Live Demo
Check out the live demo of our project here.
Project: AI Fitness Trainer
Problem Statement
The importance of proper form in fitness is paramount to avoid injuries. Self-assessing form is difficult, creating a need for a personalized fitness coach. This project leverages deep learning and computer vision to build a real-world application that addresses these challenges.
Solution
We developed a computer vision-based fitness trainer using Mediapipe to analyze squat exercises, providing real-time feedback based on the input format and difficulty mode.
Application Details
The AI Fitness Trainer uses MediaPipe Pose for high-fidelity body pose tracking. It infers 33 3D landmarks and provides feedback based on the analysis of these landmarks during squat exercises. The application calculates angles of the hip-knee, knee-ankle, and shoulder-hip lines with the verticals to determine the states and perform the appropriate feedback messages.
State Diagram
The application maintains a state sequence to determine whether a correct or incorrect squat is performed, providing real-time feedback to ensure proper form.
Scope of Improvement
Future enhancements include adding different sets of exercises, integrating multiple camera views, and incorporating more advanced techniques such as a Human Action Recognition system using a CNN-LSTM model. We also plan to use wearable sensors like Inertial Measurement Units (IMUs) for time-series analysis.