Uma abordagem de visão computacional para correção de postura em tempo real da técnica de agachamento

Abstract

Performing physical exercises correctly is essential to avoid injuries and seek effective results, especially for squat movements. However, a person cannot always be accompanied by a professional in this activity. With this in mind, the objective of this work is to present a computer vision approach for real-time posture correction of the squat technique. The Me Exercite web app was developed in Python using MediaPipe Pose Landmark Detection resources to detect human poses. The application integrates different back-end technologies (OpenCV, FastAPI, WebSockets and Nginx) and front-end (Flutter, BLoC and JavaScript), which facilitate its use by the end user. The work also carried out a pilot study to evaluate Me Exercite by volunteers, using the technology acceptance model, Likert scale 1-5. Ten gym students and two physical education professionals participated in the experiment. The results indicated positive acceptance for both groups, with the potential to support new exercises in the future. The evaluations highlighted the ease of use of the solution (x(a) = 4.90, x(p) = 5.00), as well as its usefulness for carrying out training (x(a) = 4.80, x(p ) = 5.00). The differentials of the proposed approach are its availability as a platform-independent application, and its reliability as a tool to support physical activities and the squat technique.