Real Time Workout Posture Monitoring System Using Computer Vision

Ong, Cheng Leong (2025) Real Time Workout Posture Monitoring System Using Computer Vision. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text RSW_Ong Cheng Leong_Fulltext.pdf Restricted to Registered users only Download (6MB)

Abstract

This project addresses the challenge of maintaining correct workout posture for individuals exercising without professional guidance, a situation that can lead to injury risks and suboptimal fitness outcomes. To tackle this, a real-time posture monitoring system was developed using computer vision and deep learning techniques, focusing on exercises such as squats and bicep curls. The system captures video input through a phone camera, applies pose estimation with MediaPipe to extract key body landmarks, and normalizes these data points for consistent analysis. An LSTM-based model then classifies each frame sequence into correct or incorrect postures (e.g., lifting heels, shallow depth), enabling instant user feedback. Structured into modules for data collection, keypoint extraction, sequence generation, model training, and real-time detection, the system provides immediate on-screen guidance as well as audio alerts. Testing involved validating posture classification accuracy, ensuring real-time responsiveness, and verifying robust rep counting even under varying frame rates. Results confirm that the system can effectively identify posture mistakes and provide clear, timely feedback, thereby reducing injury risks and improving exercise efficiency. While performance remains high for straight side-view scenarios, the system’s accuracy may diminish if critical body landmarks are obscured or if users deviate significantly from the calibrated camera setup. Overall, this project demonstrates a practical, accessible approach to digital fitness coaching, offering a strong foundation for future enhancements such as multi-view analysis and an expanded exercise library. By bridging the gap in real-time form correction, it contributes a cost-effective and user-friendly solution for safer, more effective home workouts.