Flowing Asanas: Enhancing Yoga Posture Detection With BiLSTM Attention and MediaPipe

Yoga is an ancient practice that the combines harmony of mind, body, and soul with discipline, relaxation, and happiness. Besides its breathing and meditation exercises, a central aspect of yoga is its asanas (postures). Keeping the importance of asanas of yoga in mind, this research paper aims to describe a novel and compact posture detection system for yoga that integrates MediaPipe for real-time skeletal landmark placement and deep learning models for posture analysis and accuracy evaluations. The architecture acquires live yoga poses and hand gestures directly from a webcam through an RGB camera where MediaPipe recognizes key body points such as the knees, eyes, hands, etc., to create a skeleton representation of the pose taken by the user. Next, these skeletal structures are evaluated by deep learning algorithms against correct yoga postures that have been pre-trained and can tell how aligned or accurate the user’s postures may be. It automatically stops the timer once it identifies the wrong posture to ensure enhanced safety and efficiency. The technique helps expand the horizons of yoga to greater potential in physical training and rehabilitation while enhancing the practice itself.