Content uploaded by Mouazma Batool
Author content
All content in this area was uploaded by Mouazma Batool on Jun 29, 2022
Content may be subject to copyright.
XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE
Fundamental Recognition of ADL Assessments
Using Machine Learning Engineering
Mouazma Batool
Dept. of Computer Science
Air University
Islamabad, Pakistan
181788@students.au.edu.pk
Madiha Javeed
Dept. of Computer Science
Air University
Islamabad, Pakistan
191880@students.au.edu.pk
Abstract—This paper describes an RGBD (Red Green Blue
Depth) image daily life activity identification system that can
monitor and detect human activities without the need of optical
markers or motion sensors. In this paper, we have developed a
practical methodology for detecting human activity in the daily
environment. For feature extraction technique, we suggested a
noble method for recognition of ADL activities based on
symmetry principle. By extracting silhouette from depth images
and performing mapping operations over RGB images, we have
been able to extract the skeleton information of the human body
in RGBD images and identify ADL activities using four critical
parameters: angle formation between hand and upper half of
the body, angle between center body point and hand, angle
formation between hand and lower half of the body, and angle
between two hands of the single silhouette. We employed the
linearly dependent concept (LDC) and long short-term memory-
recurrent neural networks (LSTM-RNN) for feature selection
and classification, compared the results to existing approaches.
The objective of our research is to not only find an effective and
useful collection of features from the silhouette, but also to
outperform current approaches. Finally, the suggested method's
testing results showed a 2.5 percent increase in accuracy with a
92.83 percent success rate, as well as a reduction in relative error
to 2.47 percent of the original dataset.
Keywords— activities of daily living, linearly dependent
concept, recurrent neural network, symmetry principle.
I. INTRODUCTION
Activities of daily living (ADL) identification has gotten
importance in a wide range of applications including
surveillance, human-computer interaction, pedestrian
detection systems, healthcare systems, homo-robotics, and
smart homes [1-7]. Most ADL recognition systems utilize
grey or RGB video sequences [8-11]. Here, the extraction of
elements of human actions in clutter background, light
variations, and camera motion is a big difficulty here [12].
Moreover, the appearance information given by RGB images
is not resilient against typical fluctuations such as illuminance
shift, making deployment of RGB-based vision systems in
real-world settings difficult [13]. Depth sensors have become
widely employed in recent years in a variety of applications,
including object recognition, object detection, scene labelling,
and action recognition [14-24]. Because depth information
provides extra information about the object's spatial
dimension, which is invariant to color fluctuation and light, it
can considerably increase object identification ability [25].
Meanwhile, it has been demonstrated that utilizing RGB-D
information for ADL improves classification accuracy
significantly [26-27].
Most academics have focused their efforts on developing
increasingly complex algorithms, while some have taken a
different approach in quest of a new sort of representation that
can better interpret the world. Cartas et al. [28] The
ADLEgoDataset of Daily Living, which contains 105,529
annotated photos, was launched. The dataset was examined
using the CNN+LSTM algorithm and attained an accuracy of
80.12%. Yu et al. [29] have proposed a knowledge-driven
multisource fusion architecture for recognizing egocentric
behaviors in everyday life (ADL). The Dezert–Smarandache
theory and a convolutional neural network has trained to
provide a collection of textual tags to distinguish ADL
activities. Experiments indicate that the proposed technique
correctly classified 15 preset ADL classes with an average
accuracy of 85.4 percent. Heidarivincheh et al. [30] has
combined classification and regression recurrent model. The
recurrent voting node has been then forecasted the frame's
relative location to display the completion moment. The
system has been tested on RGBD-AC dataset and has
achieved an accuracy of 89% for both complete and
incomplete sequences.
Miron et al. [31] presented temporal convolutional neural
network (Res-TCN) in assessing the correctness of a human
motion or action. For every movement type, a different model
has been trained with the explicit purpose of differentiating
between optimal and non-optimal movements. This paper
produced 71.2% results for certain actions but can also fall
into the trap of classifying an incorrect execution of an action
as a correct execution of another action used. Wang et al. [32]
presented stacked auto encoder (SAE) deep learning model as
a fundamental building blocks for ADL activities
identification. The SAE has been then trained using greedy
layer-wise. In order to fine-tune the network settings, a fully
connected layer and a classification layer have been connected
on top of the SAE. The model had achieved an accuracy of
90.43% on the publicly accessible ADL-based dataset. Xiao
et al. [33] presented a deep convolutional model and a self-
attention model-based end-to-end framework for recognizing
human action from skeletal sequences. To extract
discriminative deep features, DCM has used CNN. The SAM
algorithm has been made to learn attention weights from
variations in motion. Additionally, the attention weight for
skeleton-based action categorization learned attentive deep
characteristics. Over the SYSU-3D dataset, the efficacy of the
suggested model has attained an accuracy of 80.36 %.
In this study, we have presented a novel type of symmetry
principle based on the silhouette of a human. We have showed
that the camera's location and the distance between the camera
and the target can change. Each kind of skeleton will be
processed once a skeleton has been retrieved from each
RGBD (Red Green Blue Depth) frame. The upper half of the
body, lower half of the body, center of the body, the position
of the hands, and the variation of the hands are the four
parameters discussed in this work.
The following is how the rest of the paper is organized:
The solution framework, which includes silhouette detection,
feature extraction, optimization, and classification, is
described in Section II. The experimental results are presented
in Section III, along with a comparison to other state-of-the-
art systems. Finally, part IV brings the paper to a close.
II. SYSTEM DESIGN
The RGB and depth sequences from the RGBD-AC
(RGBD-Action-Completion-2016) [34] dataset has been used
in the proposed ADL system. The replacement operation,
scaling operation, median filter, canny edge filter, and
mapping depth image across RGB images have been
employed in the preprocessing step to extract silhouette from
RGBD images. These RGBD silhouettes have been then used
to extract symmetry principal [35-42] features. With linearly
dependent concept (LDC) [43] and long short-term memory-
recurrent neural networks (LSTM-RNN) [44], these
properties have been further improved and categorized.
Finally, the leave one subject out (LOSO) method has
been utilized to test and train the sequence of all ADL tasks.
The basic diagram of our suggested ADL framework is
shown in Fig. 1.
Fig. 1. Architecture diagram of the presented methodology for activities of
daily living (ADL) system.
A. Silhouette extraction of RGB-D images using depth
silhouettes
The pre-processing step accomplished through five phases.
1) Substitution operation: In the first step, missing depth
values are interpolated by substitution operation [45]. The
substitution operation for missing values of the depth images
search through neighboring data on both left and right pixels
[46]. Later, missing value is replaced with the larger of the
two.
2) Scaling operation: In the second step, depth image has
been linearly scaled [47] within a range of 0 to 255 as:
′
′ (1)
Where, depicts depth values of images before scaling.
While ′ is the depth values of images after scaling.
Moreover, and represents minimum and maximum
values in depth images before scaling. The
′depicts the
lower bound of image, which is 0, and depicts the
upper bound of depth image which is 255.
3) Median Filter: In the third step, noise of the rescaled
depth image has been removed using median filter [48].
4) Canny Edge Filter
In the fourth step, canny edge filter has been applied
that use multi-stage algorithm to detect and correct depth
silhouette edges [49-50].
5) Mapping over RGB image
In the last step, resulting depth silhouette is mapped to
RGB images using affine transformation to extract
silhouette from the RGB frames [51-57].
The advantage of extracting silhouette from depth
images is that border information of silhouette can be
detected even under poor light conditions [58-60]. The
results of RGB and depth silhouette is illustrated in Fig. 2.
Fig. 2. Examples of silhouette extraction and background removal in RGBD
images through five phases
B. Feature Extraction
The technique of extracting important information from
raw data is known as feature extraction [61]. Detecting
human operations including drink, open, pick, plug, switch,
and pull requires several essential places in the human body
[63]. Angles between two hands, angle between hand and
upper half of body, angle between hand and lower half of
body, angle between hand and center point of the body, and
other aspects can all be used as motion indicators [64-68].
Some of the indicators are more stable than others. By
stability, we mean that some points in the human body do not
change because of movement [69-70]. For example, we found
that the position of the head fluctuates relatively little during
cyclic motions in all our samples [71-72]. In this paper, five
features of the human body have been extracted: (1) The
thirteen key points of the body has been detected; (2) The
angle of hand and upper half of the body; (3) The angle
between centers line of the body and hand; (4) The angle
between lower half of the body and hand; (5) The angle
formation between two hands. The implementation technique
for our feature extraction approach has been described below.
1) Thirteen key points detection: Fig. 3 depicts a
comprehensive overview of the human body key point
detection model, which includes thirteen human body points
that are classified as three key skeleton fragments: lower
body, mid-point of the body, and upper body, and has been
based on link of neck, head, shoulders, wrist, hand points, and
elbow with knees, hips, foot, and ankle. Each part of the body
plays a role in the completion of a specific task.
The central torso point has been calculated using the
human silhouette's outer shape for the identification of human
body points. The point 1/4 between the foot and the knee
points has been used to determine the human ankle posture
[73]. To estimate the wrist point, we divided the distance
between the hand and elbow points by ¼ . The comprehensive
explanation of human body part detection has been seen in
Algorithm 1.
Fig. 3. Illustration of thirteen key points detection on extracted silhouette.
Algorithm1: Human silhouette key point detection
Input: Human silhouette
Output: Nineteen key points detection as neck, head, shoulders,
wrists, hands, and elbows, knees, hips, foot, ankle, and torso.
HBS = Human Body Silhouette
H = Height
W = width
L = Left
R = Right
Jh = Head
Jn = Neck
do For i = 1 to N
Ih = Get_Head_Point(Jh)
In = Get_Upper_Point(Jn)
Im = Get_Mid_Body_Point(H, W)
If = Get_Bottom_Body_Point(HBS)
Iknp = Get_Midpoint(Im, If)
Ihnp = Get_Midpoint(Ih, In)
Ielp = Get_Midpoint(Ihnp, Ih)
Iwrp = Get_Midpoint(Ihnp, Ielp)/2
Ihip = Im
Ianp = Get_Midpoint(Iknp, If)/4
While ((search (HBS) & search (L, R))!=NULL)
return 13_Body_Points (head, neck, shoulders, elbows, wrists, hands,
hips, torso, feet, knees, and ankles).
2) Angle between upper half of the body and hand: In
RGBD-AC dataset, the angle formation in some activities
have been formed between hand and upper half of the body
such as switch, open and drink. Therefore, angle forms
between upper half of the body and hand, is the most
important feature of measuring activities performed by
silhouette. It can be seen as meeting the decision requirement
for the occurrence of the event at the given time. Because
ADL activities require a quick transition from start to finish
in RGBD-AC dataset, the time required for event detection
has been set at once per fifteen adjacent frames, with a 0.5s
time interval. The angle detection process [74] has been
depicted in Fig. 4. The coordinates of the upper body point
and hand is andℎ . At
time t, the angle is expressed as;
(2)
where represents the satisfying the decision condition
of the occurrence of the event at time t.
Fig. 4. Angle formation between upper body part and hand illustration on
the silhouette.
3) Angle between center body point and hand: The angle
creation between the center body point and the hand is the
most visible aspect of silhouette in the process of ADL
activities [75] in the RGBD-AC dataset as shown in Fig. 5.
The characteristics of the degree of angle creation between
the central body point and the hand while performing
activities such as open and pull have been set at once every
twenty-five adjacent frames. The coordinates of the center
body point and hand is and ℎ
. At time t, the angle between center body point and
the hand is expressed as:
(3)
where represents the satisfying the decision condition
of the occurrence of the event at time t.
Fig. 5. Angle formation between center body point and hand illustration on
the silhouette.
4) Angle between hand and lower half of the body: Angle
formation has been generated between hand and lower part of
the body in several activities in the RGBD-AC dataset, such
as plug, switch, and pick. As a result, the most important
element of assessing actions conducted by silhouette is the
angle established between the lower half of the body and the
hand as shown in Fig. 6. In the RGBD-AC dataset, the time
necessary for event detection has been adjusted to once every
fifteen adjacent frames, with a 0.5s time gap, because ADL
activities require a rapid transition from start to conclusion.
The lower body point and hand coordinates are expressed in
(4). The coordinates of the lower body point and hand
is andℎ . At time t, the
angle is expressed as;
(4)
where represents the satisfying the decision condition
of the occurrence of the event at time t.
Fig. 6. Angle formation between hand and lower half of the body
illustration on the silhouette.
5) Formation of angle between hands: The position of
hands in each class of activities change systematically [76].
Therefore, formation of angle between hands postures can be
used as a motion indicator. The coordinates of the right and
left hands is andℎ . At
time t, the angle is expressed as;
(5)
where represents the satisfying the decision condition of
the occurrence of the event at time t. The angle formation
between hands has been shown in Fig. 7.
Fig. 7. Angle formation between hands illustration on the silhouette
C. Linearly dependent concept (LDC) optimization
In this paper, linearly dependent concept (LDC) has been
used to reduce redundant features and to improve the
accuracy of classifier [77]. There have been 793 features that
have been considered as vectors. First, the vector set of
features has been set as 1, 2, . . . , ∈ where is a vector
set of features. The has been considered as a homogeneous
linear combination system for all vectors and has been
expressed as;
(6)
where, is the element of real set and is an element of
. The Gaussian elimination approach has been used to obtain
the constant values. After implementing the suggested
technique, 365 features have been chosen from the features
vector, while 428 have been eliminated.
D. Long short-term memory-recurrent neural networks
(LSTM-RNN)
Due to the lack of memory components in traditional
neural networks, they may be unable to relate earlier
knowledge to the current task to conduct reasoning about
previous occurrences [78]. Recurrent neural networks
(RNNs) are designed to processing sequential incoming data
by allowing information to survive across loops in the
network topology. Most of these achievements may be
ascribed to the usage of LSTMs, a type of RNN capable of
learning long-term dependencies [79]. As a result, we
employed LSTM-RNNs to predict probable behaviors based
on sequential sensor data observation.
We employed a simple LSTM model with one input layer,
one hidden layer, and one output layer in our research. For
the input, hidden, and output layers, the number of neurons
was set to 10, 42, and 9, respectively. During training, the
learning rate and batch size were set at 0.005, 1600,
respectively. The results of LSTM-RNN classifiers have been
depicted in Fig. 8.
Fig. 8. Long short-term memory-recurrent neural networks (LSTM-RNN)
visual results over RGBD-AC dataset.
III. EXPERIMENTAL RESULTS
The dataset description, experimental findings,
recognition accuracy, and a comparison of our technique to
existing state-of-the-art ADL recognition systems are all
presented in this part.
A. RGBD-AC Dataset
The RGBD-AC [34] dataset contains both complete and
incomplete examples of various activities. A Microsoft
Kinect v2 was used to create 414 sequences in the collection.
Plug: plug socket, switch: turn off the light, open: open the
jar, drink: drink from a cup, pull: pull the drawer, and pick:
choosing item from the desk are the six actions captured in
the sequences. The RGBD-AC dataset requirements for each
action such that it could not be performed as; switch: subjects
were asked to act as if they had forgotten to turn off the light,
plug: subjects were given a plug that did not fit the socket,
open: a lid was glued to the jar to prevent it from being
opened, yank: a drawer that was locked could not be yanked.
A total of eight individuals (3 females and 5 men) conducted
at least four full and four partial sequences for each activity.
B. Experimental Results
With training and testing data, the suggested system was
tested using the leave one subject out (LOSO) cross
validation approach. To distinguish distinct postures and
movements, the human activity classification system was
tested using recall, precision, and F-measure. Table I depicts
the confusion matrix of the RGBD-AC dataset for six
different activities. The F-measure combines recall and
precision respectively. The classification result of 92.83%
has been achieved on F-measure over RGBD-AC (RGBD-
Action-Completion-2016), reported in Table II. Finally,
Table III shows a comparison of the proposed ADL approach
with existing framework techniques on the RGBD-AC
dataset. Overall, the findings revealed that our proposed
strategy outperformed existing state-of-the-art methods.
TABLE I. CONFUSION MATRIX RESULTS OF SIX DIFFERENT ADL
ACTIVITIES OBTAINED OVER RGBD-AC DATASET
RGBD-
AC
Dataset
Drink
Open
Pick
Plug
Switch
Pull
Drink
95.50
0
4.5
0
0
0
Open
0
92.00
3.5
2.5
1.5
0.5
Pick
2.5
0
94.50
0.5
0.5
2.0
Plug
0.5
1.5
0
93.50
4.5
0
Switch
0
0.5
2.5
5.5
90.00
1.5
Pull
0.5
0.5
3.5
3.5
0.5
91.50
Mean Accyuracy = 92.83%
TABLE II. PRECISION, RECALL, AND F-MEASURE RESULTS OBTAINED
BY PROPOSED METHOD USING LOSO VALIDATION SCHEME
RGBD-Action
Completion-2016
dataset
Precision
Recall
F-measure
Drink
0.964
0.947
0.955
Open
0.931
0.910
0.920
Pick
0.951
0.940
0.945
Plug
0.950
0.922
0.935
Switch
0.900
0.902
0.900
Pull
0.915
0.917
0.915
TABLE III. COMPARISON OF RECOGNITION ACCURACY ON RGBD-
AC DATASET WITH OTHER EXISTING FRAMEWORKS AND METHODS
Methods
Recognition
Accuracy (%)
Res-TCN [31]
71.20
CNN+LSTM [28]
80.12
Deep Convolutional Model and a
Self-Attention Model [33]
80.36
Dezert–Smarandache theory [29]
85.40
Classification and regression
recurrent model [30]
89.00
Stacked auto encoder deep learning
model [32]
90.43
Proposed approach
92.83
IV. CONCLUSION
Using RGBD silhouettes, we suggested an effective
symmetry principle-based feature extraction methodology for
ADL. The method we have suggested for extracting features
from RGBD silhouettes is a combination of points and body
form. On the publicly available RGBD-AC dataset, we
trained and tested an ADL system uses LOSO and have
achieved a mean recognition rate of 92.83%. Many
applications can benefit from the suggested feature extraction
method, including smart homes, autonomous video
monitoring, and health care. We intend to construct our own
ADL dataset in future for an online continuous environment
that will make it more useful for ADL applications. We will
also employ these characteristics for exceedingly compound
situations.
REFERENCES
[1] A. Jalal, S. Lee, J. Kim, and T. Kim, “Human activity recognition via
the features of labeled depth body parts,” in Proceedings SHHT, pp.
246-249, 2012.
[2] A. Jalal, J. T. Kim, and T.-S Kim, “Development of a life logging
system via depth imaging-based human activity recognition for smart
homes,” in Proc. of the ISSHB, pp. 91-95, 2012.
[3] A. Jalal, J. T. Kim, and T.-S. Kim, “Human activity recognition using
the labeled depth body parts information of depth silhouettes,” in Proc.
of the ISSHB, pp. 1-8, 2012.
[4] A. Jalal, N. Sharif, J. T. Kim, and T.-S. Kim, “Human activity
recognition via recognized body parts of human depth silhouettes for
residents monitoring services at smart homes,” Indoor and Built
Environment, vol. 22, pp. 271-279, 2013.
[5] A. Jalal, Y. Kim, and D. Kim, “Ridge body parts features for human
pose estimation and recognition from RGB-D video data,” in Proc. of
the ICCCNT, pp. 1-6, 2014.
[6] A. Jalal and Y. Kim, “Dense depth maps-based human pose tracking
and recognition in dynamic scenes using ridge data,” in Proc. of the
AVSS, pp. 119-124, 2014.
[7] F. Farooq, A. Jalal, and L. Zheng, “Facial expression recognition using
hybrid features and self-organizing maps,” in Proc. of the ICME, July
2017.
[8] M. Mahmood, A. Jalal, and H. A. Evans, “Facial expression
recognition in image sequences using 1D transform and Gabor wavelet
transform,” in Proc. of the ICAEM, 2018.
[9] A. Jalal, Majid A. K. Quaid, and A. S. Hasan, “Wearable sensor-based
human behavior understanding and recognition in daily life for smart
environments,” in Proc. of the FIT, 2018.
[10] M. Gochoo, I. Akhter, A. Jalal, and K. Kim, “Stochastic remote sensing
event classification over adaptive posture estimation via multifused
data and deep belief network,” Remote Sensing, 2021.
[11] A. Jalal, M. A. K. Quaid, and M. A. Sidduqi, “A triaxial acceleration-
based human motion detection for ambient smart home system,” in
Proc. of the ICAST, 2019.
[12] A. Jalal, M. Mahmood, and A. S. Hasan, “Multi-features descriptors
for human activity tracking and recognition in Indoor-outdoor
environments,” in Proc. of the ICAST, 2019.
[13] A. Jalal and M. Mahmood, “Students behavior mining in E-learning
environment using cognitive processes with information technologies,”
Education and Information Technologies, 2019.
[14] A. Jalal, A. Nadeem, and S. Bobasu, “Human body parts estimation
and detection for physical sports movements,” in proc. of C-Code,
2019.
[15] A. A. Rafique, A. Jalal, and A. Ahmed, “Scene understanding and
recognition: statistical segmented model using geometrical features
and Gaussian naïve bayes,” in proc. of ICAEM, 2019.
[16] M. Batool, A. Jalal, and K. Kim, “Sensors technologies for human
activity analysis based on SVM optimized by PSO algorithm,” IEEE
ICAEM conference, 2019.
[17] A. Ahmed, A. Jalal, and A. A. Rafique, “Salient segmentation based
object detection and recognition using hybrid genetic transform,” IEEE
ICAEM conference, 2019.
[18] A. Jalal, M. A. K. Quaid, and K. Kim, “A wrist worn acceleration based
human motion analysis and classification for ambient smart home
system,” JEET, 2019.
[19] K. Kim, A. Jalal and M. Mahmood, “Vision-based human activity
recognition system using depth silhouettes: A Smart home system for
monitoring the residents,” JEET, 2019.
[20] A. Ahmed, A. Jalal, and K. Kim, “Region and decision tree-based
segmentations for multi-objects detection and classification in outdoor
scenes,” in proc. of FIT, 2019.
[21] A. A. Rafique, A. Jalal, and K. Kim, “Statistical multi-objects
segmentation for indoor/outdoor scene detection and classification via
depth images,” in proc. of IAST, 2020.
[22] A. Ahmed, A. Jalal, and K. Kim, “RGB-D images for object
segmentation, localization and recognition in indoor scenes using
feature descriptor and Hough voting,” in proc. of IAST, 2020.
[23] M. Mahmood, Ahmad Jalal, and K. Kim, “WHITE STAG model: Wise
human interaction tracking and estimation (WHITE) using spatio-
temporal and angular-geometric (STAG) descriptors,” Multimedia
Tools and Applications, 2020.
[24] M. Quaid and Ahmad Jalal, “Wearable sensors based human
behavioral pattern recognition using statistical features and reweighted
genetic algorithm,” Multimedia Tools and Applications, 2019.
[25] A. Nadeem, A. Jalal, and K. Kim, “Human actions tracking and
recognition based on body parts detection via Artificial neural network,”
in proc. of ICACS, 2020.
[26] S. Badar, A. Jalal, and M. Batool, “Wearable sensors for activity
analysis using SMO-based random forest over smart home and sports
datasets,” IEEE ICACS, 2020.
[27] S. Amna, A. Jalal, and K. Kim, “An accurate facial expression detector
using multi-landmarks selection and local transform features,” IEEE
ICACS, 2020.
[28] A. Cartas, P. Radeva and M. Dimiccoli, “Activities of daily living
monitoring via a wearable camera: Toward real-world applications,” in
IEEE Access, vol. 8, pp. 77344-77363, 2020.
[29] H. Yu, W. Jia, Z. Li, et al., “A multisource fusion framework driven by
user-defined knowledge for egocentric activity recognition,”
EURASIP Journal on Advances in Signal Processing, vol. 14, 2019.
[30] F. Heidarivincheh, M. Mirmehdi, and D. Damen,”Action completion:
A temporal model for moment detection,” arXiv, 2018.
[31] A. Miron, and C. Grosan, “Classifying action correctness in physical
rehabilitation exercises,” arXiv, 2021.
[32] A. Wang , S. Zhao, C. Zheng , J. Yang , G. Chen, C. Y. Chang, et al.,
“Activities of daily living recognition with binary environment sensors
using deep learning: A comparative study,” IEEE Sensors Journal, vol.
21, no. 4, 2021.
[33] R. Xiao, Y. Hou, Z. Guo, C. Li, P. Wang, W. Li, et al., “Self-attention
guided deep features for action recognition,” in Proceedings of the
2019 IEEE International Conference on Multimedia and Expo (ICME),
pp. 1060–1065, 2019.
[34] F. Heidarivincheh, M. Mirmehdi, and D. Damen, “Beyond action
recognition: Action completion in RGB-D Data,” British Machine
Vision Conference (BMVC), 2016.
[35] S. Badar, A. Jalal, and K. Kim, “Wearable inertial sensors for daily
activity analysis based on Adam optimization and the maximum
entropy Markov model,” Entropy, vol. 22, no. 5, pp. 1-19, 2020.
[36] A. Jalal, N. Khalid, and K. Kim, “Automatic recognition of human
interaction via hybrid descriptors and maximum entropy markov model
using depth sensors,” Entropy, 2020.
[37] A. Ahmed, A. Jalal, and K. Kim, “A novel statistical method for scene
classification based on multi-object categorization and logistic
regression,” Sensors, 2020.
[38] M. Batool, A. Jalal, and K. Kim, “Telemonitoring of daily activity
using accelerometer and gyroscope in smart home environments,”
JEET, 2020.
[39] A. Jalal, M. A. Quaid, S. B. Tahir, and K. Kim, “A study of
accelerometer and gyroscope measurements in physical life-log
activities detection systems,” Sensors, 2020.
[40] A. Rafique, A. Jalal, and K. Kim, “Automated sustainable multi-object
segmentation and recognition via modified sampling consensus and
Kernel sliding perceptron,” Symmetry, 2020.
[41] A. Jalal, M. Batool, and K. Kim, “Stochastic recognition of physical
activity and healthcare using tri-axial inertial wearable sensors,”
Applied Sciences, 2020.
[42] A. Jalal, I. Akhtar, and K. Kim, “Human posture estimation and
sustainable events classification via Pseudo-2D stick model and K-ary
tree hashing,” Sustainability, 2020.
[43] A. Jalal, M. Batool, and K. Kim, “Sustainable wearable system: Human
behavior modeling for life-logging activities using K-Ary tree hashing
classifier,” Sustainability, 2020.
[44] S. B. Tahir, A. Jalal, and K. Kim, “IMU sensor based automatic-
features descriptor for healthcare patient’s daily life-log recognition,”
in proc. of IAST, 2021.
[45] I. Akhter, A. Jalal, and K. Kim, “Pose estimation and detection for
event recognition using sense-aware features and Adaboost classifier,”
IEEE IBCAST, 2021.
[46] A. Jalal, M. Batool and B. Tahir, “Markerless sensors for physical
health monitoring system using ECG and GMM feature extraction,”
IEEE IBCAST, 2021.
[47] J. Madiha, A. Jalal, and K. Kim, “Wearable sensors based exertion
recognition using statistical features and random forest for physical
healthcare monitoring,” IEEE ICAST, 2021.
[48] K. Nida, M. Gochoo, A. Jalal, and K. Kim, “Modeling two-person
segmentation and locomotion for stereoscopic action identification: A
sustainable video surveillance system,” Sustainability, 2021.
[49] P. Mahwish, A. Jalal, and K. Kim, “Hybrid algorithm for multi people
counting and tracking for smart surveillance,” IEEE IBCAST, 2021.
[50] A. Ahmed, A. Jalal, and K. Kim, “Multi‑objects detection and
segmentation for scene understanding based on texton forest and kernel
sliding perceptron,” JEET, 2020.
[51] J. Madiha, M. Gochoo, A. Jalal, and K. Kim, “HF-SPHR: Hybrid
features for sustainable physical healthcare pattern recognition using
deep belief networks,” Sustainability, 2021.
[52] S. Amna, A. Jalal, M. Gochoo and K. Kim, “Robust active shape model
via hierarchical feature extraction with SFS-optimized convolution
neural network for invariant human age classification,” Electronics, vol.
10, no. 4, 2021.
[53] A. Jalal, A. Ahmed, A. Rafique and K. Kim “Scene semantic
recognition based on modified fuzzy c-mean and maximum entropy
using object-to-object relations,” IEEE Access, 2021.
[54] A. Jalal, M. Mahmood, and M. A. Sidduqi, “Robust spatio-temporal
features for human interaction recognition via artificial neural network,”
IEEE Conference on FIT, 2018.
[55] A. Hira, A. Jalal, M. Gochoo, and K. Kim, “Hand gesture recognition
based on auto‐landmark localization and reweighted genetic
algorithm for healthcare muscle activities,” Sustainability, 2021.
[56] N. Amir, A. Jalal, and K. Kim, “Automatic human posture estimation
for sport activity recognition with robust body parts detection and
entropy markov model,” Multimedia Tools and Applications, 2021.
[57] I. Akhter, A. Jalal, and K. Kim, “Adaptive pose estimation for gait
event detection using context‑aware model and hierarchical
optimization,” JEET, 2021.
[58] M. Gochoo, S. Badar, A. Jalal, and K. Kim, “Monitoring real-time
personal locomotion behaviors over smart indoor-outdoor
environments via body-worn sensors,” IEEE Access, 2021.
[59] P. Mahwish, G. Yazeed, M. Gochoo, A. Jalal, S. Kamal et al., “A smart
surveillance system for people counting and tracking using particle
flow and modified SOM,” Sustainability, 2021.
[60] M. Gochoo, S. R. Amna, G. Yazeed, A. Jalal, S. Kamal et al., “A
systematic deep learning based overhead tracking and counting system
using RGB-D remote cameras,” Applied Sciences, 2021.
[61] Y. Ghadi, I. Akhter, M. Alarfaj, A. Jalal, and K. Kim, “Syntactic
model-based human body 3D reconstruction and event classification
via association based features mining and deep learning,” PeerJ
Computer Science, 2021.
[62] W. Manahil, A. Jalal, M. Alarfaj, Y. Ghadi, S. Tamara, S. Kamal, and
D. Kim, “An LSTM-based approach for understanding human
interactions using hybrid feature descriptors over depth sensors,” IEEE
Access, 2022.
[63] Y. Ghadi, J. Madiha, M. Alarfaj, S. Tamara, A. Suliman, A. Jalal, S.
Kamal, and D. Kim, “MS-DLD: multi-sensors based daily locomotion
detection via kinematic-static energy and body-specific HMMs,” IEEE
Access, 2022.
[64] A. Usman, Y. Ghadi, S. Tamara, A. Suliman, A. Jalal, and J. Park,
“Smartphone sensor-based human locomotion surveillance system
using multilayer perceptron,” Applied Sciences, 2022.
[65] A. Ayesha, Y. Ghadi, M. Alarfaj, A. Jalal, S. Kamal, and D. Kim,
“Human pose estimation and object interaction for sports behaviour,”
CMC, 2022.
[66] B. Mouazma, Y. Ghadi, A. Suliman, S. Tamara, A. Jalal, and J. Park,
“Self-care assessment for daily living using machine learning
mechanism,” CMC, 2022.
[67] Y. Ghadi, W. Manahil, S. Tamara, A. Suliman, A. Jalal, and J. Park,
“Automated parts-based model for recognizing human-object
interactions from aerial imagery with fully convolutional network,”
Remote Sensing, 2022.
[68] Y. Ghadi, R. Adnan, S. Tamara, A. Suliman, A. Jalal, and J. Park,
“Robust object categorization and Scene classification over remote
sensing images via features fusion and fully convolutional network,”
Remote Sensing, 2022.
[69] S. Tamara, J. Madiha, M. Gochoo, A. Suliman, Y. Ghadi, A. Jalal, and
J. Park, “Student’s health exercise recognition tool for E-learning
education,” IASC, 2022.
[70] Y. Ghadi, W. Manahil, M. Gochoo, A. Suliman, S. Chelloug, A. Jalal,
and J. Park, “A graph-based approach to recognizing complex human
object interactions in sequential data,” Applied Sciences, 2022.
[71] A. Alam, S. Abduallah, A. Israr, A. Suliman, Y. Ghadi, S. Tamara, and
A. Jalal, “Object detection learning for intelligent self automated
vehicles,” IASC, 2022.
[72] Y. Ghadi, A. Israr, A. Suliman, S. Tamara, A. Jalal, and J. Park,
“Multiple events detection using context-intelligence features,” IASC,
2022.
[73] H. Sadaf, Y. Ghadi, M. Alarfaj, S. Tamara, A. Jalal, S. Kamal, and D.
Kim, “Sensors-Based Ambient Assistant Living via E-Monitoring
Technology,” CMC, 2022.
[74] M. Jamil, A. Shaoor, A. Suliman, Y. Ghadi, S. Tamara, A. Jalal, and J.
Park, “Intelligent Sign Language Recognition System for E-learning
Context,” CMC, 2022.
[75] S. Tamara, A. Israr, A. Suliman, Y. Ghadi, A. Jalal, and J. Park,
“Pedestrian Physical Education Training over Visualization Tool,”
CMC, 2022.
[76] M. Alarfaj, W. Manahil, Y. Ghadi, S. Tamara, A. Suliman, A. Jalal,
and J. Park, “An Intelligent Framework for Recognizing Social
Human-Object Interactions,” CMC, 2022.
[77] Y. Ghadi, B. Mouazma, M. Gochoo, A. Suliman, S. Tamara, A. Jalal,
and J. Park, “Improving the Ambient Intelligence Living using Deep
Learning Classifier,” CMC, 2022.
[78] R. Hammad, M. Muneed, A. Suliman, Y. Ghadi, A. Jalal, and J. Park,
“Home Automation-based Health Assessment along Gesture
Recognition via Inertial Sensors,” CMC, 2022.
[79] M. Mushhood, S. Shizza, Y. Ghadi, A. Suliman, A. Jalal, and J. Park,
“Body Worn Sensors for Health Gaming and e-learning in Virtual
Reality,” CMC, 2022.
