Smart Yoga – A Study of Different Standing Yogasanas With AI-Based Technique

Abstract

Human activity recognition is well managed by human pose estimation. Hunan body parts can be detected by two ways. First is human pose estimation that detects the each key point of the human body. And second is human activity recognition that takes a series of inputs to train the model and test it with new input and give the accuracy. So according to human pose estimation history, human activity recognition estimates the pose first then judges the activity of the human body. Human pose estimation is useful in Human activity recognition. And Human activity recognition has been done with two methods. The first is with the pose key points and the second is with the 3dcnn model. Most of the work in human activity recognition assumes a figure centric scene where the actor is free to perform anything. The system is proficient to classify the activity with low error and high accuracy. It is a challenging task due to background problems, changes in scale, lightning, and frame resolution. Some actions are impulsive as well as under habit so might not be accurate as desired [1]. Human pose estimation is one of the most essential applications in computer vision. Human pose estimation helps in AI technology. Human pose estimation typically follows the assumption of human body parts. In this study, we have applied a system for performing smart yoga standing postures. Tadasana, Vrikshashana, Virabhadrasana, Utkatasana, Hasta padangusthasana are studied to recognize all the body movements. The system has recognized every bend of hands and legs, and suggested the correction to be made. The system is successfully running on Jetson tegra-x2 computing device.