idea behind this yoga pose detection project using deep learning or neural network learning is that yoga popularity is increasing day by day because of its benefits. Doing yoga helps us physically, mentally as well as spiritually. Because of this many people nowadays are doing it regularly. The main idea of this project is to help the people to recognize which yoga pose they are doing with the help of this detection technique. Yoga which involves 8 rungs and limbs of it, which includes Yama, Niyama, Asana, Pranayama, Dharana, Dhyana and Samadhi. To easily help people understand which pose they are performing via images, video recording by classifying it, we are implementing this project because of this people will incline towards doing more as they will get help to identify which pose they are doing very easily.

1. Project Title
“Deep learning based Yoga pose
detection”
Group Members Name
Aniruddha Ghewade
Kalyankuamr Chintguntla
Swapnil Prasad
Omkar Patil
Guide Name
Prof. Minal Zope

2. Contents/Agenda
1. Problem statement
2. Introduction
3. Literature survey
4. Motivation
5. Goal and objective
6. Scope of project
7. Design Diagrams
8. System requirements
9. Proposed system architecture
10. Methodology used
11. Algorithm
12. Input to project
13. Dataset Used
14. Expected output
15. Applications
16. Conclusion
17. Future Scope
18. References

3. Problem statement
• The proposed device is used with convolutional neural network for
recognizing human action based on yoga pose classification using image
processing and deep learning. The objective of pose estimation for
monitoring the movement of human topics for distinct exercises

4. Introduction
• Here we are using real time detection which can use both video and
images for detection. We need to place the camera in some distance to
capture the images or videos perfectly. Here we will be having a static
home page where we can do registration if we want and then we can login
by using user name and password credentials. If we are using video for
detection the video can be a recorded video i.e. mp4 file by using
recorded video path or simply put zero in my video section to on the
camera i.e. webcam to use a real time video detection. Here when the
video is on it will detect the pose only if the background is white and give
output as the name description of the pose. Here the webcam needs to
cover the complete body for the pose detection. Here we are using CNN
(convolutional neural network) algorithm for video and image processing.

5. Literature Survey
• 1. “AI Human Pose Estimation: Yoga Pose Detection and Correction” -by Rutuja
Gajbhiye, Snehal Jarag, Pooja Gaikwad, Shweta Koparde 2. Publishing year of this
IEEE research paper - 5 May 2022 3. The Paper is about Yoga Pose Detection and
Correction which introduce us about various implementation methods for
detection of yoga pose. the dataset is collected first and after that preprocessing
is done on that data. The proposed system is implemented in python using the
OpenCV library. A dataset containing 84 yoga asanas sets in a typical yoga
posture is selected by the system using a regular webcam and made publicly
available. A novel hybrid approach based on machine learning and deep learning
classifiers. Step one contains, a support vector machine (SVM) is planned. This
classifier uses machine learning prediction to improve the performance of ML
algorithms. The second step is a convolutional neural network, which captures
the human skeleton of poses and the user's target poses, and compares the two
poses to obtain similarities. they imported libraries such as OS, Time, Keyboard,
Array, and Mediapipe. 4. The paper has contributed in our knowledge about
SVM algorithm and various python libraries like OS, Time, Keyboard, Array and
Mediapipe which helped us to study about various methods and libraries which
can give us most possible accuracy.

6. • 1. “A Proposal of Yoga Pose Assessment Method Using Pose Detection for
Self-Learning” -by Maybel Chan Thar1, Khine Zar Ne Winn1, Nobuo
Funabiki2 2. Author introduced this research paper based on the idea of
pose detection for self-learners. Proposed method is for yoga pose
assessment. Here the proposed method first detects a pose using
OpenPose i.e., a pose recognition library. Then, it calculates the difference
of the specified body angles between the pose of an instructor and that of
a user. Then, it calculates the difference of the specified body angles
between the pose of an instructor and that of a user, and suggests the
correction. The model is trained using self-learning. It works by analyzing a
dataset and looking for patterns that it can draw conclusions from. 3.
What we are doing different compared to this research paper is that
instead of OpenPose we are using feature extraction and classification
using CNN. Feature extraction will be used first to extract the features via
layers and then classify the images via classification using CNN

7. • 1. “Yoga Pose Detection and Validation” -by Ayush Gupta, Dr. Ashok Jangid
2. Publishing year of this IEEE research paper - 21 September 2021 10 3.
The Paper has introduced Yoga pose detection and validation. In this paper
OpenPose is used to for key point detection in human body in
preprocessing. OpenPose is a real-time multi-person system presented by
the Perceptual Computing Lab of Carnegie Mellon University (CMU) to
jointly detect a human body, hand, facial, and foot key points on single
images. After that Classification is done using a proposed dataset for
training the classification model. For this purpose, images of the poses
were extracted from the video of yoga instructors performing the yoga
asanas and from the google image search results. The SVM and Random
Forest algorithms are used here for training of classification model 4. This
research paper inspired us to use various algorithms for feature extraction
and image classification like OpenPose, PoseNet or CNN can be used for
feature extraction and for image classification SVM, Random Forest or
CNN can be used according to their accuracy percentage.

8. Motivation
• Motivation behind the project of Yoga pose detection is because yoga practicing
yields profound life changing benefits from improved sleep to reduction of the
stress.
• So we would love to create a yoga pose detection web application to help people
identify which yoga pose they are doing

9. Goal and Objective
 Goal:
Pose estimation is a computer vision technique to track the movements of a person or
anobject. This is usually performed by finding the location of key points for the given
objects.
 Objective:
Based on these key points we can compare various movements and postures and draw
insights. Pose estimation is actively used in the field of augmented reality, animation,
gaming, and robotics.

10. Scope of Project
• Image Recognition
• Deep Learning

11. Design Diagrams
• Class Diagram:

12. • Use case Diagram:

13. • Sequence Diagram:

14. • Activity Diagram:

15. • State Diagram:

16. System Requirement
Software:
Libraries:
• NumPy
• Sklearn
• Pandas
IDE:
– Jupyter notebook
Framework:
– Flask
Hardware:
A 64-bit Windows/Linux Machine with a support of at least:
• Quadcore Processor.
• 4 GB DDR4 RAM.
• 2GB NVIDIA GPU.

17. Proposed System Architecture

18. Methodology Used
• Dataset:
The quality of the dataset is an important metric that determines the
accuracy ofthe model. In the majority of the research done in this area, the
dataset has been self-generated. What self-generated means is that the
author of the dataset either himselfor with the help of volunteers performed
a certain Yoga Pose and then captured an image of it with the use of a
webcam, A good quality DSLR, Mobile Phone camera, etc.

19. • Normalising the Inputs:
Data normalization is an important step which ensures that each input
parameter (pixel, in this case) has a similar data distribution. This
makes convergence faster while training the network. Data
normalization is done by subtracting the mean from each pixel and then
dividing the result by the standard deviation.

20. • Data Augmentation:
Another common pre-processing technique involves augmenting the
existing data-set with perturbed versions of the existing images.
Scaling, rotations and other affine transformations are typical. This is
done to expose the neural network to a wide variety of variations. This
makes it less likely that the neural network recognizes unwanted
characteristics in the data-set.

21. • Feature Extraction:
Feature extraction in CNN - CNN is a neural network that extracts input image
features and another neural network classifies the image features. The input
image is used by the feature extraction network. The extracted feature signals
are utilized by the neural network for classification.

22. • Image classification using CNN:
• After the feature extraction using CNN we need to classify the
images or videos(i.e., collection of frames).
• CNN uses various layers to classify the images
• Steps for classification:
1. Convolution
2. Non Linearity (ReLU)
3. Pooling or Sub Sampling
4. Classification (Fully Connected Layer)

23. Algorithms
1) CNN :
A CNN is a kind of network architecture for deep learning
algorithms and is specifically used for image recognition and tasks that
involve the processing of pixel data. There are other types of neural
networks in deep learning, but for identifying and recognizing objects,
CNNs are the network architecture of choice.
• A CNN has two basic roles–
1) feature extraction
2) image classification

24. Input to project
• We are giving input to this yoga pose detection project as
1. Image- We can give input as image i.e., the image of the pose that
the user is doing
2. Recorded video- We can give input as recorded video (i.e.,
mp4) via video path
3. Real time video- We can open webcam to capture real time
video

25. Dataset Used
• Dataset for Image analysis :- The dataset used for this project is a part of
the Open Source collection and is publicly available.
1. Sample set (link:-
https://www.kaggle.com/datasets/niharika41298/yoga-poses-dataset)
2. Kaggle

26. Expected Output
• We are expecting the output of this yoga pose detection project is that we
are getting a recognized image or video of the pose of the yoga of the user
with the description of the name of that yoga pose and Accuracy.

27. Applications
• With the help of this method the users can select their favorite pose for
practice and then upload their recorded video on this image detection
platform.
• Then the user poses are sent to train model and then the minor mistakes
that were performed by mistake are called out.
• With the help of these outputs, the system advice the user to correct their
pose by telling them that what is going wrong.

28. Conclusion
• Over the past few years, a lot of research works in done on image
detection. This human pose detection is that it detects the pose of the
human based on the structure of the human body. The research work
done yet is abundant in this evolving sector. Here deep learning methods
have been used for human/yoga pose detection. Here we are using CNN
and its feature extraction technique. In feature extraction technique there
are different layers in which we shortlist key points that will be finally
resulted in a recognized image. Yoga pose detection has received a lot of
attention in the pattern recognition communities.

29. Future Scope
• The future of image detection is now in a learning phase and in process of
proving its importance. It is much like the original Industrial Revolution, it
has the potential to free the people from stressful jobs of recognizing the
pose as that can be done in more efficiently and effectively manner by
machines. Here we are using CNN (convolutional neural network) which is
new in market, trending technology and performing very well nowadays
and in a big demand because it has a very bright future.

30. References
• “AI Human Pose Estimation: Yoga Pose Detection and Correction” by
Rutuja Gajbhiye, Snehal Jarag, Pooja Gaikwad, Shweta Koparde
• “Deep Learning Based Yoga Pose Classification” by Shakti Kinger, Abhishek
Desai, Sarvarth Patil, Hrishikesh Sinalkar, Nachiket Deore
• “Implementation of Machine Learning Technique for Identification of Yoga
Poses” by Yash Agrawal, Yash Shah, Abhishek Sharma
• “richYoga: An Interactive Yoga Recognition System Based on Rich Skeletal
Joints” by Yu-Hsuan Lo, Chun-Cheih Yang, Hsuan Ho
• “Yoga Posture Recognition By Detecting Human Joint Points In Real Time
Using Microsoft Kinect” by Muhammad Usama Islam, Hasan Mahmud,
Faisal Bin Ashraf
• “INFINITY YOGA TUTOR : YOGA POSTURE DETECTION AND CORRECTION
SYSTEM” by Fazil Rishan, Binali De Silva, Sasmini Alawathugoda

31. Thank You...