🧍 Human Activity Recognition (HAR) using Live Video and Deep Learning

This project implements a real-time Human Activity Recognition system using a pre-trained deep learning model built with TensorFlow/Keras. The system captures live video from a webcam and classifies human actions such as walking, sitting, standing, and more.

---

📚 Table of Contents

• 🧠 Project Overview
• 📁 Project Structure
• 🔍 Model Details
• 🚀 How It Works
• 📦 Requirements
• ▶️ How to Run
• 🧪 Sample Output
• 📌 Notes
• 🧰 Utility Function: Max Subsequence
• 📷 Screenshots (Optional)
• 🤝 Contributing
• 📄 License

---

🧠 Project Overview

The goal of this project is to recognize and classify human activities using a Convolutional Neural Network (CNN) trained on image data. The model is capable of identifying various physical states from webcam video in real time.

Activities Detected

• 🧍 Standing
• 🪑 Sitting
• 🛏️ Sleeping
• 🚶 Walking
• 🧗 Walking on Stairs
• ⏹ Control (Default/Idle state)

---

---

🔍 Model Details

• Framework: TensorFlow / Keras
• Input Size: 150x150 RGB images
• Normalization: Pixel values scaled to [-1, 1]
• Output: Softmax classification across 6 activity classes

---

🚀 How It Works

1. Captures live video feed using OpenCV.
2. Each frame is:
   • Converted to a PIL image
   • Resized to 150x150 pixels
   • Normalized and reshaped to match the model's input format
3. Model predicts the activity from the processed frame.
4. Predicted label is printed to the console with a timestamp.

---

📦 Requirements

Install the dependencies using pip:

pip install numpy opencv-python tensorflow pillow

▶️ How to Run

Ensure your webcam is connected and functional.

Run the main script:

Copy Edit

python main.py

Press q to exit the live video feed.