Human Activity Recognition using CNN + LSTM Deep Learning Project

This project implements a Human Activity Recognition (HAR) system using a combination of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. The goal is to recognize and classify human activities based on video input.

Table of Contents

• Introduction
• Project Overview
• Requirements
• Installation
• Usage
• Models
• Evaluation
• Action Recognition on YouTube Videos
• Streamlit User-Friendly Version
• Contributing
• License

Introduction

Human Activity Recognition (HAR) is a field of study that involves the identification and classification of human activities based on data such as video sequences. In this project, we use a combination of CNN and LSTM layers to build an effective HAR model.

Project Overview

The project is divided into two main parts:

1. Part 1: CNN + LSTM Model Training

   • Train a model using CNN and LSTM layers for Human Activity Recognition.
   • Plot model loss and accuracy curves.
   • Implement the Long-term Recurrent Convolutional Network (LRCN) approach.

2. Part 2: Action Recognition on Videos

   • Download YouTube videos for testing.
   • Create functions for action recognition in videos.
   • Evaluate the trained model on test videos.
   • Implement a user-friendly Streamlit version.

Requirements

• Python 3.x
• TensorFlow
• Keras
• OpenCV
• MoviePy
• Pafy
• Matplotlib
• Streamlit (for Streamlit version)

Installation

1. Clone the repository:

   git clone <repository-url>
   cd <repository-directory>

2. Install the required dependencies:

   pip install -r requirements.txt

Usage

1. Part 1: Model Training

   • Open and run the Jupyter notebook for Part 1.
   • Follow the instructions in the notebook to train the CNN + LSTM model.

2. Part 2: Action Recognition on Videos

   • Open and run the Jupyter notebook for Part 2.
   • Follow the instructions in the notebook to download YouTube videos and perform action recognition.

Models

• Two models are implemented in this project:

  1. ConvLSTM Model
  2. Long-term Recurrent Convolutional Network (LRCN) Model

• The models are trained for Human Activity Recognition.

Evaluation

• Model performance is evaluated on test datasets.
• There are 6676 videos in UCF50 dataset.
• Loss and accuracy curves are plotted for model evaluation.

Action Recognition on YouTube Videos

• Functions are provided to download YouTube videos for testing.
• Action recognition is performed on these videos using the trained model.

Streamlit User-Friendly Version

• A Streamlit version is created for easy interaction.
• Users can recognize actions in a video with a single click.

Contributing

Contributions are welcome! If you have suggestions or find issues, please open an issue or create a pull request.

License

This project is licensed under the MIT License.

Please make sure to replace `<repository-url>`, `<repository-directory>`, and `<link-to-issues>` with the appropriate values. Additionally, include the license file (e.g., `LICENSE`) in your project directory. Adjust the structure and content as needed for your specific project.