Um método para monitoramento e geração de feedbacks em atividades físicas repetitivas baseado em Máquinas de Boltzmann Restritas

Abstract

The practice of physical activities, often carried out in environments such as gyms and physiotherapy sessions, requires precise execution of movements to ensure effective results and prevent injuries. Currently, approaches that combine wearable technologies and Artificial Intelligence (AI) are employed to identify the correct execution of movements. However, these approaches have limitations as they are tied to pre-programmed physical activities and do not provide specific guidance to correct movements. This thesis proposes a disruptive approach to generate execution-time models capable of offering adjustment suggestions to users, aiming for the correct execution of movements. Using data from inertial sensors, such as accelerometers and gyroscopes, the approach monitors, learns patterns, analyzes, and provides correction suggestions for the inertial data of each body segment through a Restricted Boltzmann Machine. The results demonstrate that the generation of these execution-time models, adaptable to different body types and user limitations, is efficient in producing adjustment guidance for movements, resulting in a similarity up to 3.68 times greater with the correct movement. This validates the effectiveness of the proposed method for its purpose.