The human pose estimation proves useful for various tasks such as healthcare, sports analysis and human computer interaction. The yoga practitioners have gained wide momentum in today’s era due to increasing health awareness. With the rising popularity of yoga and the increasing demand for technology-assisted learning platforms, there is an increasing need for accurate and effective methods for recognizing yoga poses. Yoga pose estimation and classification plays a crucial role in automated yoga training systems, enhancing the accessibility and effectiveness of yoga practice. Accurate recognition of yoga poses is essential for safe and effective yoga practice. Incorrect posture can result in serious injury to the body, emphasizing the critical need for precise pose detection and classification. This concern motivated our research to explore automated systems for yoga pose recognition. The paper aims to facilitate accurate identification of yoga poses, thereby enhancing accessibility to yoga instruction and minimizing the risk of injury associated with improper form. This research paper focuses on the comparison of two models namely, Ultralytics and MoveNet for detecting the keypoints in yoga poses. The keypoints are subsequently utilized for classification into five different yoga poses: downdog, goddess, plank, tree, and warrior2. Evaluation shows MoveNet achieving a superior accuracy of 93% compared to Ultralytics 88%. Precision, recall, and F1 scores are analyzed through confusion matrices for a performance analysis. This study advances automated yoga pose recognition, providing insights into the capabilities and limitations of current deep learning approaches.