Form Perfector

Abstract

Home-based workouts are known to have significant health and fitness benefits, but it could be dangerous if not performed with proper posture. This could be due to a lack of sufficient training and bad habits, or the inability to afford a personal trainer to fix the individual's posture. With the current Artificial Intelligence technology, we are now able to achieve workout analysis by using computer vision approach without any involvement of personal trainer. This report discussed the approach of keypoint detection, human pose estimation and pose analysis that will be utilized to produce the proposed project through Rapid Application Development (RAD) methodology, which will monitor these procedures phase by phase. Pretrained model from MediaPipe was chosen as the main human key point detection, and various machine learning algorithms such as Linear Regression and Random Forest were studied to find the suitable algorithm to be implemented in detecting and analyze home workout postures. Furthermore, this research also studies on geometrical approach to determining proper and improper posture for analysis. These approaches were then compared to see which will produce the best results and user experience. Since machine learning approach does not produce the best result due to lack of training data and variety of other factors, the geometrical approach was chosen as the final approach before implementing it into desktop application. Keyword: Personal Trainer, Keypoint Detection, MediaPipe, Rapid Application Development.