Github

Motivation

Goal: Build a home security system that uses facial recognition to detect an intruder.

How it Works

If a roommate is detected (with or without someone else) don’t ask for a password. If someone is detected and it is not a roommate, ask for a password, if the password is incorrect notify a roommate by sending an email with pictures as attachemnts and a text message. If the password is correct don’t notify roommate.

Tools Used

  • Hardare: Raspberry Pi, Keyboard, Camera

  • Language: Python

  • Packages

    • OpenCV

    • OpenCV-Contrib

    • Numpy

    • Python-PiCamera

    • Twilio

    • Memory_Profiler

Demo

Analysis

Memory Analysis

Time Analysis

Security

Assumptions

  • Using minimum number of characters (6 required)

  • Using QWERTY keyboard (101, 104, or 107 possible characters)

  • Using Efficiency Constant = 0.99 Using encryption constant 45898 (from a John the Ripper benchmarks)

Number of combinations: 1.615202*10^12 - 1.5007304*10^12

Brute Force Times

Algorithm: Effective Cores: 1/1-Efficiency Constant)+(Efficiency Constant/Processor Cores GFLOPS: Processor Frequency * Effective Cores Keys per second: GFLOPS/Encryption Constant Seconds = Combos/Keys per second

  • Average Computer (3 ghz, quad core)

    • Effective Cores: 3.883495

    • GFLOPS: 13980582524.3

    • Keys per second: 304601.12694

    • Time (for 101 possible characters): 3484951.65026 seconds, 968.04213 hours, 40.335 days

    • Time (for 107 possible characters): 4926870.80667 seconds, 82114.5134445 minutes, 1368.57522408 hours, 57.024 days

  • 1,000 Core Computer

    • Effective Cores: 90.9918

    • GFLOPS: 1.6196542*10^14

    • Keys per second: 3528812216.48

    • Time (for 101 possible characters): 300.815 seconds

    • Time (for 107 possible characters): 425.2792 seconds

Improvements

  • Use database to store test data

  • Create setup script to setup household

  • To avoid a brute force attack monitor speed of password entry (if speed not humanly possible, exit)