Yoga Pose Estimation with Machine Learning

Yoga pose estimation involves the use of computer vision algorithms to automatically identify and track yoga poses from images or videos. This study focuses on improving the accuracy and performance of pose estimation systems through the application of OpenPose keypoint detection, SMOTE, and LightGBM classification. OpenPose is utilized for keypoint detection, enabling the identification of specific points on the body and resulting in more precise pose estimation. To address class imbalance issues, SMOTE is employed to ensure a balanced representation of poses by oversampling minority classes. Additionally, LightGBM classification is utilized to enhance model performance, benefiting from its ability to handle large datasets, faster training speed, and high accuracy. The research utilizes two datasets: the Yoga Pose Image Classification dataset and a self-collected dataset, consisting of 5994 and 5431 images, respectively. The proposed method achieved an accuracy of 56.18% on the Yoga Pose Image Classification dataset with 107 classes and 71.47% on the self-collected dataset with 50 classes, outperforming existing models.