Page 363 - Kaleidoscope Academic Conference Proceedings 2024

Innovation and Digital Transformation for a Sustainable World

Figure 10– Flow Chart of Application

13. RESULTS & CONCLUSION

As it can be seen in the screenshots below, at first our
model is detecting all the 17 landmarks on the body and
draw connections between them in white colour. Further, if
our current posture reaches the accuracy of threefold above
70% probability then, the colour turns to the green and also
it counts the time for holding the posture correctly and
notifies us by playing sound.

Figure 12–Transfer learning models with accuracy

Possible benefits of developing AI model for yoga posture
correction are
1. Real-time Feedback: With an AI model,
practitioners can receive immediate feedback on
their posture as they perform yoga poses. This real-
time feedback allows them to make adjustments
and corrections on the spot, leading to more
effective and safer practice sessions.
2. Personalized Feedback: While attending a yoga
class, it can be challenging for instructors to
Figure 11– Screenshot of Application provide individualized feedback to each student
due to time constraints and class size. AI model
We have exploited these different transfer learning models could offer personalized feedback tailored to each
(TL-VGG16, TLMobileNetV2, TL-InceptionV3, and TL- practitioner's unique needs and abilities.
Resnet50, TLInception-ResnetV2, TLEfficientNetB0, TL-
MoveNet Thunder) as shown in figure12. As we can see the 3. Continuous Improvement: AI models can be
accuracies all other models are lower than Movenet trained on large datasets of yoga poses, allowing
Thunder for classification task. Hence we can conclude that them to learn from a wide range of examples and
the MoveNet Thunder is optimal model for the yoga continuously improve over time. This means that
correcting system, based on evaluation metrics. As a result, the accuracy and effectiveness of the posture
the TL-MoveNet model was selected as the optimal model, correction AI can increase as more data is
showing validation accuracy of 87%, precision 0.87, recall collected and the model is refined.
of 0.86, and validation loss of 0.4958. 4. Supplementary Learning Tool: Even for
individuals attending yoga classes with instructors,
AI model can serve as a supplementary learning
tool, providing additional support and guidance
outside of class hours.
5. Consistency: Human instructors may vary in their
teaching styles and levels of expertise, leading to
inconsistencies in the feedback provided to
students. AI model can provide consistent and
objective feedback, ensuring that practitioners
receive accurate guidance regardless of who is
teaching them.
6. Accessibility: Not everyone has access to a yoga
instructor or classes, especially in remote or

– 319 –