Computer Vision-Based Yoga Pose Recognition Using Hybrid Deep Learning Model
Hukam Chand Saini, Renu Bagoria, Praveen Arora · Advances in engineering research/Advances in Engineering Research · 2024
Human action recognition is a critical aspect of computer vision research and has various practical applications.In this paper, we focus on a specific type of action recognition, yoga pose recognition, and propose a computer vision-based model using deep learning.Our proposed model is a hybrid of Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) and is designed to aid individuals in their self-practice of yoga.Mediapipe pose estimation is used to extract body keypoint as a feature of yoga poses.The Convolutional Neural Network (CNN) layer is utilized for extracting features from the keypoints, and the Gated Recurrent Unit (GRU) layer follows it to understand the sequence of frames for making predictions.The model is trained on video dataset of yoga poses carried out by various individuals.Model performance is evaluated based on its ability to accurately recognize the poses.The integration of Mediapipe and the combination of CNN and GRU offers a unique approach to yoga pose recognition and provides new insights into the field of human action recognition.