Year: 2026
Type: article
Abstract: The rise of digital platforms for physical activity in the post-pandemic period has driven the demand for accessible and automated solutions to support and correct exercise performance remotely. Recent advances in computer vision have enabled the development of systems that provide real-time feedback using only conventional cameras, eliminating the need for specialized sensors. Using the Design Science Research (DSR) methodology, a web-based solution for real-time exercise assessment powered by machine learning was developed as an artifact. This work covers the entire development pipeline, including data collection and labeling, training of machine learning models, conversion and optimization for inference in the browser environment, and integration into a fully functional and responsive prototype web application. Furthermore, a modular software design is proposed to support scalability and extensibility, allowing the seamless inclusion of new validation strategies. Results show that all trained models achieved accuracies above 90% and executed efficiently in the browser, with inference times below 5 ms across different devices.
Source: Journal on Interactive Systems
Authors: Rafael Santos, Andrei Sant’anna, Luan Machado, Lucas Santos, Márcio Soussa
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7135179765