jaswalabhijeet/hiitpi

A workout trainer Dash/Flask app that helps track your HIIT workouts by analyzing real-time video streaming from your sweet Pi using machine learning and Edge TPU..

HIIT PI

HIIT PI is a Dash app that uses machine learning (specifically pose estimation) on edge devices to help track your HIIT workout progress in real time (~30fps). The backend runs everything locally on a Raspberry Pi while you interact with the app wherever there is a web browser connecting to the same local network as the Pi does.

How it works

• Video demo
• Blog post

Hardwares

• Raspberry Pi (Pi 4 recommended)
• Raspberry Pi Camera Module v2
• Google's Coral USB Accelerator (Edge TPU)

Softwares

• Raspbian 10 Buster
• Python 3.7+
• TensorFlow Lite
• Edge TPU runtime
• Dash, Flask, Plotly
• Redis
• PiCamera, OpenCV, ...(more in requirements.txt)

Usage guides

1. SSH into your Raspberry Pi and clone the repository.
2. Set up a working environment with dependencies listed above before running the app by

   $ python app.py
   

3. Go to <your_pis_ip_address>:8050 on a device in the same LAN as the Pi's, and then enter a player name in the welcome page to get started.
4. The live-updating line graphs show the model inferencing time (~50fps) and pose score frame by frame, which indicates how likely the camera senses a person in front.
5. Selecting a workout from the dropdown menu starts a training session, where your training session stats (reps & pace) are updating in the widgets below as the workout progresses. Tap the DONE! button to complete the session, or EXIT? to switch a player. Click LEADERBOARD to view total reps accomplished by top players.

Notes

• This project currently has implemented a couple of workouts to play with, and we're planning to add more later.