Pose Estimate Based Yoga Instructor

Abstract

Wanting a private trainer to assist track our fitness goals, we have a tendency to patterned we have a tendency to might build our own. The goal was to make associate degree application that might track however we have a tendency to were physical exercise and commenced with Yoga as a straightforward context. we have a tendency to dubbed our 1st iteration of this application as Pose Estimation based yoga trainer to evaluate our construct, we start by gathering pictures of Yoga poses with a picture explore for terms like: ‘yoga tree pose’, ‘yoga triangle pose’, etc. we have a tendency to selected yoga as a result of the movements square measure comparatively static compared to different athletic maneuvers, this makes the constraints on frame rate of abstract thought less tightened. we will quickly separate extraneous photos and maybe refine our queries to make a corpus of one or two thousand yoga create pictures. Pose estimation gets USA half means. to comprehend Pose Estimation based yoga trainer, we’d like to feature one thing new. we’d like an operate that takes USA from cause estimates to yoga position categories. With up to fourteen body key points, every of our couple thou- sand pictures may be drawn as a vector in a very 28-dimensional real linear area. By convention, we’ll take the x and y indices of the mode for every key purpose slice of our cause estimation model belief map. In different words, the cause estimation model can output a tensor formed like (1, 96, 96, 14) wherever every slice on the ultimate axis corresponds to a 96×96 belief map for the situation of a selected body key purpose. Taking the max of every slice, we discover the foremost doubtless index wherever that slice’s key point is positioned relative to the framing of the input im- age. Our 1st plan here was to concatenate the create vectors from a pair of or three ordered time steps and take a look at to coach the tree to acknowledge a motion. to stay things easy, we have a tendency to begin by framing a need to differentiate be- tween standing, squatting, and forward bends (dead-lift). These classes were chosen to check each static and dynamic maneuvers. Squats and Dead-lifts continue to exist similar planes-of-motion and area unit leg-dominant moves though’ activating opposing muscle teams.