An Analytical Comparison of Deep Learning Frameworks for 2D Image-Based Hatha-Yoga Pose Identification

The comprehensive practice of yoga seeks to improve one's mental, bodily, and spiritual well-being. Among the many forms of yoga, hatha yoga is a conventional style that uses physical postures and breath exercises to balance the body and mind. We can reap its greatest health benefits by adopting the ideal postures and adhering to the recommended techniques and sequencing. However, adopting improper postures while practising yoga can result in a number of health issues such as short-term chronic issues or acute muscle discomfort. Therefore, there is a need for scientific evaluation of yoga posture recognition in order to help people practise yoga effectively. To support self-learning, we'll provide a model for classifying poses using posture detection in this paper. Here we have taken a 2D image dataset of 20 famous hathayoga poses and we comparatively evaluated the performance of several cutting-edge deep learning architectures and DenseNet201 yields the highest accuracy of 90.4%.