AJAY KUMAR GARG ENGINEERING COLLEGE, GHAZIABAD

A Project Synopsis on

HUMAN ACTIVITY RECOGNITION USING S DATASET

Submitted for fulfillment of award of

Bachelor of Technology

In

Information Technology

By

Radhika Agarwal (2000270130078)

Vishal Mishra (2000270130189)

Yash Verma (2000270130194)

Yashaswi Srivastava (2000270130195)

Under the Guidance of

Dr. Veena Bharti

(2023-24)
 INDEX

Content Page No.

1. Problem Statement 3

2. Scope of the Project 4

3. Objective of the Project 6

4. Why is this Particular Topic is Chosen ? 7

5. Literature Review 8

6. Gantt Chart 10

7. Outcome of the Project 11

8. References 13
 Problem Statement

Human Activity Recognition (HAR) is presently considered a dynamic and burgeoning field in
both computer vision and human-computer interaction. Still, it remains a really complicated
assignment, due to unsolvable difficulties like:

1. Diverse Activity Recognition: Human activities can be diverse, ranging from simple
activities of daily living (e.g. running, sitting) to complex actions (e.g. cooking, dancing).
The system needs to accurately classify and differentiate among these various activities.
2. Data Collection: Gathering high-quality labeled data for training the HAR model can be
challenging. This includes collection of data from different individuals, age groups, and
cultural backgrounds to ensure the model's robustness.
3. Sensor Variability: Data collected from various sensors, such as accelerometers and
gyroscopes, can vary in quality, accuracy and precision. Handling this sensor variability
is crucial for consistent recognition.
4. Real-Time Processing: In many applications like healthcare and security, real-time
recognition is essential. The system must process data and make predictions with low
latency.
5. Resource Constraints: When implementing HAR on resource-constrained devices like
smartphones, power efficiency and computational requirements become significant
challenges.
6. Privacy and Data Security: Collecting and processing sensor data, especially in
healthcare and lifestyle tracking, raises privacy concerns. Ensuring data security and user
consent is vital.
7. Model Robustness: The HAR system must work across different environments, lighting
conditions, and user behaviors, making the model's robustness a critical factor.

Addressing the issues in the domain of Human Activity Recognition (HAR) mentioned above is
of utmost importance in order to foster the development of precise, dependable, and efficient
systems that are capable of accurately identifying and categorizing human activities across
multiple domains.
Moreover, this would lead to the enhancement of real-time processing capabilities, the ability to
handle sensor variability, and the resolution of ethical and privacy concerns.
These challenges, once surmounted, will serve to facilitate the widespread implementation and
application of HAR in fields such as healthcare, sports analytics, smart environments, and other
domains, thereby resulting in substantial benefits.
 Scope of the Project

The Human Activity Recognition system will contribute in making our society safer, healthier,
and more efficient by accurately recognizing and understanding human activities.
The future scope of Human Activity Recognition (HAR) systems is promising and is likely to
see significant advancements and broader adoption in various domains. Some of the key aspects
of the future scope of HAR are:

1. Healthcare and Aging Population: As the global population continues to age, HAR
systems will play an essential role in healthcare. They will assist in remote patient
monitoring, fall detection, medication adherence tracking, and providing valuable
insights into the daily activities of elderly individuals.

2. Preventive Healthcare: HAR can contribute to preventive healthcare by identifying

early signs of health issues through changes in activity patterns. It may help in the early
detection of conditions such as Alzheimer's disease, Parkinson's disease, and depression.

3. Wearable Technology: Wearable devices equipped with sensors for HAR will
continue to evolve. Smaller, more comfortable, and more accurate sensors will become
commonplace, allowing for continuous monitoring of activities and health parameters.

4. Mental Health: HAR can be applied to mental health monitoring by recognizing

patterns of behavior related to stress, anxiety, or mood disorders. It may provide timely
interventions and support for individuals facing mental health challenges.

5. Personalized Services: HAR will enable highly personalized services and

recommendations in various domains, including fitness, entertainment, and marketing.
Systems will adapt to users' preferences and activities to provide tailored experiences.

6. Smart Environments: Smart homes and smart cities will integrate HAR technology to
create more efficient and responsive environments. Lighting, heating, and security
systems will adapt based on occupants' activities and needs.

7. Human-Robot Collaboration: HAR is crucial for improving human-robot

collaboration, especially in fields like manufacturing, where robots need to work safely
alongside humans. HAR can help robots understand and predict human actions.
8. Sports and Performance Optimization: Athletes and sports teams will increasingly
rely on HAR for performance optimization, injury prevention, and personalized training
programs. It can provide valuable insights into biomechanics and training efficiency.

9. Security and Surveillance: HAR will continue to enhance security and surveillance
systems by automating the detection of suspicious activities and improving response
times to security threats.

10. Education and Skill Assessment: In educational settings, HAR can assess students'
engagement levels and provide feedback on their learning activities. It can also evaluate
physical skills in vocational training.

11. Data-Driven Decision-Making: HAR-generated data will be used for data-driven

decision-making in various industries. It will inform business strategies, urban planning,
and public policy.

12. Ethical and Privacy Standards: Research in HAR will focus on addressing ethical
and privacy concerns related to data collection and usage. Developing robust standards
and practices for responsible HAR systems will be essential.

13. Integration with AI and Machine Learning: Advancements in AI and machine

learning will further enhance HAR capabilities. Deep learning models and reinforcement
learning techniques will lead to more accurate and robust activity recognition.

14. Global Adoption: HAR systems will become more accessible and globally adopted,
improving healthcare and quality of life in various regions, including underserved areas.

15. Interdisciplinary Research: The future of HAR lies in interdisciplinary

collaboration, bringing together experts from fields such as computer science, healthcare,
psychology, and engineering to develop innovative solutions.
 Objective of the Project

The objective of Human Activity Recognition (HAR) using smartphone datasets is to develop
andimplement machine learning or deep learning algorithms that can automatically identify and
classify different human activities based on data collected from smartphone sensors. This field of
research and application has several important goals:

Activity Classification: The primary objective is to accurately classify various human activities,
such as walking, running, cycling, sitting, standing, or even more specific activities like climbing
stairs or descending stairs. The goal is to recognize these activities in real-time or from recorded
data.
Health and Fitness Monitoring: HAR using smartphone data can be used to monitor an
individual's physical activity and health. This information can be valuable for personal fitness
tracking, healthcare applications, and assessing the impact of physical activity on overall
well-being.
Research and Insights: HAR datasets and models can be used for research purposes to gain
insights into human behavior, activity patterns, and trends. This information can be valuable for
various fields, including psychology, sociology, and urban planning.
Sports Performance Analysis: In sports, HAR can be used to analyze the performance of athletes.
It can help in tracking their movements and activities to improve training regimens and prevent
injuries.
In summary, the primary objective of Human Activity Recognition using smartphone datasets is
to develop accurate, efficient, and practical algorithms and applications that can automatically
recognize and classify human activities based on sensor data from smartphones. These
applications have a wide range of potential uses, from personal fitness tracking to improving
healthcare and enhancing the user experience in various contexts.
Why is the Particular Topic Chosen ?
 Literature Review

● In [1] Lin, Guiping, et al. discussed in detail that The paper presents a system that uses a
smartphone and off-the-shelf WiFi router for human activity recognition with high
accuracy (97.25%), while recognising small scale motions was (97.57%) which is
particularly useful for gesture recognition. The system utilizes customized firmware and
developed software to capture WiFi channel state information (CSI) data and extract
features associated with specific human activities. The system is able to classify 20 types
of human activities with different scales, including small, medium, and big motions. The
paper evaluates the system's performance by collecting data from 60 participants over
140 hours, resulting in 36,000 data points.Three machine learning models, including
convolutional neural networks (CNNs), decision tree, and long short-term memory, are
compared, with CNN achieving the best predicting accuracy. The system's potential
applications include gesture control and sign language recognition, offering a more
ubiquitous and mobile approach to human activity recognition.

● In [2]Subasi, Abdulhamit, et al. proposed The paper presents a smartphone-based human

activity recognition (HAR) system using Bagging and Adaboost ensemble classifiers,
which can be used for intelligent and automatic monitoring of daily clinical activities,
remote life assistants, and preventive care in healthcare applications. The Adaboost
ensemble classifiers algorithm, combined with SVM, significantly improved the
performance of smartphone-based HAR, achieving an accuracy of 97.44%. Naive Bayes
showed low performance initially, but its accuracy increased to 86.72% when using
AdaBoost. The accuracy of the Random Tree classifier did not improve with the
Adaboost ensemble classifier. The results demonstrate that Adaboost performs well in
improving the accuracy of HAR, especially for elderly and disabled patients who require
continuous care. The paper concludes by providing guidelines for future work in the field
of smartphone-based activity recognition.

● In[3] Hoang, Le Uyen Thuc, et al. The paper reviews and discusses video-based
technologies for human activity recognition, focusing on core technology, human activity
recognition systems, and applications in surveillance environments, entertainment
environments, and healthcare systems. It mentions the use of different classification
algorithms for human activity recognition, such as dynamic time warping (DTW), hidden
Markov models (HMMs), discriminative models like support vector machines (SVMs),
relevance vector machines (RVMs), and artificial neural networks (ANNs) . Enhancing
the reliability of background models to handle moving or cluttered backgrounds and
inconsistent lighting conditions is an important area for future work. Continuously
improving the accuracy and reliability of human activity recognition systems in
real-world scenarios is a key direction for future work.
● In[5] Mukherjee, Debadyuti, et al. worked on an ensemble model called EnsemConvNet,
which combines three different CNN-based models (CNN-Net, Encoded-Net, and
CNN-LSTM) to extract features and predict human activities using smartphone sensors in
healthcare applications. It achieved superior performance compared to other deep
learning models, with an overall accuracy of 99.6% on benchmark datasets . The authors
propose that future research can explore the use of both motion and non-motion types of
human activities and apply transfer learning to design customized models for
sensor-based HAR.

● In[6] Mrs. G. Sangeethal , Shantha Kumar et al The paper focuses on recognizing human
activity from the body using a DNN classifier and feature analysis, with the goal of
accurately classifying a wide range of data. The project aims to connect facial
expressions to fundamental emotions and utilize body behavior analysis to assess people's
behavior& it can recognize and categorize human facial photos performing different
activities into different expression classes. The FP-Growth Algorithm, a
divide-and-conquer approach was employed for finding frequent item sets .

● In[7] The article provides a thorough review of human action recognition in intelligent
living environments, focusing on five key areas: sensor technology, multimodality,
real-time processing, interoperability, and resource-constrained processing. It highlights
the need for more research and discusses the current state of HAR in smart living. The
study also reviews publicly available datasets and proposes a smart living Temple
framework to explore the interplay between factors like sensing modalities, feature
extraction approaches, and classification algorithms. It emphasizes the importance of a
comprehensive approach to sensing technologies in smart living, focusing on strengths,
limits, and synergies. Future research should also focus on real-time processing,
interoperability, and resource-constrained processing. The study is an important resource
for developing human action recognition in smart living, offering practical insights and
guiding future research and development efforts.
Gantt Chart (TimeLine Graph)
 Outcome of the Project

The efficacious establishment of a system for Human Activity Recognition (HAR) may engender
an array of advantageous results and gains in diverse areas. Presented below are several of the
principal effects of a prosperous HAR system:

1. Enhanced Healthcare and Well-Being: Improved healthcare monitoring for patients

with chronic conditions or the elderly. This can lead to early detection of health issues
through changes in activity patterns. It will reduce hospital readmissions and healthcare
costs.

2. Improved Public Safety and Security: Enhanced security through the automated
detection of suspicious activities. It will lead to faster response times to security threats
and emergencies. Safer transportation systems with driver monitoring for drowsiness or
distraction.

3. Efficient Smart Environments: More energy efficient and responsive smart homes
and buildings. Adaptive lighting, heating, and security systems that respond to occupants
activities and needs. Enhanced convenience and comfort in smart living spaces.

4. Optimized Sports Performance: Improved sports performance through data-driven

insights into athletes' movements and techniques. Reduced risk of injuries and more
effective training programs. Enhanced coaching and sports science practices.

5. Efficient Human: Robot Collaboration: Safer and more efficient collaboration

between humans and robots in industrial and manufacturing settings. Increased
productivity and automation in sectors like logistics and assembly.
6. Mental Health Support: Early detection of mental health issues through behavior
patterns. Timely interventions and support for individuals facing stress, anxiety, or mood
disorders.

7. Data-Driven Decision-Making: Valuable data for data-driven decision-making in

various industries like business strategies, urban planning and traffic management.

8. Enhanced User Experiences: More intuitive and natural human-computer

interactions, especially in augmented reality and virtual reality applications. Improved
gaming experiences with gesture-based controls and immersive gameplay.

9. Global Adoption and Accessibility: Widespread adoption of HAR systems globally,

benefiting underserved areas and diverse populations. Improved quality of life and access
to advanced healthcare and technology.

10. Ethical and Privacy Standards: Development and adherence to ethical and privacy
standards for data collection and usage. Responsible implementation of HAR systems
that respect users' privacy and data rights.
 References

1. Lin, Guiping, et al. "Human activity recognition using smartphones with WiFi signals."
IEEE Transactions on Human-Machine Systems 53.1 (2022): 142-153.

2. Subasi, Abdulhamit, et al. "Smartphone-based human activity recognition using bagging

and boosting." Procedia Computer Science 163 (2019): 54-61.

3. Hoang, Le Uyen Thuc, et al. "A review on video-based human activity recognition."
(2013).

4. Shikha, Ms, et al. "Human activity recognition." Int J Innov Technol Explor Eng (IJITEE)
9.7 (2020).

5. Mukherjee, Debadyuti, et al. "EnsemConvNet: a deep learning approach for human

activity recognition using smartphone sensors for healthcare applications." Multimedia
Tools and Applications 79 (2020): 31663-31690.

6. Bashar, Syed K., Abdullah Al Fahim, and Ki H. Chon. "Smartphone based human
activity recognition with feature selection and dense neural network." 2020 42nd Annual
International Conference of the IEEE Engineering in Medicine & Biology Society
(EMBC). IEEE, 2020.

7. Bulbul, Erhan, Aydin Cetin, and Ibrahim Alper Dogru. "Human activity recognition using
smartphones." 2018 2nd international symposium on multidisciplinary studies and
innovative technologies (ismsit). IEEE, 2018.
8. Diraco, Giovanni, et al. "Review on Human Action Recognition in Smart Living: Sensing
Technology, Multimodality, Real-Time Processing, Interoperability, and
Resource-Constrained Processing." Sensors 23.11 (2023): 5281.