Deep Learning-Enabled Yoga Pose Detection: Current Advances and Future Directions

Abstract

Yoga pose detection, an essential component of fitness monitoring and instruction, has garnered significant attention due to its potential to revolutionize online yoga practice. This review delves into the state-of-the-art deep learning methodologies employed for yoga pose recognition, critically examining their strengths and limitations. Drawing from a comprehensive survey of relevant datasets and evaluation metrics, we highlight research gaps and propose potential avenues for future work. Key challenges, including the need for large-scale labeled data and the complexity of spatiotemporal feature extraction, are discussed. Our findings underscore the importance of combining spatial and temporal information for accurate pose detection using existing datasets[10][1] [10a][2], achieving accuracies exceeding 80% in certain cases. This review positions itself as a valuable resource for researchers and practitioners alike, fostering further innovation in yoga pose detection technologies. We conclude by discussing the implications of our findings for future research and the computer vision community, emphasizing the significance of this work in advancing online yoga instruction and fitness monitoring.

[1] https://www.kaggle.com/datasets/nandwalritik/yoga-pose-videos-dataset

[2] https://figshare.com/articles/dataset/Yoga_Pose_Dataset/15112320