AI Surveillance System with ESP32-CAM

Real-time multi-camera surveillance system with AI-powered detection of people and weapons. Built with ESP32-CAM, YOLOv8, Flask, and Qt.

IEEE AIoT 2025 OSAKA pending . . .

Overview

This project implements a scalable AI-powered surveillance system consisting of:

• ESP32-CAM Nodes: Capture JPEG images and send to the server.
• Python Flask Server: Runs object detection using YOLOv8 and manages multiple camera streams.
• Qt GUI Client: Displays live annotated feeds with status indicators and IP labels.

All AI computations are performed on the server, enabling lightweight camera nodes.

---

Components

ESP32-CAM Firmware

• Written in C++ (PlatformIO)
• Captures frames and sends them to a configurable list of server IPs.
• Lightweight and optimized for battery-powered operation.

Folder: esp32cam-firmware/

Python Flask Server

• Accepts frames from ESP32 nodes.
• Performs:
  • Weapon detection (pistol, rifle, knife) using a custom-trained YOLOv8m model.
  • Person detection using pre-trained YOLOv8l.
• Annotates images and broadcasts live feeds.

Folder: server/

Qt GUI Client (C++)

• Developed using Qt Creator.
• Fetches active camera IPs.
• Displays annotated feeds with:
  • Smooth scaling and alignment.
  • IP tag and live/offline status indicator.
• Automatically refreshes feed every 100 ms.

Folder: client/

---

AI Models

• YOLOv8 used for object detection via Ultralytics.
• Trained on Roboflow Weapon Detection Dataset.
• Training:
  • 100 epochs
  • Model: YOLOv8m
  • Optimized for pistols, rifles, and knives

---

System Architecture

All detection is centralized — ESP32 nodes are lightweight, cost-efficient, and easy to deploy.

---

How to Run

Prerequisites

• Python 3.11+
• ESP32-CAM module (AI Thinker or similar)
• PlatformIO (VSCode)
• Qt Creator (for GUI client)
• Ultralytics library: pip install ultralytics

1. Flash ESP32-CAM

cd esp32cam-firmware/
pio run --target upload

Update SSID, password, and server IP in the firmware before flashing.

2. Start the Server

Update the Telegram bot token in notification.py

cd server/notification
python notification.py

and then

cd server/
python app.py

Ensure the correct YOLO model weights exist under server/weights.

3. Run the Client GUI

Update the server IP in clienteye.cpp

Open clienteye.pro in Qt Creator and run the project.

---

File Structure

AI-surveillance-system-esp32/
├── esp32cam-firmware/       ESP32 source code
├── server/                  Flask + YOLOv8 detection server
│   ├── app.py
│   ├── detect.py
│   ├── camclass.py
│   └── weights/             YOLOv8 models
├── client/                  Qt GUI Client
│   └── clienteye.cpp/h
└── assets/
    └── system_architecture.png

---

Sample Output

Client:

Notifications:

---

Features

• Realtime multi-camera support
• AI detection of people and weapons
• Telegram notifications via bot in case of threat detection
• Offline ESP32 processing (no detection load)
• Smooth cross-platform GUI
• Modular and extensible system: The ESP32-CAM costs around $7, making it a cost-effective option for covering larger areas.

---

Future Improvements

• HTTPs instead of HTTP
• Cloud upload and remote access
• Mobile version of GUI client
• Add motion detection and smart frame skipping
• Web dashboard (Flask + Bootstrap or React)

---

License

MIT — Feel free to use, modify, and expand upon.

---

Author

Lev Vedrov
[email protected]

GitHub: @levvedrov