Bio-mechanics of yogasanas : A study of alignment and proficiency

Abstract

BACKGROUND Yoga is a traditional Indian way of keeping the mind and body fit, through physical postures (asanas), voluntarily regulated breathing (pranayama), meditation and relaxation techniques. The recent pandemic has seen a huge surge in the number of yoga practitioners, many practicing without proper guidance which leads to unexpected injury. AIMS AND OBJECTIVES The present study was designed to develop a technology-driven pose estimation method to assess and evaluate yoga postures to understand the level of accuracy. It will assist practitioners to perform any yoga posture with the support of a machine learning algorithm. METHODS Participants Twenty practitioners in the age group of 18 to 60 years performing different postures in real time were captured and fed separately to the proposed architectures and a comparison of the estimated accuracy was done. Design The present exploratory study included a group of 20 yoga practitioners to assess the accuracy and proficiency of yoga postures. Assessments: The image of a yoga practitioner performing an asäna is captured by a camera and fed separately to the four deep learning architectures, which then estimate the pose performed by the practitioner by comparing it with the pre-trained model. An error is shown if it does not match any of the five asanas. Intervention: The five yoga poses considered for posture estimation are (a) Ardhacandräsana (Half-moon pose) (b) Täòäsana (Mountain pose) (c) Trikoëäsana (Triangular pose) (d) Vérabhadräsana (Warrior pose-II) (e) Våkñäsana (Tree pose) Results: In this work, four distinct deep learning architectures-Epipolarpose, Openpose, Posenet, and Mediapipe-were utilized to evaluate yoga postures. The results show that, despite only utilising one camera, Mediapipe outperforms the other approaches in terms of accuracy. Five yoga postures have had their poses estimated using various suggested methods. Following the model's validation, the posture correctness of 20 real-time sample photos was estimated using the model. Conclusions: The health and fitness industry can employ human pose estimation efficiently. The huge range of poses with high degrees of freedom, the occlusions caused by the body or other objects blocking limbs as viewed from the camera, and the wide range of appearances or clothes make pose assessment for fitness applications particularly difficult. The mediapipe design offers the best estimation accuracy, according to this study, which evaluates the estimation of five different postures