International Review on Computers and Software

(IRECOS)

Aims and scope

  

The International Review on Computers and Software (IRECOS) is a peer-reviewed journal that publishes

original papers on all branches of the academic Computer Science and Engineering communities. Thematic areas include, but are not limited to: Computer Science Theory, Methods and Tools

Software engineering, algorithms and complexity, computational logic, formal methods, heuristics, mathematics and

models of computation, programming languages and semantics.

  Computer and Communications Networks and Systems

Network and distributed architectures and protocols, traffic engineering, resource management and Quality of Service,

network monitoring and traffic measurements, wireless networks, personal and body area networks, vehicular networks,

content and service-centric networking, multimedia communications and standards, energy efficient/green networks,

opportunistic and cognitive networks.

  Computational Intelligence, Machine Learning and Data Analytics

Human computer interaction, computational science, pattern recognition, computer vision, speech processing, machine

intelligence and reasoning, web science, databases, information retrieval, visualisation, current applications domains,

e.g. Healthcare and BioInformatics, and emerging application domains, e.g. big data.

  Security in Computer Systems and Networks

Computer systems security, hardware and embedded systems security, security protocol design and analysis,

cryptography and cryptanalysis, intrusion detection systems and techniques, user authentication techniques and

systems.

  Hardware Design Computer architectures, parallel architectures, operating systems and signal processing.

  

Instructions for submitting a paper

The journal publishes invited tutorials or critical reviews; original scientific research papers (regular

papers), letters to the Editor and research notes which should also be original presenting proposals for a new

research, reporting on research in progress or discussing the latest scientific results in advanced fields; short

communications and discussions, book reviews, reports from meetings and special issues describing

research in any of the above thematic areas. All papers will be subjected to a fast editorial process. Any paper will be published within two months from the submitted date, if it has been accepted. Papers must be correctly formatted, in order to be published. An Author guidelines template file can be found at the following web address:

www.praiseworthyprize.org/jsm/?journal=irecos

  Manuscripts should be sent via e-mail as attachment in .doc and .pdf formats to:

editorialstaff@praiseworthyprize.com

  

The regular paper page length limit is defined at 15 formatted Review pages, including illustrations,

references and author(s) biographies.

  Pages 16 and above are charged 10 euros per page and payment is a prerequisite for publication.

  REPRINT Abstracting and Indexing Information:

  Cambridge Scientific Abstracts (CSA/CIG) Academic Search Complete (EBSCO Information Services) Elsevier Bibliographic Database - SCOPUS Index Copernicus - IC Journal Master List 2012: ICV 6.45

  Autorizzazione del Tribunale di Napoli n. 59 del 30/06/2006

  1828-6003(201605)11:5;1-7 REPRINT

  Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved