YAP_LSTM: yoga asana prediction using pose estimation and long short-term memory

Yoga is a healthy practice aimed at improving a person's physical, mental and spiritual well-being. Computational research is a great opportunity in all human societies due to the latest technological developments. However, applying artificial intelligence and machine learning techniques to a medium-sized topic like yoga is difficult. This study offers a method for accurately classifying different yoga asanas using deep learning approaches. Deep learning techniques were used in this study to build a system that can recognize yoga positions from a video or an image. Real-time videos are used to extract functions from the key points of every frame, derived from the pose estimation algorithm, and long short-term memory is applied to provide tentative predictions. A data set comprising ten yoga asanas created with a standard webcam, which is publicly available, were considered. The end-to-end deep-learning pipeline was used in this work to discover yoga from videos. In a single frame, it achieves 98 percent test accuracy, and after sampling data, it achieves 99.5 percent accuracy in 55 frames of video. When the method is applied to temporal data, it gathers knowledge from earlier frames to produce a correct and accurate result. The method was tested in real time for various groups of ten people (five men and five women) and was found to be 99.02 percent accurate. The results of the experiments provide a quality assessment of the process and a state-of-the-art comparison.