Design and Development of Robust Yoga Pose Classifier using Embedded Machine Learning

This paper proposes a novel system leveraging Edge Impulse to integrate Tiny Machine Learning device nano BLE into the domain of yoga pose recognition. Utilizing a diverse dataset of yoga pose images, our approach focuses on five key poses, employing digital signal processing pipelines and neural network architectures for model optimization. While achieving a notable accuracy of 88.44%, challenges such as pose variability and data quality lead to uncertainties in classification. Through analysis and experimentation, strategies are identified for mitigating uncertainty, emphasizing the importance of addressing these factors to enhance the reliability of pose classification systems. This research contributes to the field of embedded machine learning and promotes health and well-being through yoga practice.