Abstract:
Maintaining proper exercise technique during home workouts is critical to achieving fitness goals and preventing injuries. However, many individuals lack access to professional guidance and proper form correction. Existing fitness applications fail to provide comprehensive pre-workout assessments, personalised feedback on exercise technique, and continuous mistake tracking, leaving users prone to ineffective workouts and potential injuries. This project addresses these challenges by developing a Fitness AI Coach that uses machine learning and computer vision to deliver personalised training with form correction, safety assessments, and progress tracking.
The application integrates pose estimation models using TensorFlow.js and MediaPipe BlazePose to analyse exercise movements from verified datasets. Using the device's camera, the system tracks user movements and compares them against professional exercise demonstrations. The system provides post-exercise feedback indicating form corrections through detailed analysis of body positioning and movement patterns. Additionally, the application incorporates pre-workout wellness checks, mistake tracking across sessions, and LLM integration for intelligent workout guidance. Progress analytics generate comprehensive reports including technique quality assessment, areas for improvement, and personalised recommendations for enhanced workout effectiveness.
The system was evaluated using chi-squared statistical analysis to assess the accuracy of exercise form detection and correction capabilities. Testing was conducted across multiple exercise categories with participants performing both correct and incorrect movement patterns. The chi-squared test yielded a statistically significant result with χ² = 12.847, p < 0.05, demonstrating that the system's pose classification accuracy significantly exceeded random chance performance.The results indicate effective differentiation between proper and improper exercise techniques, with the BlazePose integration successfully identifying form deviations and providing targeted corrective feedback.
