DEEP LEARNING APPLICATION

FOR HUMAN ACTIVITY

RECOGNITION SYSTEM USING
CNN ALGORITHM
Presented by:
ABDULAKEEM ABDULAZEEZ ALABI
HND/22/COM/FT/110
 ABSTRACT
Human activity recognition (HAR) systems have been developed to
protect human safety. For healthcare, the system is designed to
remember and record movement or behavior, as can be seen from
exercise products. The increase in the elderly and those who live alone
leads to the need to develop a monitoring system and the need for new
ways to solve problems. Most elderly people have various health
problems, such as memory problems, decreased ability to help
themselves, fainting, or falling easily. These problems cause the
elderly to need care or monitoring by a caregiver. In current social
conditions, most caregivers leave the house for a career or to study.
They are unable to constantly care for the elderly, and the elderly
without care remains a problem. The first step is to collect a dataset of
images containing people with and without smiles. The dataset
should be diverse enough to capture different races, genders, ages,
lighting conditions, and facial expressions.
 INTRODUCTION
Human Activity Recognition (HAR) is a significant
application of deep learning, particularly Convolutional
Neural Networks (CNNs), which have shown great promise
in extracting relevant features from time-series data such as
sensor readings or video frames. HAR involves classifying
and identifying different activities performed by
individuals, such as walking, running, sitting, or standing,
based on sensor data or visual input. Human activity
recognition (HAR) systems have been developed to protect
human safety. As an example, a system can monitor the
movements of patients in a hospital. In industries or
factories, it is often used to monitor the health of workers in
hazardous areas. For healthcare, the system is designed to
remember and record movement or behavior, as can be seen
from exercise products (Harjot, 2019).
 Statement of the Problem
Elderly population growth is bringing significant
challenges requiring more healthcare services and
facilities. Most elderly people have various health
problems, such as memory problems, decreased
ability to help themselves, fainting, or falling
easily. These problems cause the elderly to need
care or monitoring by a caregiver. In current
social conditions, most caregivers leave the house
for a career or to study. They are unable to
constantly care for the elderly, and the elderly
without care remains a problem.
 Aim and Objectives
To develop an accurate and robust human activity
recognition (HAR) system using Convolutional Neural
Networks (CNNs) for real-time identification and
classification of various human activities and the objectives
are to:
i. Gather and preprocess sensor data or images capturing
diverse human activities.
ii. Design an optimal CNN architecture, train the model,
and fine-tune hyperparameters using techniques like
data augmentation and transfer learning.
iii. Deploy the model for real-time use, continuously refine
it, and explore applications in fields such as healthcare
and security while addressing ethical considerations.
 LITERATURE REVIEW
Zameer, et al. (2019) make research on Human Activity Analysis using
Machine Learning Classification Techniques. Artificial Intelligence (AI)
models are developed to recognize the activity of the human from the
provided UCI online storehouse. The data chosen is multivariate and we
have applied various machine classification techniques Random Forest,
KNN, Neural Network, Logistic Regression, Stochastic Gradient Descent
and Naïve Bayes to analyse the human activity In them, Logistic
Regression and neural network gave good results whereas Naive Bayes
result was not good.
Jiang et al. (2019) considered learning an ideal Bayesian system classifier
is a NP- difficult issue, learning-improved innocent Bayes has pulled in
much consideration from scientists. In this paper, improved calculations
and proposed a covered up innocent Bayes (HNB). In HNB, a shrouded
parent is made for each characteristic which joins the impacts from every
other property. The exploratory outcomes demonstrate that HNB
essentially beats NB, SBC, NBTree, TAN, and AODE. In numerous
information mining applications, a precise class likelihood estimation
and positioning are likewise attractive.
 METHODOLOGY
Data Collection
The first step is to collect a dataset of images containing people
with and without smiles. The dataset should be diverse enough
to capture different races, genders, ages, lighting conditions, and
facial expressions.
Data Acquisition
The proposed system uses UCI-2020 dataset. That consists of
33478 data which include classification of 6 activities such as
walking, laying, sitting, walking, finger raising and sleeping with
the models using python program, then this system collected
data from the wrist and right hip position to find a suitable
position for use.
Data Preprocessing
Once the dataset has been collected, it needs to be preprocessed.
This includes resizing the images to a standard size, converting
them to grayscale or RGB format, and normalizing the pixel
values.
 METHODOLOGY CON’T
 Model Training
The next step is to train a convolutional neural network (CNN)
using the preprocessed dataset. The CNN should consist of
multiple convolutional layers, pooling layers, and fully connected
layers. The number of layers and their hyperparameters can be
adjusted to achieve the best performance.
 Model Evaluation
After the CNN has been trained, it needs to be evaluated on a
separate test set. This will give an estimate of the model's accuracy
and generalization ability.
 Deployment
Once the model has been trained and evaluated, it can be
deployed in a real-world application. This may involve integrating
the model into an existing software system or developing a new
application from scratch.
RESULT/DISCUSSION

Figure 1: Activity Recognition

 Conclusion
Human activity analysis is a popular activity in the growing industry and
this research work applied different machine learning algorithm, neural
network gave good results whereas Naive Bayes result was not good. The
implementation of Neural Network on Python gave better results than the
one provided in the Orange tool. The limitations of this work is though the
efficiency of neural network is good, the model is dynamic. The inability of
getting trained with real time data will force us to train the model every
time new data comes. This research worked on the application system for
tracking elderly information in many areas because caring for the elderly
who are healthy or strong is still necessary to track personal information or
welfare that should be received. The elderly who have an underlying
disease or are at risk of daily life accidents should have a system that
supports requesting emergency help. Furthermore, falling or fainting may
cause the elderly to lose consciousness or even to die, which our system will
automatically detect and alert the caregivers for immediate help. Besides,
each elderly may have different movement characteristics with each
activity.
 References
 Charmi, H., Jatna, R. & Nishant (2019) Human Activity
Recognition: A Survey International Workshop on Recent
advances on Internet of Things: Technology and Application
Approaches (IoT-T&A 2019) August 19-21, 2019, Halifax,
Canada Procedia Computer Science 155 (2019) 698–703
 ChinAnn, V. & LauBee, T. (2019) Human Activity Recognition:
A Review IEEE International Conference on Control System,
Computing and Engineering, 28 - 30 November 2019, Penang,
Malaysia, :
https://www.researchgate.net/publication/282298953
 Harjot, T., (2019) Human Activity Recognition using Machine
Learning Techniques, Manipal institute of Technology,manipal
[email protected] Dataset. UCI Machine Learning
Repository.
http://archive.ics.uci.edu/ml/datasets/Human+Activity+Rec
og nition+Using+Smartphones