Covert Ops: AI for Real-Time Crime Detection

📌 Abstract

Covert Ops is an AI-powered surveillance system designed to enhance real-time crime detection capabilities. Using a DenseNet121 backbone with transfer learning, this project identifies and classifies suspicious or criminal incidents from video feeds. Activities such as arson, vandalism, traffic accidents, and fire are recognized via a robust model fine-tuned on crime-specific datasets.

🧠 Key Features

• 📹 Real-time video classification using CNN
• 🔍 Fine-tuned DenseNet121 model
• 🔄 Data preprocessing and augmentation
• 📊 Model testing and evaluation via Jupyter Notebooks
• 📁 Modular structure for reproducibility and scalability

🛠️ Tech Stack

• Languages: Python 3.8+
• Libraries: TensorFlow, Keras, OpenCV, NumPy, Matplotlib
• Model: DenseNet121 (with transfer learning)
• Interface: Jupyter Notebooks

🔧 Installation

✅ Prerequisites

• Python 3.8 or above
• pip
• virtualenv (optional but recommended)

⚙️ Setup

# Clone the repository
git clone https://github.com/GodFWarsion/Covert-Ops.git
cd Covert-Ops

# (Optional) Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows use: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

🔗 Datasets

This project uses two publicly available datasets:

🔥 Fire Detection Dataset (Kaggle)

• Source: https://www.kaggle.com/datasets/phylake1337/fire-dataset
• Structure: Contains fire and non-fire image classes.
• Place it under:

🎥 UCF-Crime Dataset

• Official site: https://www.crcv.ucf.edu/projects/real-world/
• Kaggle alternative input: https://www.kaggle.com/code/odins0n/video-anomaly-detection/input
• You’ll need to manually download and extract videos.
• More Crime DataSet - Arson,Vandalism,Murder,Explosion ... on the way

⚠️ Note: These datasets are large and cannot be included in this GitHub repo. Please download them manually and place them in the recommended directory structure.