🏃‍♂️ Human Pose Estimation

A powerful, interactive application that detects and visualizes human body keypoints and skeletons in real-time. This project leverages the power of OpenCV DNN and Google MediaPipe to provide accurate pose estimation for both static images and live video streams.

🌟 Features

• Image-Based Estimation: Upload any image (JPG, JPEG, PNG) and visualize the estimated pose skeleton using OpenCV's DNN module.
• Real-Time Video Processing: Utilize MediaPipe for high-performance, real-time pose tracking from video files or live webcam feeds.
• Interactive Controls: Adjust detection thresholds on-the-fly using the Streamlit dashboard.
• Deep Learning Powered: Uses pre-trained TensorFlow models (graph_opt.pb) and MediaPipe's optimized pipelines.

🛠️ Tech Stack

• Frontend: Streamlit
• Computer Vision: OpenCV, MediaPipe
• Deep Learning: TensorFlow (Inference via OpenCV DNN)
• Language: Python 3.x

🚀 Getting Started

Prerequisites

Ensure you have Python installed. We recommend creating a virtual environment:

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

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/human-pose-estimation.git
   cd human-pose-estimation

2. Install the required dependencies:

   pip install streamlit opencv-python mediapipe numpy pillow

3. Ensure you have the model weights:

   • The project uses graph_opt.pb for OpenCV-based detection.

🖥️ Usage

1. Streamlit Web App (Image Estimation)

Launch the interactive web interface to process static images:

streamlit run st1_basic.py

• Upload: Use the sidebar or main area to upload an image.
• Threshold: Adjust the slider to fine-tune the confidence level for keypoint detection.
• Results: View the original and processed images side-by-side.

2. Jupyter Notebook (Video Estimation)

For video processing and MediaPipe-based real-time estimation:

• Open video_pose.ipynb in your preferred editor (VS Code, JupyterLab).
• The notebook is configured to process run1.mp4 by default but can be toggled to use your Webcam (0).

📊 Keypoints Detected

The model detects 18 key body parts, including:

• Nose, Neck
• Shoulders (R/L), Elbows (R/L), Wrists (R/L)
• Hips (R/L), Knees (R/L), Ankles (R/L)
• Eyes (R/L), Ears (R/L)

📁 Project Structure

.
├── graph_opt.pb          # Pre-trained TensorFlow model weights
├── st1_basic.py          # Streamlit application source code
├── video_pose.ipynb      # MediaPipe video processing notebook
├── run1.mp4              # Sample video for testing
├── stand.jpg             # Sample image for testing
└── README.md             # Project documentation

🤝 Contributing

Contributions are welcome! If you have ideas for improvements or find any issues, feel free to open a Pull Request or create an Issue.

---

Developed with ❤️ by [Rupkatha]