Smart AI Security Camera using ESP32-CAM & YOLOv8

An advanced, real-time AI security camera system built with ESP32-CAM, Python, YOLOv8, and Face Recognition. This system detects humans, identifies known/unknown faces, and sends instant photo alerts to a Telegram group.

Features

• Live Video Streaming: Uses ESP32-CAM to stream video over a local network.
• Human Detection: Utilizes Ultralytics YOLOv8n to accurately detect humans while ignoring other movements (pets, fans, etc.).
• Face Recognition: Uses the face_recognition (dlib) library to match detected faces with a local database.
• Zero-Lag Processing: Implements multi-threading, frame-skipping (processing every 3rd frame), and dynamic frame resizing (0.5x) for smooth, real-time performance.
• Telegram Alerts: Automatically captures a photo of an "Unknown" person and sends it to a Telegram group with date and time. Built-in 5-second cooldown prevents spamming.
• Auto-Storage Management: A background daemon thread automatically deletes backup images older than 24 hours to save disk space.

Hardware Required

1. ESP32-CAM Module (with OV2640 Camera Sensor)
2. FTDI Programmer (CP2102 or CH340) for uploading code
3. Jumper Wires
4. 5V Power Supply

Software & AI Models

• Language: Python 3.x, C++ (Arduino IDE)
• AI Models: YOLOv8n (yolov8n.pt), dlib Face Recognition Model
• Libraries: OpenCV (cv2), ultralytics, face_recognition, requests

Complete Setup Guide

Step 1: Hardware & Driver Setup

1. Connect the ESP32-CAM to the FTDI programmer (Make sure IO0 is connected to GND during upload).
2. Install the CH340 or CP210x USB to UART bridge driver on your PC based on your FTDI module.

Step 2: ESP32-CAM Setup (Arduino IDE)

1. Open Arduino IDE and add the ESP32 board manager URL in Preferences.
2. Install the ESP32 board package and select AI Thinker ESP32-CAM.
3. Open esp32_cam_code.ino, update your Wi-Fi SSID and PASSWORD.
4. Upload the code, open the Serial Monitor (115200 baud), and copy the IP address (e.g., http://192.168.137.143:81/stream).

Step 3: Python Server Setup

1. Clone this repository:

git clone [https://github.com/yourusername/Smart-AI-Security-Camera.git](https://github.com/yourusername/Smart-AI-Security-Camera.git)
cd Smart-AI-Security-Camera

2. Install required Python packages:

pip install opencv-python ultralytics face_recognition requests cmake dlib

3. Add images of known people inside the database/ folder. Name the file with the person's name (e.g., rumon.jpg).
4. Update the STREAM_URL in main.py with your ESP32-CAM IP address.

Step 4: Telegram Bot Setup

1. Create a Bot: Open Telegram, search for @BotFather, and send the /newbot command. Follow the steps to name your bot. You will receive an HTTP API Token.
2. Create a Group: Create a new Telegram group and add your friends and the newly created bot to it.
3. Make Bot Admin: Go to the group settings and promote your bot to an Administrator (allow it to send messages).
4. Get Group Chat ID:
   • Send a test message (e.g., "Hello") in the group.
   • Open your web browser and go to this URL (replace YOUR_BOT_TOKEN with your actual token):

https://api.telegram.org/botYOUR_BOT_TOKEN/getUpdates

• Look for "chat":{"id": -100xxxxxxxxxx} in the text. Copy that entire negative number (including the minus sign).

5. Update Code: Open main.py and update the TOKEN and CHAT_ID variables with your newly generated details.

Step 5: Run the System

Start the AI security camera by running:

python main.py

Press q on the video window to stop the system.

How It Works (System Architecture)

1. Capture: The ESP32-CAM continuously captures frames and streams them over HTTP.
2. Read & Optimize: A background Python thread reads the stream using cv2.VideoCapture with a minimal buffer size. Frames are downscaled by 50%.
3. Detect: YOLOv8 scans the frame for the person class (class 0) with a confidence threshold of 0.3.
4. Recognize: If a person is found, face_recognition runs (every 3rd frame to prevent lag). It generates face encodings and compares them against the database.
5. Alert: If the face does not match, it is flagged as "Unknown". A red bounding box is drawn, and the frame is saved to backups/. The requests module sends this image to the Telegram API.