Prediction of Yoga Pose from YouTube Dataset using Skeleton Feature Extraction Based ISDL Model

The ancient Indian discipline of yoga has spread over the world as a means of improving one's physical, mental, and spiritual health. Computational probing has a huge potential in all areas of society, thanks to the rapid development of related technologies. It has been shown that pose detection methods may be utilised to both recognize the postures and help people do yoga with greater precision. Due to the difficulty of detecting posture in real time and the limited availability of datasets, posture recognition is a difficult task. This study use a deep learning model to categorise yoga postures according to their difficulty. This research suggests a methodology for automatic detection and classification based on deep learning. The steps of the proposed technique are called preprocessing, feature extraction, and classification. At first, a preprocessing stage is executed to remove any excess noise from the edges. The next step is an equalisation based on the histogram to isolate the usable parts of the image. Skeleton-based models are used to extract features; an improved synergic deep learning (ISDL) model is then used to categorise input photographs into different postures; and last, the Archimedes optimisation algorithm (AOA) is used to further refine the model. The provided SDL model is validated using data extracted from YouTube videos. Experiments showed that the projected ISDL model outperformed the state-of-the-art models in accuracy.