ESP32-Cam Surveillance System with YOLOv5 and Telegram Integration

This project combines an ESP32-Cam, YOLOv5, and Telegram API to create a smart surveillance system capable of detecting people and sending alerts directly to your phone. The ESP32-Cam streams video to your computer, where YOLOv5 processes it in real time. When a person is detected, the system captures an image, sends it to a Telegram bot, and logs relevant details.

---

Features

• ESP32-Cam Integration: Configured to act as a video stream server.
• Real-Time Object Detection: Powered by YOLOv5 for accurate and fast person detection.
• Telegram Alerts: Sends detection messages and captured images directly to your Telegram chat.
• Customizable WiFi Modes: Station or Access Point.
• Simple and Modular Code: Easy to extend or modify for your needs.

---

Requirements

Hardware

• ESP32-Cam (LilyGo T-Camera S3 used in this project)
• USB to Serial Adapter for programming the ESP32-Cam

Software

1. PlatformIO (for programming the ESP32-Cam):
   • PlatformIO installation guide
2. Python (3.8 or later)
   • Required libraries:
     • torch
     • opencv-python
     • requests
     • colorama
3. Telegram Bot API:
   • Follow the Telegram Bot API Documentation to set up a bot.

---

Installation and Setup

Step 1: Set Up ESP32-Cam

1. Clone this Repository:

   git clone https://github.com/cyprienbf/ESP32_YS.git
   cd ESP32_YS/ESP32_Camera
   code

   • Follow the instructions to install the necessary libraries.

2. Configure the platformio.ini and the partitions.csv File if needed:

   [platformio]
   src_dir = src
   
   [env:t-cameras3]
   platform = espressif32@6.9.0
   framework = arduino
   board = esp32s3box
   
   board_build.partitions = partitions.csv
   upload_speed = 921600
   monitor_speed = 115200
   
   build_flags =
       -DBOARD_HAS_PSRAM
       -DARDUINO_USB_CDC_ON_BOOT=1

3. Flash the Code:

   • Adjust the WiFi SSID and password in the WiFiSettings section of the code, located in the ESP32_Camera.ino file.
   • Use PlatformIO to upload to the ESP32-Cam.

4. Test the Camera Stream:

   • Configure the OV2640 camera via the short URL (e.g., http://<ESP32_IP>).
   • Access the stream via the URL shown in the serial monitor (e.g., http://<ESP32_IP>:81/stream).

Step 3: Configure the Python Script

1. Install Required Libraries:

   cd ../YoloSurveillance/
   pip3 install -r requirements.txt
   code

2. Install Required Tools

   • PyTorch (Ensure compatibility with your CUDA version, e.g., CUDA 12.4).

   pip3 install torch torchvision torchaudio

   • CUDA Toolkit (if using Nvidia GPU acceleration).

3. Edit main.py:

   • Update the following fields:

     class Camera:
         SOURCE = "http://<ESP32_IP>:81/stream"  # Replace with your ESP32-Cam stream URL
     
     class TelegramAPI:
         BOT_TOKEN = "<YOUR_TELEGRAM_BOT_TOKEN>"  # Replace with your bot token
         CHAT_ID = "<YOUR_TELEGRAM_CHAT_ID>"      # Replace with your chat ID

   • Train and use your own model or use YOLOv5's pretrained model (yolov5s is recommended and as default in this project). See model = torch.hub.load("ultralytics/yolov5", "yolov5s")

4. Run the Python Script:

   python main.py

Step 4: Test the System

1. Point the ESP32-Cam at an area with people.
2. The script will detect individuals and send an alert with the following:
   • Detection confidence
   • Coordinates of the detected object
   • Captured image via Telegram.
3. Adjust the confidence threshold and detection delay in main.py to suit your requirements:

   class Settings:
       CONFIDENCE = 0.6  # Adjust confidence threshold
       DETECTION_DELAY = 0.5  # Adjust delay in seconds

---

Usage Notes

• Detection Targets Accuracy:
  • Training your own model or using a better pretrained one.
• Expanding Functionality:
  • Add a web dashboard for real-time monitoring.
• Power Management:
  • Consider using a battery-powered ESP32-Cam with sleep modes for a portable solution.

---

References

1. YOLOv5 GitHub Repository
2. ESP32-Cam GitHub Repository
3. Telegram Bot API Documentation

---

License

This project is open-source and available under the MIT License.