AutoGait

This project performs video anomaly detection using an autoencoder trained with a video containing normal frames. It utilizes the scikit-learn library to develop the autoencoder for anomaly detection. The project includes an alerting section that displays the anomalous video frames on the screen, save them in a directory and sends an alert via email with attached images of the detected anomalies.

This initial stage consists of developing an autoencoder that can be used to recognize when a person falls. Afterwards, the ability to recognize a person by their gait and distinguish different dangerous situations in a domestic environment will be added. The ultimate goal is to monitor the elderly automatically in their home environment.

Project Structure

The project is structured as follows:

AutoGait/
├── data/
│ └── input_videos
│   ├── normal_video1.mp4
│   └── normal_video2.mp4
│ └── test_video.mp4
├── models/
  └── autoencoder_model.pkl
├── src/
│ ├── alerting.py
│ ├── autoencoder.py
│ ├── detector.py
│ └── video_utils.py
├── test/
│ ├── anomaly.jpg
├── utils/
│ └── email_sender.py
├── Project_lineup.md
├── README.md
├── autoencoder_train.py
├── main.py
└── requirements.txt

The main components of the project are:

• data/: This directory contains the video files used for training the autoencoder and detecting anomalies. The input_videos/ directory contains normal videos used for training, while the test_video.mp4 file is used for anomaly detection.
• models/: This folder contains the autoencoder trained models. They can be loaded into the script to run tests with different models.
• src/: This directory contains the Python scripts that implement the autoencoder, detector, and alerting functionality.
• test/: This directory contains the detector outputs.
• utils/: This directory contains a utility script for sending email alerts with attachments (work in progress).
• Project_lineup.md: This file explains how to organize and implement the anomaly detection project applied to the gait.
• README.md: This file provides an overview of the project, instructions for setting up and running the project, and other relevant information.
• requirements.txt: This file lists the dependencies required to run the project.

Steps

1. Set up a virtual environment:

   $ python3 -m venv venv
   $ source venv/bin/activate
   
   

2. Install the dependencies:

   $ pip install -r requirements.txt
   
   

3. Fine tune the project:

   • Replace videos in the data/ directory.
   • Adjust the autoencoder architecture, threshold, and other parameters.  
      

4. Train the autoencoder:

   $ python3 autoencoder_train.py
   
   

5. Run the project:

   Run main.py script to detect anomalies and display/send alerts.

   $ python3 main.py
   
   

6. Evaluate the results:

   • Monitor the console output for any detected anomalies during the execution of the main script.
   • The script will display anomalous video frames on the screen and save them in the test/ directory.
   • Fine tune the autoencoder and the detector in order to obtain the expected result.