Review Paper on Yoga Pose Detection using Machine Learning

Abstract

Estimating human stance is a difficult challenge in the world of computer vision. It is concerned with the portrayal of human joints in skeletal form in a photograph or film. Image scale and resolution, illumination variation, backdrop clutter, garment variations, surroundings, and human interaction with the environment all play a role in automatically determining a person's pose in an image. Exercise and fitness are one use of pose estimation that has piqued the interest of many academics. Yoga is an ancient form of exercise with complex postures that originated in India but is today known around the world for its numerous spiritual, physical, and mental benefits. The problem with yoga, like any other exercise, is that it must be done correctly, as any incorrect posture during a yoga session can be detrimental, if not dangerous. As a result, a teacher must monitor the session and help the client improve their posture. Because not everyone can afford a yoga instructor, an AI-based application might be used to identify yoga poses and provide personalised feedback to help people improve their form. Deep learning has benefited human posing estimate significantly in recent years, with considerable gains in performance. Deep learning methods give a more straightforward approach of mapping the structure than manually dealing with the connections between structures. Deep learning was used to identify pull ups, swiss ball hamstring curls, push ups, cycling, and walking. However, using this approach for yoga poses is a relatively novel application. The purpose of this study is to understand more about the various approaches to yoga posture classification and to obtain insight into the following: What exactly is a posed estimation? So, exactly what is deep learning? Deep learning can be used to classify yoga poses in real time, but how? This study makes use of references from conference proceedings, published papers, technical reports, and journals. Figure 1 depicts graphically the concerns discussed in this work. The first section of the project discusses the history and relevance of yoga. The second portion goes deeper into posture estimation and the many types of pose estimation methods, as well as discriminative methods.