Yoga Pose Estimation Using Rule-based Approach

Abstract

Due to high computing requirements and a lack of available datasets, precise pose detection in yoga is a challenging task. Since even small modifications can have negative effects, suggestions should be made precisely[1–2]. Pradhan is a rule-based technique which is used to guide people who practice yoga at the convenience of their homes. Pose estimation’s primary goal is to foretell human poses by identifying key points like elbows, knees, wrists, and so on. In this paper, we have proposed a system which uses rule-based techniques on every frame that was processed by the Mediapipe Framework. In this technique, the user can either upload images or perform a yoga posture in front of a camera which is then used to classify the posture from a set of 10 pretrained postures. The yoga poses used are Dolphin Pose, Half-Moon Pose, High Lunge Pose, Mountain Pose, Side Plank Pose, T-Pose, Tree Pose, Upward Salute Pose, Warrior-II Pose and Warrior-III Pose. When the work is put into practice, a real-time video feed from the user’s computer’s webcam is collected, and the yoga pose’s estimation is done. This research work used angle heuristics to categorize various yoga postures for pose detection, and we were able to achieve a combined accuracy of 93%.