Naïve Bayes Classifier for Hand Gestures Recognition

Imanuel Simatupang, Daniel Sutopo Pamungkas*, and Sumantri K. Risandriya

Mechatronics Dept, Politeknik Negeri Batam, Indonesia

Keywords: EMG, Myo Armband, Mobile Robot, Naive Bayes

Abstract: This paper provides recognizing the five gestures of the fingers using Naïve Bayes method. The

electromyography signal (EMG) is utilized to recognize the fingers movement. A myo armband is used to

obtain the signal. The average success rate of the system is about 90.61%. To verify the results, the outputs

of the system are used to control a mobile robot. The results show that the system is able to control the

movement of the robot.

1 INTRODUCTION

Every movements of the human generate a signal

from the muscles known as Electromyography

(EMG) (Eason, Noble, & Sneddon, 1955). Signal of

the muscles activities captures using the electrodes

placed in the skin of the human. The EMG signals are

utilized by the researchers for diverse objectives. In

the health applications, one of the purposes of this

signal is to known the human muscles condition

(Montoya, Henao, Muñoz, 2017). In the engineering

applications, EMG signals are used to identify the

movement of the human body e.g. the gestures of the

hands. One application in robotics is to control robot

movement using the recognizing system (Morais, G

et al. 2016).for example to control the movement of

the robot hand (Andrean, Pamungkas, & Risandriya,

2019). The robot fingers are controlled by the

movement of the fingers of the operator. This system

enables to help the disabilities people to substitute

their hand (Risandriya and Pamungkas, 2018).

To identify the signals of the muscles actions,

there are several recognizing algorithm have been

used by the researchers. For instance: Neural

Network algorithm (Risandriya & Pamungkas, 2018),

Fuzzy (Gogić, Miljkovic, & Đurđević, 2016),

Adaptive Neuro-Fuzzy Inference System

(Caesarendra, Tjahjowidodo, & Pamungkas, 2017),

Linear Discriminant Analysis (Zhang, 2012), K-

Nearest Neighbor (Kaya & Kumbasar, 2018), etc.

For this study, the Naïve Bayes algorithm is used

to recognize the gesture of the fingers of the subjects.

The root mean square (RMS) of the EMG signal is

used to be processed in this algorithm. Five fingers

postures are examined to be identified. These fingers

poses are: relax, all fingers are open, all fingers are

close, wave out and wave in. These gestures are used

to control the mobile robot in the certain track.

To provide a complete explanation, this article is

organized as follows: next section objective is to

provide an explanation of the method, also Naïve

Bayes. Then proceed with the next in section III,

which presents experiments on the method proposed

to identify hand movements. This is followed by a

comparison between the two methods, while the last

section is given conclusions obtained from

experiments conducted.

2 BACKGROUND

Naïve Bayes classifier is a classifier algorithm based

on probability theorem. The Bayesian rule, or known

as the conditional probability, is used for this

classifier. Equation (1) and equation (2) shows the

Bayes rules. To classify the classes, this algorithm

calculates the possibility of each of the categories.

The group which has the most significant number of

probabilities is the event is in that group

𝑃



𝑟

𝑞



𝑃



𝑞

𝑟



.𝑃



𝑟



𝑃



𝑞

⁄

(1)

And

𝑃



𝑟

𝑄



𝑃



𝑞



𝑟



.𝑃



𝑞



𝑟



…𝑃



𝑞



𝑟



(2)

Where:

110

Simatupang, I., Pamungkas, D. and Risandriya, S.

Naïve Bayes Classiﬁer for Hand Gestures Recognition.

DOI: 10.5220/0010352601100114

In Proceedings of the 3rd International Conference on Applied Engineering (ICAE 2020), pages 110-114

ISBN: 978-989-758-520-3

P(a|x) is the probability of posterior class a with

attribute x

P(x|a) is the likelihood which the probability of

the predictor given class

P(x) is the probability of class x

P(a) is the probability of class a

X = (x

, x

…., x

)

In this classification, all features/values are

assumed to be values independent of each other in

each class. Because of this condition, even if a value

depends on one another, this classification will

consider all propositions to be independent so that it

gives the probability effect. One of the advantages of

Naive Bayes is that only the average value and

standard deviation of variables are required to

estimate the parameters of the classification. The

equation of mean and standard deviation are shown in

(3) and (4) respectively.

𝜇





∑

𝑞







(3)

𝜎









∑

𝑞



𝜇







 (4)

Where:

µ is the mean of the attribute in a class

σ is the variance of the attribute in a class

The decision making rule for classification is to

choose one more possibility based on the results of

the probability density values such as the Gauss

Density equation. [5]

𝑓



𝑥





√

2𝜋𝜎

𝑒















(5)

For the experiments the Root Mean Square

(RMS) of the EMG signals is used as signals which

will be processed in the Naïve Bayes method. EMG

is a method to known electrical signal from the

muscle activity e.g. contraction or relaxation. To

obtain this signal, the electrodes are used. The Root

Mean Square (RMS) value of the EMG has a linear

relation and the activity of the muscle (Fukuda, T, Y.

et al 2010). The RMS values enable to obtain the

effective value of EMG signal. RMS equation (6) as

follows:

𝑅𝑀𝑆





𝑥





𝑁





(6)

Where:

𝑥





= the square of EMG signal amplitude

N = the data from EMG value

3 METHODS AND RESULTS

To establish the ability of Naïve Bayes algorithm the

block diagram of the system shown in Figure 1 is

made. This system aims are to recognize the gestures

of the hand also to drive a mobile robot.

The myo sensors are placed in the upper hand of

the subject. For this equipment we used myo

armband. A computer with 64 GB RAM and Intel

processor i-3 are used for performing Naïve Bayes

algorithm. The outcome of the algorithm is used for

controlling a custom mobile robot. This robot must be

follow in the specified path.

To control the robot hand gestures of the user are

used. Fist pose is used to stop the robot (Figure 2 (a)),

while the relaxed pose is used to transition from one

command to other command, as shown in Figure 2(b).

To make robot move forward, open fingers gestures

is applied (figure 2 (c)). Figure 2 (c) and Figure 2 (d)

are the gesture which are used to control robot to

move to the right and turn left respectively.

Figure 1: Block diagram of the system.

The algorithm which has a high achievement to

recognize the pose of the user's hand is calculated.

Naïve Bayes Classiﬁer for Hand Gestures Recognition

111

Also, the time of the robot for accomplish the mission

are measured.

The sensor is used using the surface EMG or

known as sEMG. For this system a myo armband is

utilized, this device is positioned on the user's arm.

Figure 3 shows users who use muscle sensors. This

equipment has eight muscle sensors and a position

sensor. These eight sensors will be used to obtain

signals from the user's muscles when moving their

fingers. One subject used for the experiment was a

male aged 23 years. The user had never used the myo

Armband sensor before.

Before the sensors signal processed in Naïve

Bayes and NN algorithm, the information is filtered.

Moving average filter is used to reduce the noise. For

the retrieval of data for the training phase, subjects

performed thirty times for each pose. In the Naïve

Bayes method, the mean and the standard deviation

of the RMS signal of the training phase are saving in

the database of the algorithm.

Figure 2: Poses of the hand (a) fist (b) relax (c) open (d)

wave right and (e) wave left.

Figure 3: The usage of the armband.

For the application phase, the subject tries the

effectiveness of both methods to recognize the pose

of the hand. The subject is doing the three times

experiment (thirty times per experiment) for each

gesture and each method. The results of these

experiments are shown in Table 1 for Naïve Bayes.

The average for the Naïve Bayes method is 90.61%,

success rate. The flow charts of the algorithms are

shown in Figure 4.

To prove the success of the system in recognizing

finger movement patterns, both algorithms are tested

to move a mobile robot. The information from the

computer to the robot is transmitted using wireless via

Bluetooth connection. This robot is equipped with

two dc motors. The dimension of the robot is 10 cm

height, a length and width are 20 cm and 15cm

respectively. The mobile robot has a free-wheel at the

front. An Arduino Uno is used to controlling the

robot. Robot figure is shown in Figure 5

The task of the robot is to follow the path, as

shown in Figure 6. Before the experiments are

performed, the subject is trained how to drive the

mobile robot using Naïve Bayes algorithm five times

each. For the experiments, every task is measured

how much time is needed for mobile robot to

accomplish the path for this algorithm. The results

can be shown in Table 2.

Table 1: Result the success recognition for Naïve Bayes

method.

Pose

Percentage success value (%)

Average

Trial 1 Trial 2 Trial 3

Fist 89.26 90.12 90.53 89.97

Relax 91.22 90.37 91.03 90.87

Open 87.61 89.27 86.47 87.78

Wave

right

93.6 90.31 92.74

92.22

Wave

left

92.92 93.28 90.41

92.20

Table 2: Time to accomplish the path.

Algorithm

Trial (second)

Average

(second)

1 2 3

Naïve Bayes 7.2 8.5 7.6 7.67

ICAE 2020 - The International Conference on Applied Engineering

112

Figure 4: Flow chart of the algorithms.

Figure 5: A mobile robot.

Figure 6: Path of the robot.

4 CONCLUSION

This article shows the ability of a recognize system to

identify fingers gesture using Naïve Bayes algorithm.

This system consists of a muscle sensor, a computer,

a controller, and a mobile robot. Data taken are RMS

values in the time domain. This system enables to

recognize the gestures around 90.6%. Moreover, this

system able to control mobile robot to follow the

certain path without any

ACKNOWLEDGEMENTS

This research was supported by Politeknik Negeri

Batam

REFERENCES

Andrean, D., Pamungkas, D. S., Risandriya, S. K., 2019.

Controlling Robot Hand Using FFT as Input to the NN

Algorithm. Journal of Physics: Conference Series.

Caesarendra, W., Tjahjowidodo, T., Pamungkas, D, S.,

2017. EMG based classification of hand gestures using

PCA and ANFIS (pp. 18-23). In International

Conference on Robotics, Biomimetics, and Intelligent

Computational Systems (Robionetics). IEEE.

Eason, G., Noble, B., Sneddon, I.N., 1955. On certain

integrals of Lipschitz-Hankel type involving products

of Bessel functions. Philosophical Transactions of the

Royal Society of London, pp.529-551.

Fukuda, T., Y., et al., 2010. Root mean square value of the

electromyographic signal in the isometric torque of the

quadriceps, hamstrings and brachial biceps muscles in

female subjects. The Journal of Applied Research, vol.

10, no. 1, pp. 32-39.

Gogić, A., Miljkovic, N., & Đurđević, Đ., 2016.

Electromyography-based gesture recognition: Fuzzy

classification evaluation. In International Conference

on Electrical, Electronic and Computing Engineering.

Kaya, E., Kumbasar, T., 2018. Hand gesture recognition

systems with the wearable myo armband. In 2018 6th

International Conference on Control Engineering &

Information Technology (CEIT). IEEE.

Montoya, M., Henao, O., Muñoz, J., 2017. Muscle fatigue

detection through wearable sensors: a comparative

study using the myo armband, In Proceedings of the

XVIII International Conference on Human Computer

Interaction, ACM.

Morais, G., et al., 2016. Application of myo armband

system to control a robot interface. In 9

International

Conference on Bio-Inspired Systems and Signal

Processing. SCITEPRESS.

Risandriya, S. K., Pamungkas, D. S., 2018. MYO Armband

sensors and Neural Network Algorithm for Controlling

Naïve Bayes Classiﬁer for Hand Gestures Recognition

113

Hand Robot. In International Conference on Applied

Engineering (ICAE). IEEE.

Zhang, D., et al., 2012. PCA and LDA for EMG-based

control of bionic mechanical hand. In International

Conference on Information and Automation. IEEE.

ICAE 2020 - The International Conference on Applied Engineering

114