Enhancing Yoga Pose Recognition Using Large Language Models with Real-World Dataset Integration for Personalized Fitness Assistance

Correct detection of yoga poses has become a central aspect of personalized fitness to promote increased performance and enjoyment. This is further extended by the implementation of large language models, possibly complemented by real-world dataset-based information, to significantly contribute to a more accurate, flexible approach toward yoga detection. Using publicly available datasets such as the Yoga-82 dataset, the research leverages a rich amount of real-world variations in pose execution, lighting conditions, and background environments for the purpose of training robust recognition models. Incorporation of LLM facilitates nuanced interpretation and contextual understanding of pose data, allowing the system to offer tailored feedback and guidance to practitioners. In addition, the approach taken in this work emphasizes scalability and accessibility, ensuring that personalized fitness assistance can be seamlessly integrated with mobile and wearable technologies. The results of the experiments show significant improvements in the accuracy of recognition and user satisfaction, further underlining the promising use of LLM combined with real-world data in next-generation fitness applications. It paves the way for more intelligent and responsive fitness p latforms that cater to individual needs by fostering sustained engagement and healthier living.