A survey on bio-signal analysis for human-robot interaction

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 12, No. 6, December 2022, pp. 5998~6009
ISSN: 2088-8708, DOI: 10.11591/ijece.v12i6.pp5998-6009  5998
Journal homepage: http://ijece.iaescore.com
A survey on bio-signal analysis for human-robot interaction
Huda Mustafa Radha1
, Alia Karim Abdul Hassan2
1
Department of Computer Science, College of Science, University of Baghdad, Baghdad, Iraq
2
Department of Computer Science, University of Technology, Baghdad, Iraq
Article Info ABSTRACT
Article history:
Received Jul 29, 2022
Revised Jun 12, 2022
Accepted Jul 5, 2022
The use of bio-signals analysis in human-robot interaction is rapidly
increasing. There is an urgent demand for it in various applications,
including health care, rehabilitation, research, technology, and
manufacturing. Despite several state-of-the-art bio-signals analyses in
human-robot interaction (HRI) research, it is unclear which one is the best.
In this paper, the following topics will be discussed: robotic systems should
be given priority in the rehabilitation and aid of amputees and disabled
people; second, domains of feature extraction approaches now in use, which
are divided into three main sections (time, frequency, and time-frequency).
The various domains will be discussed, then a discussion of each domain's
benefits and drawbacks, and finally, a recommendation for a new strategy
for robotic systems.
Keywords:
Bio-signals
Health care
Human-robot interaction
This is an open access article under the CC BY-SA license.
Corresponding Author:
Huda Mustafa Rada
Department of Computer Science, Baghdad University
Baghdad, Al-Jadria Street, Iraq
huda.rada@sc.uobaghdad.edu.iq
1. INTRODUCTION
Human-robot interaction (HRI) is a rapidly expanding field of study and application. There are
numerous complex problems in this field, and solutions that have a positive societal impact are possible. due
to their interdisciplinary nature, researchers in the discipline must understand their studies from a larger
perspective [1]. HRI is an interdisciplinary topic that encompasses essential research in domains such as
human-robot interaction (HCI), artificial intelligence (AI), control systems, pattern recognition, psychology,
electronics, mechanics, social communication, behavioral expression systems, and neuroscience, among
others. Most HRI experiments involve the development of a robot that interacts with the environment or a
specific item. They devised a movement strategy for a particular situation (based on physical attributes)
without considering the impulse or stream of thinking that characterizes human activity [2].
To develop the robotics hardware and software needed to create a successful human-robot interface.
A range of fields must collaborate, analyze human behavior when interacting with robots in various social
circumstances, and design the aesthetics of the robot's embodiment and behavior and the domain knowledge
required for specific applications. Due to the numerous disciplinary jargon and processes [3]. Monitoring and
evaluating the patient's physiological data is crucial in physical therapy to assess treatment effects and
regulate assistive devices during the rehabilitation process. These two forms of data are detected using
various sensors, including electromechanical sensors (such as accelerometers) [4]. And biosensors (such as
electromyography (EMG) [5], as well as force sensors [6], Electroencephalography (EEG), and
magnetoencephalography (MEG) have been used. Electromechanical sensors can effectively detect
biological data [5], [7].
The number of possible bio-signals is enormous, given that there are many physiological systems of
relevance. Bio-signals include everything from a visual evaluation of the patient to bodily indicators captured

Int J Elec & Comp Eng ISSN: 2088-8708 
A survey on bio-signal analysis for human-robot interaction (Huda Mustafa Radha)
5999
by sensors in a larger sense [8]. Machine learning (ML) offers powerful algorithms for perception and
knowledge that may be used to create solid and versatile high-level recognition systems. Image processing
techniques and entropy distance, wearable cameras, and inertial sensors were used to recognize human
actions and detect falls. A 99% adaptable system based on decision trees and four sensors coupled to the
body system was constructed to identify daily behaviors such as standing, sitting, and walking [9].
This research paper is divided into the following sections: the second section explains how existing
studies' research was gathered. In the next area, we will go over how to evaluate human-robot interactions.
Following that, the model for gesture recognition techniques is presented. The last section compares domains
for feature extraction in pattern recognition techniques. The study's conclusion was then discussed.
2. RESEARCH METHOD
Unlike previous surveys, this work provides a thorough review of current advancements in
prosthesis control based on upper-limb myoelectric control in terms of model robustness, adaptation, and
reliability. Efforts will be made in three areas in this context: i) primary consideration should be given to
robotic systems in the rehabilitation and assistance of amputees and disabled persons; ii) domains of features
extraction techniques now in use, categorized into three types (time, frequency, and time-frequency); and iii)
a review of the benefits and drawbacks of each domain that has been evaluated. Elsevier, PubMed, IEEE,
SpringerLink, Google Scholar, and Wiley Online Library were used to conduct literature searches for this
review. Search terms included surface EMG (sEMG), myoelectric control, bio-signals analysis,
rehabilitations, assistances, classification, regression, features extraction, features domains, robustness,
reliability, domain adaptability, multi-modal, sensor fusion, confidence estimate, uncertainty analysis, and so
on. Publications from 2010 to 2021 were favored, while the time period was extended in some situations.
After doing a literature search, we carefully examine each article to exclude those that do not meet the
following criteria: i) the literature must focus on upper-limb motion estimates utilizing machine learning
control and ii) technical contributions should be relevant to one of the three targets, namely robotic systems
in the rehabilitation and assistance of amputees and disabled people, feature extraction, and their benefits and
drawbacks. For these objectives, we first chose 74 linked papers.
3. HUMAN-ROBOT INTERACTION BASED ON BIOMEDICAL SIGNALS
Bio-signals are the critical source of information on biosystem functioning and provide
communication between them. It is, as all signals require the transmission of energy. Bio-signals can be
evaluated directly from their biological source; however, external power is frequently used to study
physiological and external power interactions. A transducer converts a bio-signal into an electric signal that
can be measured. In most cases, the analog signal is converted to a digital (discrete-time) signal [10].
3.1. Electroencephalography signal
The EEG is a technique for detecting electrical activity in the brain. Because of its exceptional
temporal sensitivity, EEG is primarily used to detect changes in brain activity. EEG testing may benefit
patients with seizures, epilepsy, or unusual spells. This comes with several benefits. EEG signals are linked
to mental functions in the brain, making them less susceptible to an amputation. Second, muscle strain has
little effect on EEG-based movement estimations. As a result, using EEG in conjunction with another
bio-signal, such as sEMG, has piqued researchers' curiosity [11]–[13], where the features of two signals can
be analyzed sequentially or concurrently.
In study [14], researchers created a new EEG dataset that was infected with real-time motion
artifacts and discovered that coherence was a better similarity metric between the motion artifact-infected
EEG and motion sensor data. This real dataset was created to aid researchers working on EEG motion artifact
reduction techniques by allowing them to compare and contrast existing methods as well as develop new
models. Hadiyoso et al. [15] presented a study comparing the coherence of EEG signals in mild cognitive
impairment (MCI) patients and healthy people. Its goal is to make each group's feature visible so that the
coherence value can be used to detect early Alzheimer's disease via EEG analysis. It looked at the inter-and
intra-hemispheric coherence of the electrode pairs. MCI sufferers' mean coherence was lower than healthy
people were. In this experiment [16], the Wavelet technique to evaluate the EEG signal successfully
classified a person's brain dominance. This is due to the Wavelet's ability to assess brain signals based on
scale and direction. The benefits of EEG signals include being less reliant on amputation conditions. On the
other hand, the disadvantages include a low signal-to-noise ratio (SNR), a slow data transfer rate, poor
estimation accuracy, and restricted user adaptability.

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 6, December 2022: 5998-6009
6000
3.2. Electromyogram signal
The EMG signal, which is frequently employed in psychophysiology to study the relationship
between cognitive expression and physiological responsiveness, measures and records the electrical potential
generated by muscle cells [17]. The two categories of EMG sensors are sEMG and intra muscular (iEMG).
Muscle cells produce EMG signals, which are bioelectric signals used in a variety of applications including
sports science, ergonomics, assistive technology, and rehabilitation [18].
The method for classifying various hand gestures that have been developed will be helpful in
human-computer interaction as well as in controlling devices such as prostheses, virtual objects, and
wheelchairs; for example, In [19], researchers used time-domain EMG features that were normalized to the
area under the averaged root mean square curve to improve gesture recognition accuracy (AUC-RMS). From
each channel's active EMG signals, the four basic time-domain features were extracted: mean absolute value
(MAV), zero crossing, waveform length (WL), and slope sign change (SSC). The researchers used five
machine-learning algorithms to classify the three different hand motions: k-nearest neighbor (k-NN),
discriminant analysis (DA), naive Bayes (NB), random forest (RF), and support vector machine (SVM). The
results showed that both normalizing approaches improved performance indicators such as accuracy, F1
score, Matthew correlation coefficient, and Kappa score compared to the original EMG features.
EMG signal is employed as a signal for intention prediction of human motion in numerous robotic
applications, particularly in human-robot interaction systems. Most significantly, it is used to calculate the
kinematics of upper limb movements, which are the most active sections of the human body and are
necessary for daily tasks [20]. Patricia et al. [21] looked at four adaptive learning algorithms, two of which
had previously been evaluated in the prosthetic field, to see if they could help amputees learn to use an
sEMG-based prosthetic hand in less time. By increasing the number of source subjects available, they
compared their performance when dealing with different types and numbers of hand movements. As a result,
they demonstrated that adaptive learning has much potential in this field.
Researchers investigate several mother wavelet functions in discrete wavelet transforms (DWT), and
continuous wavelet transforms (CWT) in [22]. The performance of several mother wavelets in DWT and
CWT is also investigated at various decomposition levels and scales. Each CWT and reconstructed DWT
wavelet coefficient has extracted its MAV and WL features. The SVM, a popular machine learning
technique, recognizes different hand movements. As a result, DWT outperforms CWT in rehabilitation and
clinical application.
3.3. Mechanomyography signal
Acoustic myography (AMG) is the term for when microphones pick up an MMG signal [23].
During contraction, they measure the mechanical responses of muscle fibers' lateral oscillation at low
frequencies (2–200 Hz). Unlike sEMG, MMG is unaffected by changes in skin impedance or sensor
positioning sensitivity [24]. MMG can be created using a variety of transducers, including piezoelectric touch
sensors and accelerometers [25]. As a result, combining these two signals has piqued people's interest. In
[26], researchers proposed a hybrid EMG and MMG acquiring method for hand motion detection and
discovered that combining EMG and MMG features reduced classification error. On the other hand,
Zhang et al. [27] developed a structure-level strategy for post-processing sEMG-based hand gesture detection
and feature-based fusion techniques using MMG signals as movement onset/offset detectors.
4. HUMAN-ROBOT INTERACTION PERFORMANCE EVALUATION METHODS
Many studies have been carried out to document user experiences and the psychological impacts of
human-computer interaction. Using features of psycho-physiological data collected in real-time, they
deduced the subject's biology. There are varieties of research findings for various study issues along these
lines. Still, very little of the accumulated data can formulate speculative hypotheses or establish more
detailed research questions [28].
Multiple events, such as subject moves, electrode disconnection, fluctuating temperatures, humidity
levels, subject-dependent physiological abnormalities, electrostatic artifacts, and other unrelated user actions,
may occur during the data collection operation, causing the sensor signal to degrade and external interference
[29]. As a result, signal-preprocessing approaches need to be applied to the unprocessed signal, typically the
initial stage in developing an ML system. It requires synchronizing the signals from various sensors,
removing data loss and null values, and perhaps other tasks: filtering, noise removal, and outlier elimination
(usually by linear interpolation). The type of sensor utilized and the goal of the study state the filter type and
characteristics, which are as [30]: HRI's principal purpose is to equip robots with various skills to interact
with people more effectively. However, as more people contact robots daily, it is becoming increasingly
necessary for untrained users to engage with and use these robots efficiently. A robot hand a wheelchair [31],

6001
or a humanoid robot [32] are examples of external mechanical equipment that have grown increasingly
common in brain robot interaction (BRI) [33]–[35].
4.1. Rehabilitation and assistance by using human-robot interaction
Rehabilitation robots are one area of study that focuses on using robotic systems to augment and
analyze rehabilitation processes. These devices have been created to assist with various therapeutic
preparation and sensory function testing methods [36]. For example, robotic rehabilitation is a valuable tool
for treating motor control in people who have had a stroke or another type of motor impairment [37].
Alternatively, assistive robotics strives to help people with impairments execute their daily life activities
(DLAs) with greater freedom. Moving, gripping, as well as handling objects, and eating, are examples.
Consequently, robotic devices are used daily for therapy and assistance. A growing number of research
efforts have recently focused on assistive and rehabilitative robots due to rapidly growing technical
breakthroughs in machine learning, sensors, data processing, computation, prototype testing, and production,
as well as massive social demands such as mobility aids for older people [38], [39] are a couple of examples.
4.1.1. Assistive robotics
Millions of people worldwide have upper or lower limb limitations, limiting their reach and ability
to control objects. Geriatric patients, older patients, patients with muscular dystrophy (MD), amyotrophic
lateral sclerosis (ALS), Multiple sclerosis (MS), cerebral palsy (CP), spinal muscular atrophy (SMA), as well
as others who have severe motor paralysis, are among the patients [40]. This emphasizes a need for a
low-cost, moderate assistive device that enables patients to grasp objects, interact safely with their
environment, and gain independence. Researchers have developed assistive robotic manipulators (ARMs) in
wheelchair-mounted and fixed desktop forms [41]. Many studies have focused on building novel user
interfaces to improve the performance of helpful robotic manipulators. Furthermore, as more embedded
system computers become available, artificial intelligence is increasingly used to aid patients in executing
activities more autonomously and successfully. A sensitive joystick, similar to intelligent wheelchairs, is the
most often used user interface for assistive robotic manipulators [42].
4.1.2. Robotic prostheses
The use of a robotic prosthesis effective Prosthesis, also known as a rehabilitation robot, is a type of
rehabilitation robot attached to a patient's body and performs the duties of the missing limb during daily
activities. Patients frequently order and regulate their function via muscle or brain impulses in real-time. The
human body interacts directly with the robotic prosthesis. As a result, designing robotic prostheses, physical
qualities, humanoids' looks, knowledge of user intent, reproducing movements, grasp patterns of the
complete human body, or force effort are essential factors to consider [43]. Many recent research efforts have
focused on producing prosthetics that resemble a biological limb's capabilities. The key to successful
improvement is acquiring a comprehensive technique for comprehending the patient's desire to perform a
task and detecting the environment [44]. The mobility and dynamic control system of the robot may gather
numerous bio-signals in order to create a recognition algorithm and achieve the user's goal [45]. Prosthetic
operating methods, which significantly impact the device's comfort, define how muscle activation data
collected by sensing modalities are used to handle the prosthesis [46].
Upper limb prosthetics are frequently controlled using pattern-recognition-based proportional
control approaches and combinations. On chosen degree of freedom (DOF) shafts for the Prosthesis,
balanced control technology also offers a position, velocity, or force control, with a motion property adjusted
proportionally to the intensity of muscle contractions [47]. Although direct scaling control of prosthetic
fingers is more natural than grip selecting, the user may find it more challenging it to modify the position of
the fingers for the work because there are usually only a few control sites available [48]. Deep learning (DL)
has only recently been used in sEMG research, even though pattern recognition approaches have been used
for decades. Because of the introduction of large sEMG data sets and the development of evolutionary
algorithms, DL had already shown significant potential in the sEMG pattern identification of an artificial
hand. sEMG and DL are employed in various foundational gesture recognition investigations [18].
5. PATTERN RECOGNITION SYSTEM
Although the concept of gesture recognition techniques is similar, each step has its own style; also,
not all study uses all phases of the standard framework. As a result, we recommend a six-step conventional
strategy: data collection, data preparation, feature extraction, classification algorithms, post-processing, and
performance evaluation. Figure 1 shows how to identify sEMG signal patterns using conventional ML-based
methodologies.

 ISSN: 2088-8708
6002
5.1. Data acquisition
Two or more electrodes were put in the front and rear of the lower arm to acquire the EMG signal,
believing that the cross-talk content for each signal would be essentially the same for each movement. The
purpose would have been to characterize the signal's varied patterns for different directions [49]. Because
there are only a few coupled sensors over the muscles in this setup, most gesture labels are assigned to
sEMG [50].
Figure 1. The general framework for describing the interactions between the user and the robotic prosthesis
5.2. Data preprocessing
A single-channel EMG signal is segmented for post-processing and production of muscle activation
patterns. Each muscle is independently recorded since not all EMG recording channels are sufficient to
account for movement cycles. The applicability of particular EMG channels is determined by the mode of
muscle activity and the number of onsets and cessations during each movement cycle. For example, muscles
in a phasic manner with only one activity interval within the movement cycle produce the most appropriate
EMG signals [51]. Because of the limited EMG voltage amplitude and poor EMG signal in stroke patients,
recognizing motion intent based on EMG presents an additional challenge due to the reduced SNR; hence,
standard signal processing may not effectively extract the user's purpose. Emerging machine-learning signal
processing algorithms may increase EMG signal processing quality while eliminating the need for onsite
manual calibration and threshold setup operations, resulting in a user-friendly solution [52]. To determine
EMG-based hand movement intent, neural network-based machine learning techniques such as a
convolutional neural network (CNN) were used, demonstrating the ability to overcome these challenges in
EMG signal feature extraction and system calibration [22].
5.3. Feature extraction
Extracting features from a signal to achieve consistent categorization is known as feature extraction.
The most important step in biomedical signal classification is feature extraction because if the features are
not extracted correctly, the classification performance will suffer. The feature extraction stage should reduce
the original data to a smaller dimension while retaining most of the useful information from the initial vector.
As a result, it is critical to identify the fundamental elements that define the collection based on its nature.
Because they are the most typical values for determining the distribution of biological signals, different
statistical features can be retrieved from each subsample data point. Biological signals' min, max, mean,
median, mode, std, variance (VAR), 1st, 3rd, and inter-quartile range (IQR) can be used as features [53]. ML
employs feature engineering to achieve this goal, time domain (TD) features, time-serial domain (TSD)
features, frequency domain features, and time-scale or time-frequency domain characteristics are all included
[54].
5.3.1. Time-domain features
The TD values are determined by the raw sEMG, which are time functions. They are commonly
used because they have a lower processing complexity than other sEMG features. MAV, root mean square
(RMS), integrated EMG (iEMG), histogram (HIST), zero crossings (ZC), standard deviation (SD), SSC,
waveform length (WL), Willison amplitude (WAMP), variance (VAR). Other time-domain components are

6003
frequently used, commonly used together [18]. Table 1 presents a brief use of Time-domain features. Hamedi
et al. [55] Use the integrated (INT), MAV, mean absolute value slop (MAVS), RMS, VAR, and WL
time-domain feature extraction methods on signals. To obtain recognition accuracy and assess the efficacy of
each feature extraction method, the features are categorized using Fuzzy C-means.
Table 1. Shows the domain-specific features
Category Feature Formula Time-domain
Time
domain
MAV MAV=
1
𝑁
∑ |𝑥𝑖|
𝑁
𝑖=1
Indication of muscle contraction levels
RMS
RMS=√
1
𝑁
∑ 𝑥𝑖2
𝑁
𝑖=1
A Gaussian random process with
amplitude modulation.
iEMG iEMG = ∑ |𝑥𝑖|
𝑛
𝑖=1 An index for detecting the onset of a
condition
ZC ZC=∑ [𝑠𝑔𝑛(𝑥𝑖 × 𝑥𝑖 + 1) ∩ |𝑥𝑖 − 𝑥𝑖 + 1| ≥
𝑛−1
𝑖=1
𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑]
𝑠𝑔𝑛(𝑥) = {
1 , 𝑖𝑓 𝑥 ≥ 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
}
A feature that makes an
approximation of frequency domain
properties.
SSC SSC=∑ [ 𝑓[(𝑥𝑖
𝑛−1
𝑖=2 − 𝑥𝑖−1) × (𝑥𝑖 − 𝑥𝑖+1 )]]
𝑓(𝑥) = {
1, 𝑖𝑓 𝑥 ≥ 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑑
A method for representing the signal's
frequency information.
WL WL=𝑙𝑜𝑔(
∑ |𝛥𝑥|
𝑛−1
𝑖=0
∑ | 𝛥2𝑥 |
𝑛−1
𝑖=0
) The signal's frequency, duration, and
amplitude are all detailed.
WAMP
𝑓(∆i) = {
1, 𝑖𝑓 ∆i ≥ 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑑
∆i = |𝑥𝑖 + 1 − 𝑥𝑖|
Indicator of muscle contraction
intensity.
VAR VAR=
1
𝑁−1
∑ 𝑥𝑖
2
𝑁
𝑖=1
The sEMG signal's power index
V-order
(V) V=(
1
𝑁
∑ 𝑥𝑖
𝑣
𝑁
𝑖=1 )
1
𝑣
A nonlinear detector for estimating
the force of muscle contractions
implicitly.
simple
square
integral
(SSI)
SSC=∑ [ 𝑓[(𝑥𝑖
𝑛−1
𝑖=2 − 𝑥𝑖−1) × (𝑥𝑖 − 𝑥𝑖+1 )]]
𝑓(𝑥) = {
1, 𝑖𝑓 𝑥 ≥ 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑑
The sEMG signal's energy index
Frequency
domain
total
power
(TP)
TP=∑ 𝑝𝑖
𝑀
𝑗 The sEMG power spectrum as a whole.
modified
phase
(MP)
MP=
1
𝑀
∑ 𝑝𝑖
𝑀
𝑗=1
The EMG power spectrum's average
power
mean
frequency
(MNF)
MNF=
∑ 𝑓𝑗𝑃𝑖
𝑀
𝑗=1
∑ 𝑃𝑖
𝑀
𝑗=1
Frequency average
median
frequency
(MDF)
∑ 𝑝𝑗 = ∑ 𝑝𝑗 =
1
2
∑ 𝑃𝑗
𝑀
𝑗=1
𝑁
𝑗=𝑀𝐷𝐹
𝑀𝐷𝐹
𝑗=1
A frequency splits the sEMG power
spectrum into two regions of equal
amplitude.
Notes: N denotes the number of sampling points; xi denotes the sEMG signal of the ith sampling point; M denotes the total number of
frequency bins; PJ denotes the sEMG power spectrum, and fj indicates the frequency of the range of the jth frequency bin.
According to Balbinot and Favieiro [56], the system was created with a small number of myoelectric
signal acquisition channels (up to eight) and a more robust artificial intelligence technique in mind. By
describing seven different motions, he was able to demonstrate the system's validity. They used an
eight-input neuro-fuzzy network with one output to calculate the RMS value for each of the eight channels.
The average accuracy for seven different moves was 86 %. On the other hand, Wu et al. [57] study the use of
a single-channel signal characteristic of a single-channel time-domain sEMG envelope in a gesture detection
system. The modified KNN algorithm and the soft margin SVM method are used to classify five different
types of motions. Using the improved KNN algorithm and the smooth margin SVM approach, they were able
to attain gesture recognition scores of 75.8% and 79.4%, respectively. Table 2 provides more information.
5.3.2. Frequency domain features
The FD features are calculated using the Fourier transform of the sEMG signal's autocorrelation
function, which is then estimated using a period gram or parameter method. Other frequency domain
properties of sEMG signals include frequency ratio (FR), TP, MP, MDF, MNF, power spectrum (PS), and

 ISSN: 2088-8708
6004
others [58]. The detailed information is shown in Table 1. Awang et al. [59] create an integrated method for
identifying brain changes during rest and writing situations using power spectral density (PSD) estimation for
selected representative EEG data. The k-NN classifier is used, which has an average accuracy of 95%
researcher in [60] used the short-time Fourier transform on raw EEG data to determine the power distribution
for each frequency band. The stationary wavelet transform was used to remove artifacts, and k-NN and SVM
were used to classify them.
The power distribution for each frequency band was determined using the short-time Fourier
transform on raw EEG data. Artifacts were removed using the stationary wavelet transform, and they were
classified using k-NN and SVM. They show that k-NN has the best results for frequency bands between 4
and 50 HZ, with an average accuracy of 94.208% and 90.816% using SVM.
Djemal et al. [61] proposed a system for extracting features. Using event-related de-synchronization
and synchronization strategies, the classification accuracy of three-class motor imagery (MI) brain to
computer interaction (BCI) is improved. The classification method uses fast Fourier transforms (FFT) and
autoregressive (AR) modeling to combine phase and amplitude aspects of brain signals. For two BCI
competition datasets, 86.06 % and 93% classification accuracy were achieved using sequential forward
floating selection (SFFS) and multi-class linear discriminant analysis (LDA) classifiers. Acharya et al. [62]
developed a computer-aided diagnostic (CAD) technique that can classify the three classes of EEG segments
with 99.7% classification accuracy by using a fuzzy Sugeno classifier and non-linear features based on
higher-order spectra (HOS), fractal dimension, and Hurst Exponent. Table 3 provides further information
time-dependent power spectrum descriptors (TD-PSD) [63] and temporal-spatial descriptors (TSD) [64] are
two new features that have recently been proposed to increase the robustness of feature extraction. Because a
single feature, such as Phinyomark's feature set, may only supply so much information, combining elements
from multiple groups is a smart concept [65].
Table 2. Summary of previous literature utilizing time domain features extraction
Ref Year Signal type Features Classification Accuracy
[55] 2012 sEMG INT, MAV, MAVS, RMS, VAR, and WL Fuzzy C-means (FCM) 90.8%
[56] 2013 sEMG RMS neuro-fuzzy 86%
[57] 2018 sEMG IEMG, MAV, VAR, Standard deviation,
Average energy, MAX, SSC, Skewness,
Kurtosis
KNN+SVM 75.8-79.4%
[66] 2021 EMG, ECG, hand and foot
galvanic skin response (hand
GSR) and (foot GSR)
statistical features SVM 88.89%
*Galvanic skin response (GSR)
Table 3. Summary of previous literature utilizing frequency domain features extraction
Ref Year Signal type Features Classification Accuracy
[59] 2012 EEG Spectral density of power (PSD) k-NN 95%
[62] 2012 EEG Higher-order spectra higher-ordertal dimension
and hurst exponent-based nonlinear features
Fuzzy Sugeno
Classifier
99.7%
[61] 2016 EEG FFT + auto regressive (AR) LDA 86.06%-93%
[60] 2020 EEG Short-time fourier transform k-NN+SVM 94.20%-90.81%
5.3.3. Time-frequency domain features
The energy of the sEMG signal in time and frequency is calculated using time–frequency domain
(TFD) features, which is a typical feature extraction approach. The frequency bands are formed by different
wavelet coefficients, and statistical indicators are extracted as TFD features. When the basic wavelet function
db4 is utilized as the wavelet transform equation, the raw sEMG signal is partitioned into three levels.
𝑊𝑇 =
1
√𝑎
∫ 𝑓(𝑡)𝜓∗
(
𝑡−𝜏
𝑎
) 𝑑𝑡
+∞
−∞
(1)
Where, a is the scale parameter, the mother wavelet is 𝜓 (t), and the translation parameter is τ. The ith level
decomposition coefficient ci (ci=one of [cD1, cD2, cD3, cA3]) has a length of ni.; μ is the average of ci [57].
5.3.4. Parameter model
The parameter model's central premise is that raw sEMG data sequence information should be
treated as a time series. So, the sEMG signal is stable on short notice. The coefficients and intercept of the
fourth autoregressive model are often used as characteristic values [67].

6005
5.3.5. Performance feature extraction domains
By thoroughly reading various publications, including primary methodologies for one-dimensional
signal linear analysis in the time, frequency, or time-frequency domain. Several common interest approaches
were examined and rated for their overall benefits and drawbacks. The advantages and disadvantages of
several feature extraction algorithms for gesture detection are shown in Table 4.
Table 4. Feature extraction approaches (pros and cons) [54], [68]–[70]
Domain name Pros Cons
Time − TD characteristics are derived directly from
raw signal time series without any
manipulations.
− Implementation is simple.
− The computational load must be kept to a
minimum.
− Because they are entirely made of time, Time-domain
windows are adaptable.
− Depending on the values of the R-R interval,
Frequency − This is a valuable tool for signal processing in
a stationary environment
− It is better for narrowband signals like sine
waves.
− It outperforms virtually all other approaches
available in real-time applications
− Non-stationary data analysis, such as EEG suffers from a
flaw.
− Because of its poor spectrum estimation, it cannot be
used to analyze short EEG signals.
− Its spectrum estimation is poor, so it cannot be used to
analyze short EEG signals.
Time-
Frequency
− It has a variable window size, with a large one
at low frequencies and a small one at higher
frequencies.
− It excels at analyzing transient and fast signal
changes.
− Better at analyzing erratic data patterns, such
as impulses that occur at irregular intervals.
− It is necessary to choose a suitable mother wavelet.
Parameter
model
− The use of augmented reality (AR) techniques
reduces spectral loss and improves frequency
resolution
− Excellent frequency resolution is provided.
− Spectral analysis based on the AR model is
helpful when analyzing small data segments.
− It is difficult to choose the model order in AR spectral
estimation.
− The AR technique produces poor spectral estimation
when the estimated model is not appropriate, and the
model's orders are incorrectly specified.
− Solid biases and even large variability can easily
influence it.
5.4. Pattern classification
The final stage of pattern recognition for EMG control of the prosthesis is pattern classification.
Following that, a specific tool can be chosen based on the system's requirements and characteristics. One of
the most widely used pattern detection methods in myoelectric signals is support vector machines (SVM),
which discover an n-dimensional hyperplane that can divide a set of extracted input features into different
classes [71].
This approach outperforms other classifiers in recognizing complicated patterns, such as artificial
neural networks (ANN) and linear discriminant analysis (LDA). On a number of occasions. The SVM is
based on the following basic concepts: i) hyperplane divider, ii) kernel function, iii) optimal separation
hyperplane, and iv) soft margins (tolerance of hyperplane). SVMs have a high classification accuracy
because they may be coupled with other classification methods to achieve diverse classification goals. They
also have a high rate of accuracy [72].
5.5. Performance evaluation
5.5.1. The metrics used for models’ evaluation
Prediction accuracy is usually one of the most important criteria for evaluating the functionality of a
pattern recognition model. Offline testing evaluates the model's performance using accuracy, recall,
precision, standard deviation, and other metrics. Error measures such as mean square error (MSE), root mean
square error (RMSE), normalized root mean square error (NRMSE), and correlation coefficient are
commonly used to classify these observations [73].
Furthermore, the data in the model's test set for each category and generation must be balanced;
otherwise, the model's accuracy will be comparable to that of subjects with more data. The system
performance evaluation can measure the accuracy of the complete system in real-time when developing the
real-time model. Fetts' law was used to calculate these results. This is used to evaluate the proposed system's
objective performance [74].

 ISSN: 2088-8708
6006
6. COMPARATIVE ANALYSIS OF MOTION DETECTION TECHNIQUE AND DISCUSSION
This paper summarizes recent human-robot interaction research and development efforts in assistive
and rehabilitation robotics. Several subdomains of assistive and rehabilitative robotic research are first
identified. The background and trends of such advancements and the types of feature extraction domains
employed by the technique are discussed. It has been discovered that feature extraction methods based on the
Time domain can yield good results if the environment (e.g., temperature, humidity, and lightning) does not
influence the recorded signals. While features-based Frequency domains perform well in real-time
applications, they struggle to analyze brief EEG signals. It is difficult to choose the model order for AR
spectral estimation. On the other hand, Feature-based parameter models can help reduce the constraints of
spectral problems and improve the technique's accuracy by increasing frequency resolution. The feature-
based Time-Frequency domain is superior for assessing rapid and transient signal changes and
comprehending irregular data patterns, but it necessitates the selection of the appropriate mother wavelet. In
addition, combining a large dataset with a robust feature selection strategy can improve the classifier’s
performance.
7. CONCLUSION
The current methodologies and measures for assessing HRI states in biological signal analysis
investigations are summarized in this paper. Different methods for estimating user states and the signals and
procedures utilized to gather these signals are identified. For example, they must be processed and analyzed
to prepare them for usage in machine learning algorithms. Furthermore, this research reviewed and analyzed
the most important literature in the HRI perspective to measure the cognitive state by examining
psychological and biological signals. After benefiting from the literature study, we want to develop a multi-
module that uses analysis of biological data such as EEG and EMG to produce a robotic limb to assist
patients with upper limbs. According to our research, the most prevalent feature extraction strategies have
been presented with their benefits and drawbacks. According to a comparative study, accuracy varies
depending on the situation. Finally, moving from feature engineering to feature learning is a good idea. On
the other hand, DL improves representation through feature learning, which extracts high-level features from
input data through multiple layers of processing blocks.
REFERENCES
[1] W. Lambrechts, J. S. Klaver, L. Koudijzer, and J. Semeijn, “Human factors influencing the implementation of cobots in high
volume distribution centres,” Logistics, vol. 5, no. 2, May 2021, doi: 10.3390/logistics5020032.
[2] C. Jost et al., Human-robot interaction, vol. 12. Cham: Springer International Publishing, 2020, doi: 10.1007/978-3-030-42307-0.
[3] F. Negrello, H. S. Stuart, and M. G. Catalano, “Hands in the real world,” Frontiers in Robotics and AI, vol. 6, Jan. 2020,
doi: 10.3389/frobt.2019.00147.
[4] L. M. S. do Nascimento, L. V. Bonfati, M. L. B. Freitas, J. J. A. Mendes Junior, H. V. Siqueira, and S. L. Stevan, “Sensors and
systems for physical rehabilitation and health monitoring—a review,” Sensors, vol. 20, no. 15, Jul. 2020, doi: 10.3390/s20154063.
[5] C. Fang, B. He, Y. Wang, J. Cao, and S. Gao, “EMG-centered multisensory based technologies for pattern recognition in
rehabilitation: state of the art andchallenges,” Biosensors, vol. 10, no. 8, Jul. 2020, doi: 10.3390/bios10080085.
[6] U. Martinez-Hernandez, I. Mahmood, and A. A. Dehghani-Sanij, “Simultaneous Bayesian recognition of locomotion and gait
phases with wearable sensors,” IEEE Sensors Journal, vol. 18, no. 3, pp. 1282–1290, Feb. 2018,
doi: 10.1109/JSEN.2017.2782181.
[7] S. S. Jasim and A. K. Abdul Hassan, “Modern drowsiness detection techniques: a review,” International Journal of Electrical and
Computer Engineering (IJECE), vol. 12, no. 3, pp. 2986–2995, Jun. 2022, doi: 10.11591/ijece.v12i3.pp2986-2995.
[8] S. N. Mohammed and A. K. A. Hassan, “A survey on emotion recognition for human robot interaction,” Journal of Computing
and Information Technology, vol. 28, no. 2, pp. 125–146, Jun. 2021, doi: 10.20532/cit.2020.1004841.
[9] H. Rezaie and M. Ghassemian, “An adaptive algorithm to improve energy efficiency in wearable activity recognition systems,”
IEEE Sensors Journal, vol. 17, no. 16, pp. 5315–5323, Aug. 2017, doi: 10.1109/JSEN.2017.2720725.
[10] A. Mohammed and L. Wang, “Advanced human-robot collaborative assembly using electroencephalogram signals of human
brains,” Procedia CIRP, vol. 93, pp. 1200–1205, 2020, doi: 10.1016/j.procir.2020.03.074.
[11] E. Rocon et al., “Multimodal BCI-mediated FES suppression of pathological tremor,” in Annual International Conference of the
IEEE Engineering in Medicine and Biology, Aug. 2010, pp. 3337–3340, doi: 10.1109/IEMBS.2010.5627914.
[12] K. KIGUCHI, T. D. LALITHARATNE, and Y. HAYASHI, “Estimation of forearm supination/pronation motion based on EEG
signals to control an artificial arm,” Journal of Advanced Mechanical Design, Systems, and Manufacturing, vol. 7, no. 1,
pp. 74–81, 2013, doi: 10.1299/jamdsm.7.74.
[13] X. Li, O. W. Samuel, X. Zhang, H. Wang, P. Fang, and G. Li, “A motion-classification strategy based on sEMG-EEG signal
combination for upper-limb amputees,” Journal of NeuroEngineering and Rehabilitation, vol. 14, no. 1, Dec. 2017,
doi: 10.1186/s12984-016-0212-z.
[14] A. Islam, E. J. Esha, S. F. Binte Ahmed, and M. Kafiul Islam, “Study and analysis of motion artifacts for ambulatory
electroencephalography,” International Journal of Electrical and Computer Engineering (IJECE), vol. 12, no. 2, pp. 1520–1529,
Apr. 2022, doi: 10.11591/ijece.v12i2.pp1520-1529.
[15] S. Hadiyoso, I. Wijayanto, and S. Aulia, “Comparison of resting electroencephalogram coherence in patients with mild cognitive
impairment and normal elderly subjects,” International Journal of Electrical and Computer Engineering (IJECE), vol. 12, no. 2,
pp. 1558–1564, Apr. 2022, doi: 10.11591/ijece.v12i2.pp1558-1564.

6007
[16] K. A. Abu Nawas, M. Mustafa, R. Samad, D. Pebrianti, and N. R. Hasma Abdullah, “K-NN classification of brain dominance,”
International Journal of Electrical and Computer Engineering (IJECE), vol. 8, no. 4, pp. 2494–2502, Aug. 2018,
doi: 10.11591/ijece.v8i4.pp2494-2502.
[17] J. S. Hussain, A. Al-Khazzar, and M. N. Raheema, “Recognition of additional myo armband gestures for myoelectric prosthetic
applications,” International Journal of Electrical and Computer Engineering (IJECE), vol. 10, no. 6, pp. 5694–5702, Dec. 2020,
doi: 10.11591/ijece.v10i6.pp5694-5702.
[18] W. Li, P. Shi, and H. Yu, “Gesture recognition using surface electromyography and deep learning for prostheses hand: State-of-
the-art, challenges, and future,” Frontiers in Neuroscience, vol. 15, Apr. 2021, doi: 10.3389/fnins.2021.621885.
[19] M. F. Wahid, R. Tafreshi, M. Al-Sowaidi, and R. Langari, “Subject-independent hand gesture recognition using normalization
and machine learning algorithms,” Journal of Computational Science, vol. 27, pp. 69–76, Jul. 2018,
doi: 10.1016/j.jocs.2018.04.019.
[20] Y. Liu, Z. Li, H. Liu, and Z. Kan, “Skill transfer learning for autonomous robots and human–robot cooperation: a survey,”
Robotics and Autonomous Systems, vol. 128, Jun. 2020, doi: 10.1016/j.robot.2020.103515.
[21] N. Patricia, T. Tommasit, and B. Caputo, “Multi-source adaptive learning for fast control of prosthetics hand,” in 22nd
International Conference on Pattern Recognition, Aug. 2014, pp. 2769–2774, doi: 10.1109/ICPR.2014.477.
[22] J. Too, A. R. Abdullah, N. Mohd Saad, N. Mohd Ali, and H. Musa, “A detail study of wavelet families for EMG pattern
recognition,” International Journal of Electrical and Computer Engineering (IJECE), vol. 8, no. 6, pp. 4221–4229, 2018,
doi: 10.11591/ijece.v8i6.pp.4221-4229.
[23] N. Siddiqui and R. H. M. Chan, “Hand gesture recognition using multiple acoustic measurements at wrist,” IEEE Transactions on
Human-Machine Systems, vol. 51, no. 1, pp. 56–62, Feb. 2021, doi: 10.1109/THMS.2020.3041201.
[24] R. B. Woodward, M. J. Stokes, S. J. Shefelbine, and R. Vaidyanathan, “Segmenting mechanomyography measures of muscle
activity phases using inertial data,” Scientific Reports, vol. 9, no. 1, Apr. 2019, doi: 10.1038/s41598-019-41860-4.
[25] M. A. Islam, K. Sundaraj, R. B. Ahmad, N. U. Ahamed, and M. A. Ali, “Mechanomyography sensor development, related signal
processing, and applications: a systematic review,” IEEE Sensors Journal, vol. 13, no. 7, pp. 2499–2516, Jul. 2013,
doi: 10.1109/JSEN.2013.2255982.
[26] P. Prociow, A. Wolczowski, T. G. Amaral, O. P. Dias, and J. Filipe, “Identification of hand movements based on MMG and EMG
Signals,” in Proceedings of the First International Conference on Bio-inspired Systems and Signal Processing, 2008,
pp. 534–539, doi: 10.5220/0001057305340539.
[27] X. Zhang, X. Li, O. W. Samuel, Z. Huang, P. Fang, and G. Li, “Improving the robustness of electromyogram-pattern recognition
for prosthetic control by a postprocessing strategy,” Frontiers in Neurorobotics, vol. 11, Sep. 2017,
doi: 10.3389/fnbot.2017.00051.
[28] M. Z. Baig and M. Kavakli, “A survey on psycho-physiological analysis & measurement methods in multimodal systems,”
Multimodal Technologies and Interaction, vol. 3, no. 2, May 2019, doi: 10.3390/mti3020037.
[29] A. Samad, D. R. Obando Nuñez, G. C. Solis Castillo, B. Laquai, and U. Vogt, “Effect of relative humidity and air temperature on
the results obtained from low-cost gas sensors for ambient air quality measurements,” Sensors, vol. 20, no. 18, Sep. 2020,
doi: 10.3390/s20185175.
[30] P. J. Bota, C. Wang, A. L. N. Fred, and H. Placido Da Silva, A review, current challenges, and future possibilities on emotion
recognition using machine learning and physiological signals, vol. 7, Institute of Electrical and Electronics Engineers (IEEE),
2019, pp. 140990–141020, doi: 10.1109/ACCESS.2019.2944001.
[31] D. A. Craig and H. T. Nguyen, “Adaptive EEG thought pattern classifier for advanced wheelchair control,” in 29th Annual
International Conference of the IEEE Engineering in Medicine and Biology Society, Aug. 2007, pp. 2544–2547,
doi: 10.1109/IEMBS.2007.4352847.
[32] A. Guneysu and H. L. Akin, “An SSVEP based BCI to control a humanoid robot by using portable EEG device,” in 35th Annual
International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Jul. 2013, pp. 6905–6908,
doi: 10.1109/EMBC.2013.6611145.
[33] H. Abdulkarim and M. Z. Al-Faiz, “Online multiclass EEG feature extraction and recognition using modified convolutional
neural network method,” International Journal of Electrical and Computer Engineering (IJECE), vol. 11, no. 5, pp. 4016–4026,
Oct. 2021, doi: 10.11591/ijece.v11i5.pp4016-4026.
[34] J. Zhao, W. Li, X. Mao, and M. Li, “SSVEP-based experimental procedure for brain-robot interaction with humanoid robots,”
Journal of Visualized Experiments, vol. 2015, no. 105, Nov. 2015, doi: 10.3791/53558.
[35] S. Bozinovski and A. Bozinovski, “Mental states, EEG manifestations, and mentally emulated digital circuits for brain-robot
interaction,” IEEE Transactions on Autonomous Mental Development, vol. 7, no. 1, pp. 39–51, Mar. 2015,
doi: 10.1109/TAMD.2014.2387271.
[36] R. Gassert and V. Dietz, “Rehabilitation robots for the treatment of sensorimotor deficits: a neurophysiological perspective,”
Journal of NeuroEngineering and Rehabilitation, vol. 15, no. 1, Dec. 2018, doi: 10.1186/s12984-018-0383-x.
[37] W. Liao, C. Wu, Y. Hsieh, K. Lin, and W. Chang, “Effects of robot-assisted upper limb rehabilitation on daily function and real-
world arm activity in patients with chronic stroke: a randomized controlled trial,” Clinical Rehabilitation, vol. 26, no. 2,
pp. 111–120, Feb. 2012, doi: 10.1177/0269215511416383.
[38] H. Tanaka, M. Yoshikawa, E. Oyama, Y. Wakita, and Y. Matsumoto, “Development of assistive robots using international
classification of functioning, disability, and health: concept, applications, and issues,” Journal of Robotics, vol. 2013, pp. 1–12,
2013, doi: 10.1155/2013/608191.
[39] S. W. Brose et al., “The role of assistive robotics in the lives of persons with disability,” American Journal of Physical Medicine
and Rehabilitation, vol. 89, no. 6, pp. 509–521, Jun. 2010, doi: 10.1097/PHM.0b013e3181cf569b.
[40] M. J. Matarić and B. Scassellati, “Socially assistive robotics,” in Springer Handbook of Robotics, Cham: Springer International
Publishing, 2016, pp. 1973–1994, doi: 10.1007/978-3-319-32552-1_73.
[41] P. Polygerinos et al., “Soft robotics: Review of fluid‐driven intrinsically soft devices; manufacturing, sensing, control, and
applications in human‐robot interaction,” Advanced Engineering Materials, vol. 19, no. 12, Dec. 2017,
doi: 10.1002/adem.201700016.
[42] A. Kumar, A. Mantri, and R. Dutta, “Development of an augmented reality‐based scaffold to improve the learning experience of
engineering students in embedded system course,” Computer Applications in Engineering Education, vol. 29, no. 1, pp. 244–257,
Jan. 2021, doi: 10.1002/cae.22245.
[43] P. Maciejasz, J. Eschweiler, K. Gerlach-Hahn, A. Jansen-Troy, and S. Leonhardt, “A survey on robotic devices for upper limb
rehabilitation,” Journal of NeuroEngineering and Rehabilitation, vol. 11, no. 1, Dec. 2014, doi: 10.1186/1743-0003-11-3.
[44] J. Cheesborough, L. Smith, T. Kuiken, and G. Dumanian, “Targeted muscle reinnervation and advanced prosthetic arms,”
Seminars in Plastic Surgery, vol. 29, no. 01, pp. 62–072, Feb. 2015, doi: 10.1055/s-0035-1544166.

 ISSN: 2088-8708
6008
[45] A. Mohebbi, “Human-robot interaction in rehabilitation and assistance: a review,” Current Robotics Reports, vol. 1, no. 3,
pp. 131–144, Aug. 2020, doi: 10.1007/s43154-020-00015-4.
[46] M. Connan, E. Ruiz Ramírez, B. Vodermayer, and C. Castellini, “Assessment of a wearable force- and electromyography device
and comparison of the related signals for myocontrol,” Frontiers in Neurorobotics, vol. 10, Nov. 2016,
doi: 10.3389/fnbot.2016.00017.
[47] D. Farina et al., “The extraction of neural information from the surface EMG for the control of upper-limb prostheses: Emerging
avenues and challenges,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 22, no. 4, pp. 797–809,
Jul. 2014, doi: 10.1109/TNSRE.2014.2305111.
[48] C. L. Semasinghe, D. G. K. Madusanka, R. K. P. S. Ranaweera, and R. A. R. C. Gopura, “Transradial prostheses: Trends in
development of hardware and control systems,” International Journal of Medical Robotics and Computer Assisted Surgery,
vol. 15, no. 1, Oct. 2019, doi: 10.1002/rcs.1960.
[49] S. Grushko, T. Spurný, and M. Černý, “Control methods for transradial prostheses based on remnant muscle activity and its
relationship with proprioceptive feedback,” Sensors, vol. 20, no. 17, Aug. 2020, doi: 10.3390/s20174883.
[50] A. Phinyomark, R. N. Khushaba, and E. Scheme, “Feature extraction and selection for myoelectric control based on wearable
EMG sensors,” Sensors, vol. 18, no. 5, May 2018, doi: 10.3390/s18051615.
[51] E. Stålberg et al., “Standards for quantification of EMG and neurography,” Clinical Neurophysiology, vol. 130, no. 9,
pp. 1688–1729, Sep. 2019, doi: 10.1016/j.clinph.2019.05.008.
[52] Y. Geng et al., “A robust sparse representation based pattern recognition approach for myoelectric control,” IEEE Access, vol. 6,
pp. 38326–38335, 2018, doi: 10.1109/ACCESS.2018.2851282.
[53] S. Siuly, Y. Li, and Y. Zhang, EEG signal analysis and classification techniques and applications. Springer International
Publishing, 2016, doi: 10.1007/978-3-319-47653-7.
[54] A. Phinyomark, F. Quaine, S. Charbonnier, C. Serviere, F. Tarpin-Bernard, and Y. Laurillau, “EMG feature evaluation for
improving myoelectric pattern recognition robustness,” Expert Systems with Applications, vol. 40, no. 12, pp. 4832–4840,
Sep. 2013, doi: 10.1016/j.eswa.2013.02.023.
[55] M. Hamedi, S. H. Salleh, A. M. Noor, T. T. Swee, and I. K. Afizam, “Comparison of different time-domain feature extraction
methods on facial gestures’ EMGs,” Progress in Electromagnetics Research Symposium, pp. 1897–1900, 2012
[56] A. Balbinot and G. Favieiro, “A neuro-fuzzy system for characterization of arm movements,” Sensors, vol. 13, no. 2,
pp. 2613–2630, Feb. 2013, doi: 10.3390/s130202613.
[57] Y. Wu, S. Liang, L. Zhang, Z. Chai, C. Cao, and S. Wang, “Gesture recognition method based on a single-channel sEMG
envelope signal,” EURASIP Journal on Wireless Communications and Networking, vol. 2018, no. 1, Dec. 2018,
doi: 10.1186/s13638-018-1046-0.
[58] F. E. Abd El-Samie, T. N. Alotaiby, M. I. Khalid, S. A. Alshebeili, and S. A. Aldosari, “A review of EEG and MEG epileptic
spike detection algorithms,” IEEE Access, vol. 6, pp. 60673–60688, 2018, doi: 10.1109/ACCESS.2018.2875487.
[59] S. Ardeenawatie Awang, M. P. Paulraj, and S. Yaacob, “Analysis of EEG signals by eigenvector methods,” in IEEE-EMBS
Conference on Biomedical Engineering and Sciences, Dec. 2012, pp. 778–783, doi: 10.1109/IECBES.2012.6498164.
[60] A. Jalilifard, A. Rastegarnia, E. B. Pizzolato, and M. K. Islam, “Classification of emotions induced by horror and relaxing movies
using single-channel EEG recordings,” International Journal of Electrical and Computer Engineering (IJECE), vol. 10, no. 4,
pp. 3826–3838, Aug. 2020, doi: 10.11591/ijece.v10i4.pp3826-3838.
[61] R. Djemal, A. Bazyed, K. Belwafi, S. Gannouni, and W. Kaaniche, “Three-class EEG-based motor imagery classification using
phase-space reconstruction technique,” Brain Sciences, vol. 6, no. 3, Aug. 2016, doi: 10.3390/brainsci6030036.
[62] U. R. Acharya, S. V. Sree, P. C. A. Ang, R. Yanti, and J. S. Suri, “Application of non-linear and wavelet based features for the
automated identification of epileptic EEG signals,” International Journal of Neural Systems, vol. 22, no. 02, Apr. 2012,
doi: 10.1142/S0129065712500025.
[63] A. H. Al-Timemy, R. N. Khushaba, G. Bugmann, and J. Escudero, “Improving the performance against force variation of EMG
controlled multifunctional upper-limb prostheses for transradial amputees,” IEEE Transactions on Neural Systems and
Rehabilitation Engineering, vol. 24, no. 6, pp. 650–661, Jun. 2016, doi: 10.1109/TNSRE.2015.2445634.
[64] R. N. Khushaba, A. H. Al-Timemy, A. Al-Ani, and A. Al-Jumaily, “A framework of temporal-spatial descriptors-based feature
extraction for improved myoelectric pattern recognition,” IEEE Transactions on Neural Systems and Rehabilitation Engineering,
vol. 25, no. 10, pp. 1821–1831, Oct. 2017, doi: 10.1109/TNSRE.2017.2687520.
[65] B. Hudgins, P. Parker, and R. N. Scott, “A new strategy for multifunction myoelectric control,” IEEE Transactions on Biomedical
Engineering, vol. 40, no. 1, pp. 82–94, 1993, doi: 10.1109/10.204774.
[66] I. Isikli Esener, “Subspace-based feature extraction on multi-physiological measurements of automobile drivers for distress
recognition,” Biomedical Signal Processing and Control, vol. 66, Apr. 2021, doi: 10.1016/j.bspc.2021.102504.
[67] A. Phinyomark, P. Phukpattaranont, and C. Limsakul, “Fractal analysis features for weak and single-channel upper-limb EMG
signals,” Expert Systems with Applications, vol. 39, no. 12, pp. 11156–11163, Sep. 2012, doi: 10.1016/j.eswa.2012.03.039.
[68] A. S. Al-Fahoum and A. A. Al-Fraihat, “Methods of EEG signal features extraction using linear analysis in frequency and time-
frequency domains,” ISRN Neuroscience, vol. 2014, pp. 1–7, Feb. 2014, doi: 10.1155/2014/730218.
[69] K. K. Patro and P. Rajesh Kumar, “A novel frequency-time based approach for the detection of characteristic waves in
electrocardiogram signal,” in Lecture Notes in Electrical Engineering, vol. 372, Springer India, 2016, pp. 57–67,
doi: 10.1007/978-81-322-2728-1_6.
[70] S. Kuila, N. Dhanda, and S. Joardar, “Feature extraction of electrocardiogram signal using machine learning classification,”
International Journal of Electrical and Computer Engineering (IJECE), vol. 10, no. 6, pp. 6598–6605, Dec. 2020,
doi: 10.11591/ijece.v10i6.pp6598-6605.
[71] Q. Li, A. Zhang, Z. Li, and Y. Wu, “Improvement of EMG pattern recognition model performance in repeated uses by combining
feature selection and incremental transfer learning,” Frontiers in Neurorobotics, vol. 15, Jun. 2021,
doi: 10.3389/fnbot.2021.699174.
[72] A. Subasi, “Medical decision support system for diagnosis of neuromuscular disorders using DWT and fuzzy support vector
machines,” Computers in Biology and Medicine, vol. 42, no. 8, pp. 806–815, Aug. 2012,
doi: 10.1016/j.compbiomed.2012.06.004.
[73] M. Hasan, M. M. Islam, M. I. I. Zarif, and M. M. A. Hashem, “Attack and anomaly detection in IoT sensors in IoT sites using
machine learning approaches,” Internet of Things, vol. 7, Sep. 2019, doi: 10.1016/j.iot.2019.100059.
[74] M. T. Ribeiro, S. Singh, and C. Guestrin, “‘Why should I trust you?,’” in Proceedings of the 22nd
ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining, Aug. 2016, pp. 1135–1144, doi: 10.1145/2939672.2939778.

6009
BIOGHRAPHES OF AUTHORS
Huda Mustafa Radha Computer Sciences Department/College of Science,
Baghdad University, Iraq. B.Sc. in Computer Sciences, Baghdad University 1995, High
Diploma Computer Sciences, University of Baghdad 2010, M.Sc. at the Computer Sciences,
University of Baghdad 2018. From 2006 to 2018 Teaching assistant at the University of
Baghdad, Science college, Baghdad, Iraq. Publication Published (8) papers in Conferences
and Journals. Current Research Interests AI, Data Mining, Biomedical Signal analysis,
Robotics, and Image processing. Currently, she is a lecturer in the Computer Science
department and a student for a Ph.D. at Computer Sciences, University of Technology. She
can be contacted at email: huda.rada@sc.uobaghdad.edu.iq.
Alia Karim Abdul Hassan Computer Sciences Department, the University of
Technology, Baghdad, Iraq. B.Sc. at Computer Sciences, University of Technology on 1993,
M.Sc. at Computer Sciences, University of Technology on 1999, Ph.D. Computer Sciences,
University of Technology on 2004. Professor since 2019. Position Dean of Computer Science
Department since Feb 2019 till now. Supervised 26 M.Sc. and Ph.D. thesis in Computer
Science since 2007. Publications Published more than (85) papers in International
Conferences and Journals. Current research interests soft computing, green computing, AI,
data mining, software engineering, electronic management, computer security. She can be
contacted at email: 110018@uotechnology.edu.iq.

A survey on bio-signal analysis for human-robot interaction

More Related Content

Similar to A survey on bio-signal analysis for human-robot interaction (20)

More from IJECEIAES (20)

Recently uploaded (20)

A survey on bio-signal analysis for human-robot interaction