Danger Detection with Computer Vision and ML

A real-time threat detection system using computer vision and machine learning to identify potentially dangerous objects (firearms and fire) and provide instant email notifications with visual evidence.

Made with ❤️, for Safety 🦺

• Real-time detection of guns, rifles, and fire in video streams
• Automated email notifications with captured images
• Easy-to-use interface for security monitoring
• Configurable alert thresholds and notification settings

• Python 3.8+
• OpenCV 4.5+
• CUDA support (optional, for GPU acceleration)
• Webcam or USB camera
• Minimum 4GB RAM
• x86_64 architecture
• Git (for cloning)

1. Clone the repository:

git clone https://github.com/phanindra-max/danger-detection-with-ml.git
cd danger-detection-with-ml

2. Create and activate a virtual environment:

python -m venv venv
venv\Scripts\activate  # Windows

3. Install dependencies:

pip install -r requirements.txt

Your requirements.txt should contain:

opencv-python>=4.5.0
numpy>=1.19.0
argparse>=1.4.0

1. Download YOLOv3 weights and config:

# Download weights
curl -L https://pjreddie.com/media/files/yolov3.weights -o yolov3.weights

# Download config
curl -L https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov3.cfg -o yolov3.cfg

2. Create obj.names file: Create a new file named obj.names with the following content:

gun
rifle
fire

1. Setup Gmail Authentication:

   • Enable 2-Factor Authentication in your Gmail account
   • Generate an App Password:
     1. Go to Google Account → Security → App Passwords
     2. Select "Mail" and "Windows Computer"
     3. Copy the 16-digit password

2. Update Email Settings:

# In main.py, update these variables:
mailfrom = "<your.email@gmail.com>"
gmailpass = "<your-16-digit-app-password>"
mailto = "<receiver@email.com>"

python main.py [--webcam True/False] [--image True/False] [--image_path PATH] [--verbose True/False]

Options:

• --webcam: Enable webcam mode
• --image: Enable image mode
• --image_path: Path to input image (default: "use-this.jpg")
• --verbose: Enable verbose output

python main.py

When prompted, enter 'Y' to start webcam detection.

• Average processing speed: 25 FPS on CPU
• GPU acceleration available with CUDA support
• Accuracy depends on image quality and lighting conditions

We welcome contributions! Please see Contributing Guidelines.

• Twitter
• BlueSky
• LinkedIn

• YOLOv3 team for the detection model
• OpenCV community for vision tools

• Add support for more object types
• Implement real-time API (Abandoned due to monetary constraints)
• Improve detection accuracy
• Add cloud deployment options

Made with ❤️ for Safety 🦺