Real-Time Human Pose Estimation Using Media-Pipe an Artificial Intelligence Applications in Health and Fitness
Ketan Totlani, Shiva Shashank Dhavala, S Sandeep Kumar Vijayarao, Yagnesh Challagundla, Bappadittya Roy, Eugenia R. Zhuo · 2024
Significant progress has been achieved in the fields of Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) in recent years. Nevertheless, a significant gap continues to exist in the effective surveillance of human actions, activities, and the recognition of human movements or poses. Machines and computers are fundamentally incapable of comprehending human activities without user intervention. Consequently, we now depend on technology to record and archive human actions through live broadcast data, videos, or images. The field of pose estimation, which is a critical area of research with applications in fields such as Augmented Reality, Animation, Gaming, Healthcare, and Sports, is expanding due to the rapid advancement of computer vision technology. This project implements innovative methods to train models and deter- mine poses, utilizing random inputs from internet images, videos, and live streaming. In particular, models that have been trained with artificial intelligence have been created for daily physical activities that necessitate intricate pose estimation. The initial step in the project is to utilize MediaPipe to estimate the poses of the human body. Subsequently, the angles between the body joints are calculated to create the desired models. Additionally, a web-based user interface (WEB-UI) was developed to enable access from any location. The methodology involves the extraction of data from critical body locations, the framing of models, and the utilization of the MediaPipe model due to its superior performance and features in comparison to other models. The objective of this method is to develop a web application for end-users, train AI models for daily physical activities, and establish an efficient system for single human pose estimation. This approach provides a cost-effective alternative to physical trainers and enables the monitoring of health and sports activities to be enhanced.