Anomaly Detection System

This project implements a video-based anomaly detection system using a Long-term Recurrent Convolutional Network (LRCN) model. The system can analyze video footage and detect potential anomalous behavior, with real-time notifications via email for high-probability anomalies.

Project Structure

your_project/
│
├── app.py                      # Flask backend
├── shoplifting_model.py        # Model implementation
├── requirements.txt            # Python dependencies
├── lrcn_160S_90_90Q.h5        # Trained LRCN model file
│
├── templates/                  # Frontend templates
│   └── index.html             
│
├── static/                     # Static files
│   ├── uploads/               # Uploaded videos
│   └── outputs/               # Processed videos
│
├── data/                      # Data directory
│   ├── input/                # Original input videos
│   └── output/               # Reference outputs
│
├── notebooks/                 # Jupyter notebooks
│   └── run.ipynb             # Development notebook
│
└── models/                    # Model storage
    └── lrcn_160S_90_90Q.h5   # Current model

Setup Instructions

1. Create a virtual environment:

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

2. Install dependencies:

   pip install -r requirements.txt

3. Run the application:

   python app.py

4. Access the web interface at http://localhost:5000

Features

• Video upload and processing interface
• Real-time anomaly detection
• Email notifications for high-probability anomalies
• Video visualization with detection results

Model Details

The system uses a Long-term Recurrent Convolutional Network (LRCN) model trained on video sequences. The model processes video frames in sequences of 160 frames, with each frame resized to 90x90 pixels.

Note

Make sure to update the email configuration in shoplifting_model.py with your own email credentials for notifications.