Home Security System

Overview

The Home Security Project is an end-to-end, AI-driven monitoring system built around an NVIDIA Jetson Nano that delivers real-time intrusion detection, identity recognition, and remote control—all for under $200 in hardware. Stereo CSI and ESP32-CAM modules capture video streams, while PIR sensors trigger motion-based alerts and guide active pan-tilt adjustments. On-board inference pipelines (MultiCue + YOLOv4-Tiny + dlib embeddings) classify individuals as Residents, Visitors, or Intruders, logging events locally and pushing snapshots and logs to Google Drive for off-site review. A companion mobile app (MIT App Inventor) provides live notifications, camera control, and event history, enabling homeowners to securely monitor and manage their property from anywhere.

Components

  • Central Computer NVIDIA Jetson Nano Developer Kit (2 GB RAM) running all decision and inference algorithms

  • Wired Camera Raspberry Pi 4 CSI Camera Module (1080p @ 30 FPS) connected via the Jetson’s CSI port

  • Wireless Cameras Two ESP32-CAM modules (5 V, up to 10 m range) each mounted on a pan-tilt servo assembly for adjustable field-of-view

  • Motion Sensors Two HC-SR501 PIR sensors (7 m range, 120° detection cone) for motion triggering

  • Pan-Tilt & Servo Motors Hobby servos driven by ESP32-CAM GPIO for remote camera aiming

Subsystems

Figure: Block diagram of the overall system and data flow.

1. Sensor Subsystem

Acquires raw data from cameras and motion detectors:

  • Wired & Wireless Cameras capture live video streams.

  • PIR Sensors detect movement and cue the cameras.
  • Pan-Tilt Assembly (servos + brackets) enables active FOV control.

2. Decision Subsystem

Processes sensor inputs and classifies each event:

  • Movement Detection via background-subtraction (MultiCue) to trigger heavier processing
  • Object & Face Detection using YOLOv4-Tiny for real-time localization of humans and pets
  • Face Recognition with dlib-based 128-dim embeddings (via face_recognition API) for identity matching
  • Classification Logic implements flowchart rules to label Residents, Visitors, Suspects, and Intruders based on database matches and behavior patterns
  • Database Management stores face encodings, timestamped entry logs, and snapshots in a local SQLite database

3. User Interface Subsystem

Allows residents to monitor, control, and review events:

  • Mobile App built with MIT App Inventor provides:
    • Real-time notifications of anomalies or intrusions
    • Remote pan-tilt control of wireless cameras
    • Log browsing and manual tagging of new faces
  • Communication over the home’s common Wi-Fi network

4. Outer Structure Subsystem

Encloses and mounts all hardware:

  • 3D-printed PLA Casings & Racks for Jetson Nano, CSI camera, ESP32-CAM platform, USB camera, and PIR sensors
  • Designs produced in Catia and FreeCAD for weatherproofing and aesthetic integration

Technologies Used

  • Hardware: Jetson Nano, Raspberry Pi CSI Camera, ESP32-CAM, HC-SR501 PIR, hobby servos, FTDI.

  • Algorithms & Libraries: MultiCue BGS for motion detection; YOLOv4-Tiny for object/face detection; dlib & face_recognition for face matching.

  • Software & Tools: Python, OpenCV, NVIDIA TensorRT optimizations; MIT App Inventor.

  • Interfaces & Protocols: CSI, GPIO, UART; Wi-Fi networking. —

Connectivity & Remote Access

  • Google Drive Integration A Python-based cloud storage automation module leverages the Google Drive API to upload captured snapshots, database backups, and event logs to a shared Drive folder—enabling off-site review and archival