Face-ID Monitor

A monitor that knows your faces.

---

Project maintained by NTUEE-ESLab Hosted on GitHub Pages — Theme by mattgraham

Face-ID Monitor

Features

• Face Detection
• Motion Detection
• Mail alarm
• Python module with tunable parameters

Scenario

The camera keeps detecting motion. When motion is detected, the authorized face must be detected within a buffer time to release the alarm; otherwise, the buzzer will ring, and Pi will send images by e-mail.

Requirements

SMTP function must be installed, which is not contained in python functions. Refer to the reference below to setup the SMTP server. Other modules that are required are:

• click==6.7
• Flask==0.12.2
• itsdangerous==0.24
• Jinja2==2.9.6
• MarkupSafe==1.0
• picamera==1.13
• Werkzeug==0.12.2
• numpy==1.13.1
• opencv-python==4.0.0

Software Structure

Main thread

Camera thread

Motion detection module

Face detection module

Indicators thread

Mail script

How to use

Face registeration

First, you have to register your face into the machine. By uploading your photos (recommended less than 2K in both width and length) into the dataset/ directory. You should create a new folder named as your name and put your photos into the folder. e.g.

• dataset/Mike/IMG_0001.JPG
• dataset/Mike/IMG_0002.JPG
• dataset/Mike/IMG_0003.JPG
• dataset/Kevin/IMG_0045.JPG
• dataset/Kevin/IMG_0049.JPG
• dataset/Kevin/IMG_0148.JPG

After that, execute python3 encode_faces.py to encode faces into a single file. Then this file will be used by the following face detection module

Setup your mail address and message

In mail.txt, you can set your own mail. Note that FROM/TO/Subject rows must be included; otherwise the mail may fail to be sent out.

run the main thread

Last, execute python3 app.py to run all the functions including http streaming, motion detection, face detection, LED and buzzer alarm, and mail alarm

Tunable parameters

• In camera_opencv.py,
  • ALARM_BUFFER: buffer time after motion is detected. This value can be set larger if false alarm is too frequent
  • AUTH_BUFFER: buffer time after face detection to prevent frequent face detection which may sometimes be annoying. Set this value lower if miss happens.
  • ALARM_DURATION: max time that alarm will last.
• In mail.txt,
  • FROM: the sender name that will be presented by the mail receiver
  • TO: the receiver mail address
  • BCC: BCC address if needed
  • Reply-To: the reply-to address for this mail
  • Subject: the subject of this mail
• In dataset/, create your own folders with your name as folder name, containing photos of yours

SWOT analysis

Strength

• Low power
• Light-weighted
• Wireless LAN supported
• Integration from capturing to alarming and selective recording
• Highly modularized funciton
• Tunable paremeters

Weakness

• Low resolution (480p) and frame rate (6fps) due to limited computing capability
• WLAN does not support enterprise security
• Haar detection is not rotation-resistant, causing low precision on face detection
• RPi has 4 threads of CPU, but few programs can fully utilize them. Despite, the clock rate of a single core is rather low.
• The model applies HOG for face recognition, which cannot beat the state-of-art method CNN. But CNN cannot be applied on RPi due to computing complexity

Opportunities

• Let the RPi concentrate at recording and streaming, put computing-intensive job at remote server

Threats

• Monitors with 1080p30 are everywhere on the market with rather low price.

Demo video

https://youtu.be/v7BJr0ogYmM

References

• Flask
• SMTP server
• Pi Face Recognition
• Pi Motion Detection
• Python Docs
• OpenCV Docs