Raspberry Pi AI Security System

This project consists of an AI-powered real-time security system using YOLOv8 and YAMNet on Raspberry Pi 5, capable of:

• Detecting guns and knives using a custom-trained YOLOv8 model
• Detecting suspicious sounds (screaming, crying, gunshot, etc.) using Google's YAMNet model
• Sending email alerts
• Activating a warning LED
• Playing a siren alarm sound
• Logging events locally

---

How It Works

Component	Description
image_detect.py	Uses YOLOv8 for weapon detection via usb camera
audio_detect.py	Uses YAMNet for screaming/gunshot detection via mic
GPIO (LED + sound)	Alerts via LED and alarm speaker
Gmail API	Sends email when a weapon or suspicious sound is detected
Log Files	Events are recorded locally for traceability

---

Hardware Setup

• Raspberry Pi 5
• Monitor
• Keyboard + Mouse
• USB Webcam (for YOLOv8 detection)
• USB Microphone (for YAMNet audio classification)
• Speakers
• LED + 470 Ohm Resistor (on GPIO pin 18)
• Breadboard + Wires

🎥 Demo Setup Video

---

Project Structure

AI-security-system/
├── yolov8_image_classification/
│ ├── image_detect.py
│ └── best.pt
│
├── yamnet_audio_classification/
│ ├── audio_detect.py
│ └── yamnet_model/
│
├── assets/
│ └── alarm.WAV
│
├── media/
│ ├── setup.jpg
│ ├── setup_video.mp4
│ ├── knife_gun_detection.mp4
│ ├── demo_with_led.mp4
| ├── scream_detection.mp4
│ ├── emails_sent_terminal_output.png
│ ├── emails_sent_screenshot.png
│ └── funny_demo.mp4
│
├── .env.example
├── requirements.txt
└── README.md

---

Installation

1. Clone the repository

git clone https://github.com/halahleis/AI-security-system.git
cd AI-security-system

2. Create and activate a virtual environment

python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

---

Environment Variables

Create a .env file in the root of your project with:

GMAIL_USER=your_email@gmail.com
GMAIL_APP_PASSWORD=your_app_password

Note: To get an app password 👉 How to create a Gmail App Password

---

Running the System

Weapon Detection via Camera (YOLOv8)

cd yolov8_image_classification
python image_detect.py

• Displays annotated detection frame
• Triggers alarm + LED + email when knife or gun is detected
• Logs detections to image_detection_log.txt

Suspicious Sound Detection (YAMNet)

cd yamnet_audio_classification
python audio_detect.py

• Plots real-time confidence of top 5 sound classes
• Triggers alarm + LED + email when suspicious sounds are detected ("crying, sobbing", "screaming", "shout", "gunshot", "explosion")
• Logs detections to audio_detection_log.txt

---

Email Alert

Terminal Output:

Emails Received:

---

Event Logging

Both detection scripts log alerts to:

• image_detection_log.txt
• audio_detection_log.txt

Each log includes a timestamp and detected event.

---

Customization

• To modify the audio classes that trigger alerts, edit these two sections in audio_detect.py:

suspicious_keywords = ["crying, sobbing", "screaming", "shout", "gunshot", "explosion"]

email_sent_flags = {
    "crying, sobbing": False,
    "screaming": False,
    "shout": False,
    "gunshot": False,
    "explosion": False
}

• The best.pt model was trained on a small custom dataset of around 700 images (after augmentation) for two classes: knife and gun. While sufficient for demonstration purposes, it is not robust for real-world deployment. For improved performance, consider training a yolo model on a larger, more diverse dataset.

---

Bonus: Demo Videos

Type	Demo
Knife/Gun Detection	🎥 Watch
Knife/Gun Detection + LED	🎥 Watch
Scream Detection	🎥 Watch
Fun Demo	🎥 Watch

I especially hope you enjoy watching the last demo as much as I enjoyed making it! 😊

---

Credits

• YOLOv8 from Ultralytics

• YAMNet from Google Research

  Note: YAMNet model is loaded locally from the yamnet_model/ directory and downloaded from https://www.kaggle.com/models/google/yamnet.

• Dataset labeling using Roboflow

---

Author

Made with ❤️ by Hala Hleiss