PosePerfect

A real-time AI-powered dance coach that uses computer vision to teach you how to Dougie (or any other dance)!

Features

• Real-time Pose Detection: Uses your webcam and MediaPipe pose detection to track key body joints
• Movement analysis: Calculates joint angles across frames and compares them to exemplar video of "The Dougie" (or your dance of choice)
• AI Feedback: Generates concise feedback with Google Gemini and plays it back with text-to-speech
• Customizability: Choose any dance to use, how much time to prepare, and how long to be recorded!

Quick Start

1. Clone the repo

   git clone https://github.com/ReeceLuo/TeachMeHowToDougie.git
   cd TeachMeHowToDougie
   

2. Create a virtual environment

   python3 -m venv venv
   source venv/bin/activate   # On Windows: venv\Scripts\activate
   

3. Install dependencies

   pip install -r requirements.txt
   

4. Setup API key
   • Create .env in your project root
   • Add your Gemini API key:

     GEMINI_API_KEY=your_api_key
     

5. Choose your dance
   • Upload an exemplar video to your project root
   • Change parameters when initalizing TeachMeHowToDougie

     app = TeachMeHowToDougie(api_key = os.getenv("GEMINI_API_KEY"), exemplar_video = "your_dance.mp4", dance_name = "your_dance", countdown_seconds, dance_duration)
     

Usage

1. Run the program

   python main.py
   

2. Start dancing!
   • Press 'r' to start recording (make sure entire figure is visible!)
   • Wait for feedback before starting again
   • Press 'q' to quit

Tech Stack

OpenCV – Video capture and rendering

MediaPipe – Real-time pose estimation

Google Gemini API – Natural language feedback generation

gTTS – Text-to-speech

Pygame – Audio playback