IEEE SENSORS JOURNAL, VOL. 21, NO.

7, APRIL 1, 2021 9515

A Hybrid Posture Detection Framework:

Integrating Machine Learning and
Deep Neural Networks
Sidrah Liaqat, Kia Dashtipour, Kamran Arshad , Senior Member, IEEE,
Khaled Assaleh, Senior Member, IEEE, and Naeem Ramzan , Senior Member, IEEE

Abstract —The posture detection received lots of attention

in the fields of human sensing and artificial intelligence.
Posture detection can be used for the monitoring health
status of elderly remotely by identifying their postures such as
standing, sitting and walking. Most of the current studies used
traditional machine learning classifiers to identify the posture.
However, these methods do not perform well to detect the
postures accurately. Therefore, in this study, we proposed a
novel hybrid approach based on machine learning classifiers
(i. e., support vector machine (SVM), logistic regression
(KNN), decision tree, Naive Bayes, random forest, Linear
discrete analysis and Quadratic discrete analysis) and deep
learning classifiers (i. e., 1D-convolutional neural network (1D-
CNN), 2D-convolutional neural network (2D-CNN), LSTM and bidirectional LSTM) to identify posture detection. The
proposed hybrid approach uses prediction of machine learning (ML) and deep learning (DL) to improve the performance
of ML and DL algorithms. The experimental results on widely benchmark dataset are shown and results achieved an
accuracy of more than 98%.
Index Terms — Posture detection, hybrid approach, deep learning, machine learning.

I. I NTRODUCTION the technology, the human has chosen the sedentary lifestyle

T HE posture detection is used in different applications

such as healthcare, surveillance, virtual environment,
indoor and outdoor monitoring, the reality for animation and
which leading to less physical activity and movement [1]–[5].
The long time sitting during the work or study leads to
decrease in muscle strength. The sedentary lifestyle have
entertainment. In addition, the posture detection can be used negative impact on body human, not caring about correct
in framework of home-human interface. With the increased posture or fault posture can lead pain in neck, back and shoul-
number of elderly population and limited healthcare resources, der. Therefore, it is important to control the human posture
it is important to propose a technology which can support the to maintain their health and safety during work or study.
remote monitoring of elderly and vulnerable people to live Considering the need, the paper reports three major contri-
more independently. Maintain the good posture is significant to butions that are outlined below:
lead the healthy life. The posture is about how the people hold
• In this paper, we implemented a novel CNN and LSTM
their body and position the limbs. Within the advancement of
architecture for automatically identify the posture detec-
Manuscript received January 8, 2021; revised January 25, 2021; tion. It is worth to mention, the deep learning classifiers
accepted January 26, 2021. Date of publication February 1, 2021; date unlike machine learning algorithms do not require hand-
of current version March 5, 2021. This work was supported in part
by the Ajman University Internal Research Grant. The associate editor
crafted features.
coordinating the review of this article and approving it for publication was • In addition, a novel hybrid approach based on DL
Dr. Qammer H. Abbasi. (Corresponding author: Sidrah Liaqat.) (1D-CNN, 2D-CNN, LSTM, BiLSMT) and ML (random
Sidrah Liaqat and Naeem Ramzan are with the School of Engineering
and Computing, University of the West of Scotland, Paisely PA1 2BE,
forest, KNN, Naive Bayes, decision tree, LDA, QDA and
U.K. (e-mail: [email protected]; [email protected]). SVM) methods developed to identify the posture.
Kia Dashtipour is with the James Watt School of Engineering, Uni- • There is an extensive comparative experimental results
versity of Glasgow, Glasgow G12 8QQ, U.K. (e-mail: kia.dashtipour@
glasgow.ac.uk).
that are conducted with state-of-the-art approaches to
Kamran Arshad and Khaled Assaleh are with the College of Engi- evaluate the performance of our proposed approach.
neering and IT, Ajman University, Ajman, UAE (e-mail: k.arshad@
ajman.ac.ae; [email protected]). This paper discuss the machine learning (ML) and deep
Digital Object Identifier 10.1109/JSEN.2021.3055898 learning (DL) methods for posture detection which is used to

1558-1748 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Sharda University. Downloaded on May 02,2022 at 13:50:19 UTC from IEEE Xplore. Restrictions apply.
9516 IEEE SENSORS JOURNAL, VOL. 21, NO. 7, APRIL 1, 2021

monitor human activity. In this study, we focus on the selection give better performances [11]. The sensor chair with pressure
of ML, DL and hybrid methods to increase the performance sensor tries to avoid wrong sitting position which may cause
of posture recognition. The activity which is recognised are disease. In this posture detection, analysis is compared with
sitting and standing. The sitting and standing is important decision tree and random forest. The classifier which gives
posture to detect because human can monitor their own activity better performance is random forest classifier [12]. For the
if they are sitting for long time, they can stand for some improvement of sitting posture, sitting posture monitoring
activity. systems (SPMSs) is used. It has mounted sensors on backrest
The paper is organised into the following sections. Section II and seat plate of a chair. For this experiment 6 sitting postures
provides detailed related work within the field of radar based are considered. Then various machine learning algorithms
motion detection. Section III presents the methodology of how (SVM with RBF kernel, SVM linear, random forest, QDA,
the experimentation of this research has been done. Section IV LDA, NB and DT) are applied on body weight ratio which
discusses the results obtained through the experimentation. is measured by SPMS. Result from SVM with RBF kernel
Section V gives the conclusion. gives better accuracy as compare to others [13]. There is
also an intelligent systems design for the posture detection
II. R ELATED W ORK of sitting person on wheel chair. A network of sensors is
In the literature, extensive research has been carried out to used for data collection using neighbourhood rule (CNN),
build different posture models. In this section, we summarize then data balancing is done with Kennard-stone algorithm
the related state-of-the-art approaches for detection of human and reduction in dimensions via principal component analysis.
postures. Finally k-nearest algorithm is applied to pre-processed and
In smart city prediction and supervision of human health, balanced data. In this amount of data is significantly reduced
using smart technology and portable system is an important but result is remarkable [14].
part. Therefore, posture recognition in this paper is determined A postural habit which has been formed cannot change
with multisensory and using LoRa (Long Range) technology. easily so it is vital to form a proper postural habit since
LoRa WAN technology has the advantage of long transmission childhood. Therefore, machine learning algorithms CNN, NB,
distances and low cost. Using these two, multisensory and DT, NN, MLR and SVM are used for posture detection. Data
loRa technology, wearable clothes are designed which is is collected with a sensing cushion which is developed with
comfortable in any given posture. In this paper multiprocessing (8 × 8) pressure sensor mat inside children chair seat cushion.
is used because LoRa has low transmitting frequency and Ten children are participated for five prescribed postures. The
data transmission size is small. Hence, multiprocessing is accuracy of CNN is the highest than other algorithms [15].
done by sliding window, feature extraction, data processing Dance is also a challenging task for posture recognition. It is
and feature selection is done with Random forest. With multimedia in nature and its duration is over time as well
three testers of 500 grouped data set better performance as space. For dance analysis few things must be undertaken
and accuracy is achieved [6]. Body postures and gestures like segment of the dance video, recognition of the detected
are also non-verbal way of communication. In this paper, action element and recognition of the dance sequences. In this
Augmented reality is used to determine the static posture paper focus is on Indian classical dance, Bharatanatyam,
by reducing cost and advanced body tracking technology. which is driven by music as well as motion for posture
Also unsupervised machine learning on Kinect body posture recognition. Then recognition is done by machine and deep
sensors are used to detect group cooperation and learning [7]. learning techniques which are GMM, SVM and CNN. In the
Posture detection has also play an important role in performing final step Hidden Markov Model (HMM) is applied for data
yoga more accurately. Posture recognition is challenging task sequence recognition. The best recognition rate is with CNN
due to real-time bases and less available data set. Therefore, classifier [16].
to overcome this issue large data set has been created with
at least 5500 images of different yoga poses. For posture III. M ETHODOLOGY
detection tf-pose estimation algorithm has been used which This section describes our proposed approach for the detec-
draws skeleton of human body on the real-time bases. Angles tion of posture. The Fig shows the overall framework of
of the joints in the human body are extracted using the tf-pose posture detection.
skeleton which is used as a feature to implement various
machine learning models (SVM, KNN, Logistic Regression,
DT, NB and random forest). Among all Random forest model A. Feature Extraction
gives best accuracy [8]–[10]. There are total of six features are used for posture predic-
In addition, there is another problem of posture in human tion. The features determine the posture are skew, percentile,
being which occurs due to maximum time in sitting position. square root (SR), standard deviation (SD), mean and kurtosis.
The poor and prolonged sitting effects physical and mental The values for each feature is calculated individually for each
health. Posture training system is designed for sitting position window size. For example, the window size of 90 seconds is
and stretch pose data collection. Then for posture recognition, selected and the aforementioned features are calculated. After
smart cushion using Artificial Intelligence (AI) and pressure the feature extraction, the new dataset is created which consists
sensing technologies is used. For more than 13 different pos- of different features. It is to be noted that, after feature extrac-
tures, supervised machine learning models are trained which tion, the most important task is to determines the combination

Authorized licensed use limited to: Sharda University. Downloaded on May 02,2022 at 13:50:19 UTC from IEEE Xplore. Restrictions apply.
LIAQAT et al.: HYBRID POSTURE DETECTION FRAMEWORK: INTEGRATING ML AND DNNs 9517

Fig. 1. Overview of the proposed framework for posture detection.

TABLE I connected layers. The 10-layered CNN framework achieved

M ACHINE L EARNING M ETHODS W ITH T HEIR PARAMETERS the most promising results. The parameters of CNN framework
are shown in Table II
Long Short Term Memory (LSTM): The long short term
memory (LSTM) proposed architecture contains input layer,
two different stacked LSTM and one output as fully connected
layer. Particularly, the LSTM architecture consists of two
different stacked bidirectional layers (contains 128 cells and
64 cells) with dropout 0.2 and a dense layer with two neurons
and softmax activation.

D. Hybrid Models
The hybrid methods consists of different classifiers and
combining their prediction to train meta-learning model. The
hybrid is used to enhance the performance of specific sys-
of best features for posture prediction in term of accuracy.
tem. In this study, the prediction of ML classifiers (logistic
In total six features and combination of these features are
regression, random forest, KNN, Naïve Bayes, decision tree,
evaluated using different ML models. In total combination of
linear discriminant analysis, quadratic discriminant analysis
features for posture prediction is time consuming therefore,
and SVM) and DL classifiers (CNN, LSTM) are used as input
we employed DL methods which dp not require any feature
of CNN, LSTM architecture. Fig 2 shows the architecture of
engineering and they usually obtain superior performance as
proposed hybrid of ML and DL for posture detection. It is to
compared to ML models.
be noted that, the parameters of each classifier has been set-up
empirically after several simulation experiments.
B. Machine Learning (ML) Methods
After feature extraction, there are different machine learning IV. E XPERIMENTAL R ESULTS
classifiers, including SVM, logistic regression, KNN, deci- In order to classify the posture prediction, standard (logistic
sion tree, naive bayes, random forest, LDA and QDA have regression, random forest, KNN, Naïve Bayes, decision tree,
been applied in order to evaluate the performance of the linear discriminant analysis, quadratic discriminant analysis
approach. Table I shows the parameters that are used to trained and SVM) and deep learning classifiers such as (1D-CNN,
the machine learning methods. The scikit-learn package is 2D-CNN, LSTM, BiLSTM) are trained. We extracted differ-
used to train the machine learning classifiers. In addition, ent features including skew, percentile, SR, SD, mean and
the training time for each models has been presented in kurtosis. It is worth to mention that, there are total thirteen
Table I. experiments have been done. In addition, the 10-fold cross-
validation is used to perform the experiments. In order to
evaluate the performance of the proposed approach, precision,
C. Deep Learning (DL) Methods
recall, F-score and accuracy metrics were used:
In next section, we discuss our proposed hybrid model
TP
which integrates the machine learning and DNN methods Pr eci si on = (1)
including 1D-CNN, 2D-CNN, LSTM and BiLSTM. The deep T P + FP
TP
learning can be used in different various application such as Recall = (2)
cyber-security, sentiment analysis, speech enhancement and T P + FN
etc. However, in this paper we proposed a novel framework Pr eci si on ∗ Recall
F_measur e = 2 ∗ (3)
to detect posture prediction. Pr eci si on + Recall
Convolutional Neural Network: For comparison, the novel TP +TN
Accur acy = (4)
CNN framework is developed. The implemented CNN con- T P + T N + FP + FN
sists of input, hidden and output layers. Our proposed CNN Human Body Posture Detection: In order to evaluate the
framework contains convolutional, max pooling and fully performance of the approach. The online widely benchmark

Authorized licensed use limited to: Sharda University. Downloaded on May 02,2022 at 13:50:19 UTC from IEEE Xplore. Restrictions apply.
9518 IEEE SENSORS JOURNAL, VOL. 21, NO. 7, APRIL 1, 2021

TABLE II
CNN A RCHITECTURE (C ONV - C ONVOLUTIONAL L AYER , M AX P OOL - M AXPOOLING L AYER , G LOBAL M AX P OOL - G LOBAL
M AX P OOLING L AYER , F C - F ULLY C ONNECTED L AYER , R E LU - R ECTIFIED L INEAR U NIT ACTIVATION

Fig. 2. Proposed Hybrid Classifier.

dataset called human body using galvanic skin response have TABLE III
been used. The data is collected for five different subjects, and S UMMARY OF L OGISTIC R EGRESSION FOR P OSTURE P REDICTION
it has been classified into three different categories such as
standing, sitting and walking. In addition, the data is recorded
at a resolution of 16 bits in samples of 5 min to 15 min and
the sampling rate is 1 MHz (maximum precision position on
the BITalino Kit) [17]. There are different machine learning
methods to train the classifiers for classifying the posture
such as standing, sitting and walking. In order to evaluate
the performance of the approach, the data used in our study is
collected from five different individuals, the dataset consists of
four males and females for different ethnicity, all the bracket
of 25 to 30 years of age.
The machine learning algorithms are trained based on the that, the scikit-learn with Tensorflow background is used to
10-fold cross-validation and train/test used Python variables implement the deep learning approaches.
containing the data and comparing the prediction of the data Table III shows the summary of results for selected features
to the actual labels of the data. There are different evaluation using logistic regression. The experimental results show the
metrics such as accuracy, precision, recall and f-measure are combination of all features achieved better performance and
used to compare the current algorithms. It is worth to mention mean feature achieved less accuracy.

Authorized licensed use limited to: Sharda University. Downloaded on May 02,2022 at 13:50:19 UTC from IEEE Xplore. Restrictions apply.
LIAQAT et al.: HYBRID POSTURE DETECTION FRAMEWORK: INTEGRATING ML AND DNNs 9519

TABLE IV TABLE VII

S UMMARY OF R ANDOM F OREST FOR P OSTURE P REDICTION S UMMARY OF D ECISION T REE FOR P OSTURE P REDICTION

TABLE V TABLE VIII

S UMMARY OF KNN FOR P OSTURE P REDICTION S UMMARY OF LDA FOR P OSTURE P REDICTION

TABLE VI TABLE IX
S UMMARY OF N AIVE B AYES FOR P OSTURE P REDICTION S UMMARY OF SVM FOR P OSTURE P REDICTION

Table IV shows the summary of results for selected features SR and percentile features achieved better performance. In the
using random forest. The experimental results show the com- other hand, combination of all feature achieved less accuracy.
bination of all features achieved better performance. In the Table IX shows the summary of results for selected features
other hand, mean and SD feature achieved less accuracy. using SVM. The experimental results show the combination
Table V shows the summary of results for selected features of all features achieved better performance. In the other hand,
using KNN. The experimental results show the combination mean and SD features achieved less accuracy.
of all features achieved better performance. In the other hand, Table X shows the summary of results for selected features
mean feature achieved less accuracy. using QDA. The experimental results show the mean feature
Table VI shows the summary of results for selected features achieved better performance. In the other hand, mean, SD and
using Naive Bayes. The experimental results show the mean SR features achieved less accuracy.
and SD and also mean, SD and SR features achieved better Table XI shows the summary of results for selected features
performance. In the other hand, mean, SD, SR and percentile using 1D-CNN. The experimental results show the raw data
features achieved less accuracy. (without any feature selection) achieved better performance.
Table VII shows the summary of results for selected fea- In the other hand, mean feature achieved less accuracy.
tures using decision tree. The experimental results show the Table XII shows the summary of results for selected features
combination of all features achieved better performance. In the using 2D-CNN. The experimental results show the raw data
other hand, mean and SD features achieved less accuracy. (without any feature selection) achieved better performance.
Table VIII shows the summary of results for selected fea- In the other hand, mean and SD features achieved less
tures using LDA. The experimental results show the mean, SD, accuracy.

Authorized licensed use limited to: Sharda University. Downloaded on May 02,2022 at 13:50:19 UTC from IEEE Xplore. Restrictions apply.
9520 IEEE SENSORS JOURNAL, VOL. 21, NO. 7, APRIL 1, 2021

TABLE X TABLE XIII

S UMMARY OF QDA FOR P OSTURE P REDICTION S UMMARY OF LSTM FOR P OSTURE P REDICTION

TABLE XI
S UMMARY OF 1D-CNN FOR P OSTURE P REDICTION TABLE XIV
S UMMARY OF B I LSTM FOR P OSTURE P REDICTION

TABLE XII TABLE XV

S UMMARY OF 2D-CNN FOR P OSTURE P REDICTION S UMMARY OF H YBRID (CNN) FOR P OSTURE P REDICTION

Table XIII shows the summary of results for selected fea- true negative 2164, the false positive 31 and the false negative
tures using LSTM. The experimental results show the raw data is 20.
(without any feature selection) achieved better performance. Table XVI shows the summary of results for selected fea-
In the other hand, mean feature achieved less accuracy. tures using hybrid 2D-CNN. The experimental results show
Table XIV shows the summary of results for selected the raw data (without any feature selection) achieved better
features using BiLSTM. The experimental results show the performance. In the other hand, mean feature achieved less
raw data (without any feature selection) achieved better per- accuracy.
formance. In the other hand, mean feature achieved less Table XVII shows the summary of results for selected
accuracy. features using hybrid LSTM. The experimental results show
Table XV shows the summary of results for selected fea- the raw data (without any feature selection) achieved better
tures using hybrid CNN. The experimental results show the performance. In the other hand, mean feature achieved less
raw data (without any feature selection) achieved better per- accuracy.
formance. In the other hand, mean feature achieved less Table XVIII shows the summary of results for selected
accuracy. It is worth to mention that, CNN frame-work consists features using hybrid BiLSTM. The experimental results show
of convolutional, max pooling and fully connected layers. the raw data (without any feature selection) achieved better
The 10-layered CNN framework achieved the most promising performance. In the other hand, mean feature achieved less
results. In addition, the true positive for hybrid CNN 1625, accuracy.

Authorized licensed use limited to: Sharda University. Downloaded on May 02,2022 at 13:50:19 UTC from IEEE Xplore. Restrictions apply.
LIAQAT et al.: HYBRID POSTURE DETECTION FRAMEWORK: INTEGRATING ML AND DNNs 9521

TABLE XVI TABLE XIX

S UMMARY OF H YBRID (2D-CNN) FOR P OSTURE P REDICTION C OMPARISON W ITH S TATE - OF - THE -A RT A PPROACHES

accuracy, precision, recall and f-measure as compared to other

approaches.
TABLE XVII
S UMMARY OF H YBRID (LSTM) FOR P OSTURE P REDICTION V. C ONCLUSION
The remote health monitoring is important for providing
independent living to elderly and vulnerable. Therefore, in this
paper, we proposed a novel architecture based on deep learning
classifiers to identify posture including standing, sitting and
walking. In addition, the novel hybrid approach are developed
based on the DL methods to identify the posture prediction.
The hybrid approach contains different prediction of machine
learning and deep learning to train the meta-learning. The
experimental results show that the proposed hybrid approach
achieved better performance as compared to DL and ML
methods.
TABLE XVIII
S UMMARY OF H YBRID (B I LSTM) FOR P OSTURE P REDICTION R EFERENCES
[1] A. Koubâa et al., “Activity monitoring of islamic prayer (salat) postures
using deep learning,” in Proc. 6th Conf. Data Sci. Mach. Learn. Appl.
(CDMA), Mar. 2020, pp. 106–111.
[2] A. Adeel et al., “A survey on the role of wireless sensor networks and
IoT in disaster management,” in Geological Disaster Monitoring Based
on Sensor Networks. Singapore: Springer, 2019, pp. 57–66.
[3] J. Lee, H. Joo, J. Lee, and Y. Chee, “Automatic classification of squat
posture using inertial sensors: Deep learning approach,” Sensors, vol. 20,
no. 2, p. 361, Jan. 2020.
[4] F. Jiang, K. Dashtipour, and A. Hussain, “A survey on deep learning
for the routing layer of computer network,” in Proc. UK/China Emerg.
Technol. (UCET), Aug. 2019, pp. 1–4.
[5] F. Jiang, B. Kong, J. Li, K. Dashtipour, and M. Gogate, “Robust visual
saliency optimization based on bidirectional Markov chains,” Cogn.
Comput., pp. 1–12, 2020.
[6] J. Han et al., “LoRa-based smart IoT application for smart city:
An example of human posture detection,” Wireless Commun. Mobile
A. Discussion Comput., vol. 2020, Aug. 2020, Art. no. 8822555.
The comparative experimental results show that the hybrid [7] I. Radu, E. Tu, and B. Schneider, “Relationships between body postures
deep learning method achieved better performance as com- and collaborative learning states in an augmented reality study,” in Proc.
Int. Conf. Artif. Intell. Educ. Cham, Switzerland: Springer, Jul. 2020,
pared to traditional machine learning and deep learning classi- pp. 257–262.
fiers. The performance achieved on the dataset is to identify the [8] Y. Agrawal, Y. Shah, and A. Sharma, “Implementation of machine
posture prediction shows that the unique hybrid DL approaches learning technique for identification of yoga poses,” in Proc. IEEE 9th
Int. Conf. Commun. Syst. Netw. Technol. (CSNT), Apr. 2020, pp. 40–43.
outperformed the ML and DL technique. In order to train [9] M. A. I. R. Ghannam and Q. Abbasi, Engineering and Technology for
the DL classifiers, the NVIDIA GeForce 940M GPU with Healthcare. Hoboken, NJ, USA: Wiley, 2021.
384 Cuda cores and 2 GB DDR3 has been used. The main [10] I. O. Hussien, K. Dashtipour, and A. Hussain, “Comparison of sentiment
analysis approaches using modern Arabic and Sudanese dialect,” in
advantage of hybrid DL classifiers is that their ability to Proc. Int. Conf. Brain Inspired Cogn. Syst. Cham, Switzerland: Springer,
identify the posture. However, the dataset is only limited to Jul. 2018, pp. 615–624.
three posture including standing, sitting and walking. The [11] K. Bourahmoune and T. Amagasa, “AI-powered posture training: Appli-
cation of machine learning in sitting posture recognition using the
main disadvantage of the approach is DL classifiers are LifeChair smart cushion,” in Proc. 28th Int. Joint Conf. Artif. Intell.,
computationally expensive. Table XIX shows the comparison Aug. 2019, pp. 5808–5814.
of state-of-the-art approaches with our proposed approach. The [12] M. Sandybekov, C. Grabow, M. Gaiduk, and R. Seepold, “Posture
tracking using a machine learning algorithm for a home AAL environ-
comparison results show that our approach achieved superior ment,” in Intelligent Decision Technologies. Singapore: Springer, 2019,
performance in term different evaluation metrics including pp. 337–347.

Authorized licensed use limited to: Sharda University. Downloaded on May 02,2022 at 13:50:19 UTC from IEEE Xplore. Restrictions apply.
9522 IEEE SENSORS JOURNAL, VOL. 21, NO. 7, APRIL 1, 2021

[13] J. Roh, H.-J. Park, K. Lee, J. Hyeong, S. Kim, and B. Lee, “Sitting Khaled Assaleh (Senior Member, IEEE)
posture monitoring system based on a low-cost load cell using machine received the B.Sc. degree in electrical
learning,” Sensors, vol. 18, no. 2, p. 208, Jan. 2018. engineering from the University of Jordan
[14] P. D. Rosero-Montalvo, D. H. Peluffo-Ordonez, V. F. Lopez Batista, in 1988, the master’s degree in electronic
J. Serrano, and E. A. Rosero, “Intelligent system for identification of engineering from Monmouth University, NJ,
wheelchair user’s posture using machine learning techniques,” IEEE USA, in 1990, and the Ph.D. degree in electrical
Sensors J., vol. 19, no. 5, pp. 1936–1942, Mar. 2019. engineering from Rutgers, The State University
[15] Y. Kim, Y. Son, W. Kim, B. Jin, and M. Yun, “Classification of children’s of New Jersey in 1993. From 2002 to 2017,
sitting postures using machine learning algorithms,” Appl. Sci., vol. 8, he was with the American University of
no. 8, p. 1280, Aug. 2018. Sharjah (AUS) as a Professor of Electrical
[16] T. Mallick, P. P. Das, and A. K. Majumdar, “Posture and sequence Engineering and the Vice Provost for Research
recognition for Bharatanatyam dance performances using machine and Graduate Studies. Prior to joining AUS, he had a nine-year R&D
learning approach,” 2019, arXiv:1909.11023. [Online]. Available: career in the Telecom Industry in the USA with Rutgers, Motorola, and
http://arxiv.org/abs/1909.11023 Rockwell/Skyworks. He is currently the Vice Chancellor for Academic
[17] A. Rizwan et al., “Non-invasive hydration level estimation in human Affairs and a Professor of Electrical Engineering at Ajman University.
body using galvanic skin response,” IEEE Sensors J., vol. 20, no. 9, He holds 11 U.S. patents and has published over 120 articles in
pp. 4891–4900, May 2020. signal/image processing and pattern recognition and their applications.
[18] W. Xu, M.-C. Huang, N. Amini, L. He, and M. Sarrafzadeh, “ECushion: His research interests include bio-signal processing, biometrics, speech
A textile pressure sensor array design and calibration for sitting posture and image processing, and pattern recognition. He has served on
analysis,” IEEE Sensors J., vol. 13, no. 10, pp. 3926–3934, Oct. 2013. organization committees of several international conferences including
[19] C. Castellini, E. Gruppioni, A. Davalli, and G. Sandini, “Fine detection ICIP, ISSPA, ICCSPA, MECBME, and ISMA. He has also served as a
of grasp force and posture by amputees via surface electromyography,” Guest Editor for several special issues of journals.
J. Physiol.-Paris, vol. 103, nos. 3–5, pp. 255–262, May 2009.
[20] J. Sanghvi, G. Castellano, I. Leite, A. Pereira, P. W. McOwan, and
A. Paiva, “Automatic analysis of affective postures and body motion
to detect engagement with a game companion,” in Proc. 6th Int. Conf.
Human-Robot Interact. (HRI), 2011, pp. 305–312.
[21] M. Winters et al., “Knowledge engineering for unsupervised canine
posture detection from IMU data,” in Proc. 12th Int. Conf. Adv. Comput.
Entertainment Technol., Nov. 2015, pp. 1–8.
Naeem Ramzan (Senior Member, IEEE) recei-
ved the M.Sc. degree in telecommunications
Sidrah Liaqat is pursuing the Ph.D. degree with from the University of Brest, France, in 2004, and
the School of Engineering and Computing, Uni- the Ph.D. degree in electronics engineering from
versity of the West of Scotland. She is working the Queen Mary University of London, London,
on remote healthcare technologies and machine U.K., in 2008.
learning. He is currently a Full Professor of Artificial
Intelligence and the Director of the Affective
and Human Computing for Smart Environ-
ment (AHCSE) Research Centre, University of
the West of Scotland (UWS), U.K. He has
authored or coauthored more than 200 research publications, includ-
ing journals, book chapters, and standardization contributions. He has
authored a book and coedited some books as well. His research interests
Kia Dashtipour received the M.Sc. degree in are cross-disciplinary and industry focused and include AI/machine
advanced computer system development and learning, affective computing and multimedia processing, analysis and
the Ph.D. degree in computing science from the communication, video quality evaluation, brain-inspired multi-modal cog-
University of Stirling in 2014 and 2019, respec- nitive technology, big data analytics, affective computing, the IoT/smart
tively. He was a Postdoctoral Research Asso- environments, natural multi-modal human–computer interaction, and
ciate with the Edinburgh Napier University from eHealth/connected Health. His article was awarded the Best Paper
2018 to 2019. He is currently a Research Asso- Award 2017 of the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR
ciate with the James Watt School of Engineer- VIDEO TECHNOLOGY and a number of conference papers were selected
ing, University of Glasgow. His main research for the Best Student Paper Award. He has been a Lead Researcher
interests include natural language processing, in various nationally or EU sponsored multimillion funded international
machine learning, and speech enhancement. research projects (total funding as PI secured over £20m). He has been
awarded the Scottish Knowledge Exchange Champion Award 2020 and
numerous other awards, such as the Staff Appreciation and Recog-
Kamran Arshad (Senior Member, IEEE) is nition Scheme (STARS) Award for Leadership in 2019 and awarded
currently the Dean of Graduate Studies and STARS award 2015 and 2017 for Outstanding Research and Knowledge
Research and a Professor in Electrical Engi- Exchange (the University of the West of Scotland) and Awarded Con-
neering with Ajman University, UAE. Prior to tribution Reward Scheme 2011 and 2009 for outstanding research and
joining Ajman University, in January 2016, he teaching activities (The Queen Mary University of London). He is a Senior
has been with the University of Greenwich, U.K., Fellow of the Higher Education Academy (HEA), the Co-Chair of MPEG
as a Senior Lecturer and the Program Director HEVC verification (AHG5) Group, and a Voting Member of the British
of M.Sc. Wireless Mobile Communications Sys- Standard Institution (BSI). In addition, he holds key roles in the Video
tems Engineering. He is also a Senior Fellow of Quality Expert Group (VQEG), such as the Co-Chair of the Ultra High
the U.K. Higher Education Academy (SFHEA). Definition (UltraHD) Group, the Co-Chair of the Visually Lossless Quality
He led a number of locally and internationally Analysis (VLQA) Group, and the Co-Chair of the Psycho-Physiological
funded research projects encompassing the areas of cognitive radio, Quality Assessment (PsyPhyQA). He is also the Co-Editor-in-Chief of
LTE/LTE advanced, 5G, optimization, and cognitive Machine-to-Machine VQEG eLetter. He has served as a Guest Editor for a number of journals.
(M2M) communications. He has contributed to several European and He is also a Founding Associate Editor of the Journal of Quality and User
international large-scale research projects. He has over 130 technical Experience (Springer) and an Associate Editor of number of journals.
peer-reviewed articles in top quality journals and international confer- He has chaired/co-chaired/organized more than 25 workshops, special
ences, received three Best Paper Awards, one best Research and sessions, and tracks in international conferences. He has developed a
Development Track Award, and chaired technical sessions in several highly innovative portfolio of post graduate studies, including the M.Sc.
leading international conferences. He is an Associate Editor of the degrees in advanced computing, big data, the IoT, and eHealth/digital
EURASIP Journal on Wireless Communications and Networking. health.

Authorized licensed use limited to: Sharda University. Downloaded on May 02,2022 at 13:50:19 UTC from IEEE Xplore. Restrictions apply.