Barbell Exercise Tracker

Overview

This project focuses on developing a barbell exercise tracking system to classify exercises, count repetitions, and detect improper form using accelerometer and gyroscope data from wearables. The system was built with the aim of providing real-time insights for strength training, similar to a digital personal trainer, leveraging machine learning for accurate tracking and feedback.

Features

• Exercise Classification: Classifies various barbell exercises using sensor data.
• Repetition Counting: Counts repetitions accurately during gym workouts.
• Form Detection: Detects improper exercise form to help users prevent injuries.

Tools and Technologies

• Programming Languages: Python
• Data Collection: Accelerometer and Gyroscope data from Empatica Embrace Plus wristband sensors.
• Libraries:
  • NumPy, Pandas: For data manipulation and preprocessing.
  • scikit-learn: For training and evaluation of machine learning models.
• Data Analysis Techniques:
  • Supervised Learning: Used for classifying exercises and detecting improper form.
  • Clustering (K-Means): To identify patterns in exercise data.
  • Feature Engineering: PCA, Low-pass Filtering, and Fourier Transformation for data preprocessing.
• Deployment Tools:
  • Anaconda: Used to create and manage the project environment.
  • Docker: For containerizing the application and ensuring consistency across environments.

How It Works

1. Data Collection: Data is collected from wristband sensors during different barbell exercises such as squats, deadlifts, and presses.
2. Data Preprocessing: The raw sensor data is cleaned, aggregated, and transformed using techniques like low-pass filtering and PCA.
3. Model Training: A variety of machine learning models, including Random Forests, were trained and evaluated to classify exercises and detect improper form.
4. Repetition Counting: Peak counting algorithms were applied to count the number of repetitions during each exercise.
5. Form Analysis: Models were trained to detect improper exercise form using collected data, providing corrective feedback.

Setup and Installation

1. Clone the Repository:

   git clone <repository-url>
   cd barbell-exercise-tracker

2. Set Up Environment:
   • Create and activate an Anaconda environment:

     conda create --name exercise_tracker_env --file conda_requirements.txt
     conda activate exercise_tracker_env

   • Install additional Python packages:

     pip install -r pip_requirements.txt

3. Run the Application:

   python main.py

Results

• Achieved 98% accuracy in exercise classification.
• Obtained a 5% error rate for repetition counting.

References

• The project is inspired by concepts from "Machine Learning for the Quantified Self" by Mark Hoogendoorn and Burkhardt Funk, leveraging sensor data for activity tracking.

Future Improvements

• Extend the system to include more types of strength training exercises.
• Implement real-time feedback for form correction via mobile or web application interfaces.