Implementación de un prototipo de estimación de pose humana en movimientos fitness

Abstract

Abstract: This document compiles research on key aspects required for the development of a prototype for human pose detection in fitness movements, leveraging computer vision techniques and neural networks such as YOLOv8 and YOLOv11 in their pose variants. The experiments were conducted on videos of three exercises across different locations, lighting conditions, and distances, with the objective of evaluating the effectiveness of each model. The developed prototype employs real-time image processing algorithms using the OpenCv, Ultraytics, Python, and Numpy libraries to classify and assess postures by identifying human body keypoints during the execution of physical exercises. In addition, Streamlit is used to develop a lightweight and intuitive interface that ensures accessibility for users. The primary objective is to provide users with accurate feedback on their movements, contributing to improved technique and injury prevention, while promoting technological advancement, health, and well being within the fitness sector.