Realtime Indoor Workout Analysis Using Machine Learning & Computer Vision

Area of Science:

• Computer Science
• Biomechanics
• Sports Medicine

Background:

• Thousands of online workout videos exist, but users struggle with correct exercise form.
• Incorrect exercise execution can lead to severe long-term injuries.
• Existing solutions lack real-time, accurate feedback on body pose alignment.

Purpose of the Study:

• To develop a system that analyzes user body posture during workouts.
• To provide visual feedback and corrective actions for accurate exercise performance.
• To compare user's body pose against a professional's reference workout.

Main Methods:

• Utilizing deep learning for human body pose estimation.
• Representing the human body as a collection of limbs and analyzing inter-limb angles.
• Employing Dynamic Time Warping (DTW) and optical flow tracking for video synchronization.

Main Results:

• The system effectively detects and locates errors in user's exercise form.
• Accuracy is achieved by analyzing deviations in limb angles compared to a reference.
• Provides corrective feedback to improve workout technique.

Conclusions:

• The proposed system offers a viable solution for real-time workout form analysis.
• It has the potential to reduce exercise-related injuries through accurate feedback.
• Future applications include aiding physicians in monitoring patient recovery.