Two-dimensional Human Pose Estimation using Key Points' Angular Detection for Basic Strength Training

Abstract

Human pose estimation (HPE) has emerged as a crucial topic in computer vision, with applications ranging from sports training to injury prevention. This paper proposes a real-time 2D pose estimation system that leverages keypoint angle detection for basic strength exercises, such as squats, bicep curls, and deadlifts. The system integrates MediaPipe to detect joint positions and analyze them against optimal movement patterns determined by fitness guidelines. Primary data were collected using standard webcam recordings of exercises performed by expert trainers, enabling the system to establish joint angle thresholds and validate user movements. The system’s performance was evaluated through user trials, where it successfully validated repetitions and provided real-time feedback. Results showed an average accuracy of 80% across all exercises, with sensitivity maintained at 100%. Pearson correlation analysis demonstrated strong validation performance, with a coefficient of 0.98. Factors affecting performance included discrepancies in body proportions and user familiarity with strength training techniques. Despite these challenges, the system effectively highlighted errors, promoting improved form and reduced injury risk. This study contributes to the development of accessible and efficient real-time HPE systems that can operate on standard hardware. It emphasizes the practical application of pose estimation in enhancing training outcomes, enabling independent users to improve their techniques while minimizing injury risks. Future work will expand the range of exercises and integrate automatic exercise detection to further improve the system’s usability and versatility.