Project Overview:

The Intelligent Home Surveillance System with Intrusion Detection is an IoT and AI-powered system designed to enhance home security. It uses motion sensors or cameras to monitor activity and detect unauthorized intrusions in real-time. AI models process video or sensor data to distinguish between normal movements (like pets or family members) and suspicious behavior (like forced entry or unknown individuals).

Upon detecting an intrusion, the system immediately triggers an alarm, sends notifications to the homeowner, and optionally starts video recording or livestreaming through a secure web interface.

Technologies Used:

Hardware:

• Microcontroller: ESP32 / Raspberry Pi

• Sensors: PIR motion sensor, ultrasonic sensor, door sensor

• Camera: USB or Pi Camera module (for video detection)

• Buzzer / Alarm: For local alert

• Wi-Fi Module: Built-in (ESP32/RPi)

Frontend:

• HTML, CSS, Bootstrap

• JavaScript (Live video via WebRTC or MJPEG stream)

• Chart.js for intrusion log analytics

Backend:

• Option 1: PHP + MySQL

• Option 2: Node.js + MongoDB

• Option 3: Java Spring Boot + PostgreSQL

AI/ML:

• Python (OpenCV + TensorFlow / Keras)

• Models:

  • Object Detection (YOLO, SSD) for recognizing humans vs pets/objects

  • Intrusion classifier (person detected outside allowed hours)

System Architecture:

1. Sensors or camera continuously monitor home premises.

2. Microcontroller / Raspberry Pi sends input data to the server.

3. If motion or object is detected, camera captures a photo or starts live video.

4. AI model analyzes the frame:

   • Identifies if it's a human

   • Compares with known faces (optional)

   • Flags unknown person or suspicious activity

5. If intrusion is confirmed:

   • Alarm is triggered

   • User is notified via email/SMS

   • Snapshot is saved and shown on dashboard

6. Web dashboard shows:

   • Live camera stream

   • Alert logs

   • Intrusion timestamps

   • Access control options (arm/disarm)

ML Model Description:

✅ Object Detection (Human vs Other):

• Input: Real-time video frame

• Model: Pre-trained YOLOv5 / MobileNet SSD

• Output: Bounding box over detected person

✅ Face Recognition (Optional):

• Model: FaceNet or Dlib

• Compare detected faces with registered family members

• If unknown face: classify as intruder