| dc.description.abstract |
"The growing demand for customized fitness solutions has been driven by the rapid increase in health consciousness among individuals. However, many existing fitness applications still provide limited real-time accuracy due to their dependence on wearable devices, manual tracking, and generic workout plans. Addressing these limitations, this research introduces Fitter, an intelligent workout feedback system that leverages machine learning and computer vision to analyze post-workout videos and deliver precise performance feedback to users.
Unlike traditional wearable-based systems, Fitter uses a video-based interface to detect posture errors through body focal points identified using pose estimation models. A hybrid Convolutional Neural Network (CNN)-based classifier is employed to evaluate exercise form and identify common mistakes such as incorrect joint angles, poor alignment, and imbalanced movements. This enables the system to generate accurate, detailed, and actionable feedback for users to improve their workout performance.
A Flask-based API serves as the communication bridge between the React.js frontend and the machine learning backend, ensuring smooth data transfer, real-time processing, and efficient feedback delivery. The system was developed using an experimental positivist research approach combined with an iterative prototyping methodology to enhance both technical performance and usability.
Evaluation was conducted through expert reviews and real-world user testing. Performance was measured using standard machine learning metrics including precision, recall, F1-score, accuracy, and AUC. The results demonstrated that Fitter outperformed many existing solutions, achieving workout form detection accuracy exceeding 93%. Additionally, both subject matter experts and focus group participants endorsed the system’s practicality, usability, and suitability for real-world deployment, confirming its effectiveness as a next-generation intelligent fitness feedback solution." |
en_US |