Edge AI Intrusion Sentinel (Offline Smart Security System)

Most DIY security systems either generate false alarms or depend heavily on cloud processing. This project builds a privacy-first, offline intrusion detection system using Edge AI.

It combines:

• PIR motion sensing (cheap + fast trigger)
• Computer vision (AI) for verification

👉 Result: Accurate + fast + no internet required

🎯 Features

• Real-time human detection (<200 ms inference)
• No cloud / no data leakage
• Snapshot capture on detection
• Audible alert system
• Expandable (Telegram, email alerts, etc.)

🔌 Wiring Diagram (Text-Based Schematic)

PIR Sensor → Raspberry Pi

• VCC → 5V
• GND → GND
• OUT → GPIO17

Buzzer → Raspberry Pi

• → GPIO27
• → GPIO27
• – → GND

Camera

• Connect via CSI ribbon cable

Coral USB

• Plug into USB 3.0 port

🧠 System Architecture

    [PIR Trigger]
      ↓
[Capture Image]
      ↓
[AI Model (Edge TPU)]
      ↓
[Person Detected?]
   ↓ YES        ↓ NO
[Alert]        [Igno

⚙️ Software Setup

1. Install OS

• Flash Raspberry Pi OS (64-bit)

2. Install Dependencies

sudo apt update
sudo apt install python3-pip
pip3 install opencv-python numpy pillow

3. Install Coral TPU Runtime

echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
sudo apt update
sudo apt install libedgetpu1-std

📦 Download Model

Use precompiled model:

• mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite

💻 Full Python Code

import cv2
import time
import RPi.GPIO as GPIO
from pycoral.adapters import common
from pycoral.adapters import detect
from pycoral.utils.edgetpu import make_interpreter

# GPIO setup
PIR_PIN = 17
BUZZER_PIN = 27

GPIO.setmode(GPIO.BCM)
GPIO.setup(PIR_PIN, GPIO.IN)
GPIO.setup(BUZZER_PIN, GPIO.OUT)

# Load model
interpreter = make_interpreter('model.tflite')
interpreter.allocate_tensors()

cap = cv2.VideoCapture(0)

def alert():
    GPIO.output(BUZZER_PIN, True)
    time.sleep(1)
    GPIO.output(BUZZER_PIN, False)

while True:
    if GPIO.input(PIR_PIN):
        print("Motion detected!")

        ret, frame = cap.read()
        if not ret:
            continue

        _, scale = common.set_resized_input(
            interpreter, (frame.shape[1], frame.shape[0]), lambda size: cv2.resize(frame, size)
        )

        interpreter.invoke()
        objs = detect.get_objects(interpreter, 0.5, scale)

        for obj in objs:
            if obj.id == 0:  # person class
                print("Human detected!")
                cv2.imwrite("intruder.jpg", frame)
                alert()
                break

    time.sleep(0.5)

🚀 Future Improvements

• Telegram alerts
• Face recognition
• Night vision optimization
• Battery backup
• Web dashboard